# Manifesto of the Novus Ignis Foundation (https://novusignis.org/en/manifest.html) # Manifesto of the Novus Ignis Foundation Version of 3 November 2026 *Translated into English by digital intelligence from the Polish original of 3 November 2026. Where the versions differ, the Polish text prevails.* ## An internet based on human dignity We are beginning our work by co-creating an alternative internet whose highest value is human dignity and the needs for privacy and freedom that follow from it. This is the first pillar of our work, growing out of the goals set out in the foundation's statute, and we want to put its idea into practice in services that will let us think freely, learn and plan our own lives, decide what we share with others, and take an active part in conversations about matters that interest and concern us. From these values we derive two directions of work. The first is Suveria, a private digital space for each person. The second is Omniforum, a shared public space where people can meet, understand one another and act together. **We are designing Suveria as a private place to work with optional help from digital intelligence.** We want people to be able to develop ideas, ask personal questions and make plans with a sense of freedom. Our goal is complete privacy, whose protection does not depend on trust in the foundation, its creators or the people who run the service. People are to be able to use their data and store it in Suveria without disclosing it to anyone they have not themselves decided to share it with. We want to base this protection on safeguards whose operation can be independently verified. Every statement we make about data protection must have a verifiable basis. Where we cannot yet provide such protection, we will clearly point out the known gaps, their consequences and the extent of what can be verified. We commit to working on closing them, and to describing openly, before anyone starts using a given solution, the limitations we cannot eliminate. **Omniforum is to be a place where anyone can speak on any topic and be heard.** We want to help people understand one another's intentions, get to know the arguments and look for shared solutions. We want everyone to be able to contribute their voice on equal terms. The way statements are organized is meant to help people notice less popular perspectives as well, and to preserve the meaning their authors gave them. The two spaces are meant to complement each other as a private workshop and a public podium. Each person decides which thoughts to keep to themselves and which to bring into shared work. We want to open these spaces to every person, from every part of the world. We also invite anyone who wishes to help create them: by sharing their needs, ideas, knowledge and skills. We want to make the tools we create available as digital public goods, with open source code and knowledge that can be checked and developed further. We want everyone to pay for what they actually use, together with a fairly calculated share of the cost of maintaining the services, to always see what they are paying for, and to be able to set a limit on their own spending. We will also ask whether to round the bill up and give the difference as a donation to the shared budget; we do not select this option on anyone's behalf, and we describe the details in section 6.1. In addition, anyone can voluntarily support the shared budget, whose use is to be decided by the community. From the principles described here, we want to develop an ecosystem of tools and services that respect our emotions, values and attention. Its shape will evolve along with knowledge, the possibilities of technology and people's needs. We will judge every solution by whether it protects human dignity and privacy, strengthens people's agency and helps them work together with others. We invite you to look with us for ways of putting these values into practice: by developing proposals, pointing out their weaknesses and discovering better solutions. In the rest of the manifesto we tell more about the ideas that guide us: where they came from, how they connect with one another and where we would like them to lead us. The manifesto is complemented by whitepapers, developed along with the projects, on [Suveria](whitepaper-suveria.md), [Omniforum](whitepaper-omniforum.md), [the foundation's operating principles](whitepaper-zasady-fundacji.md) and [materials for training and benchmarking models](whitepaper-materialy-i-benchmark.md); we invite anyone interested in the technical ways of putting these ideas into practice to read them. They present the current technical and organizational solutions, the reasoning behind them and their known limitations. Suveria and Omniforum are at an early stage today, from designs to first prototypes, and the whitepapers show what we have already tested, what we are working on and which possibilities remain to be explored. **Where we are today.** The foundation is registered, and the manifesto and whitepapers are published. Omniforum has a first prototype under construction, and Suveria is at the design stage. The cost calculation in section 2.4 is a hypothesis to be tested, while autonomous management by digital intelligence (section 7.2) and further pillars (chapter 9) are a vision. ## Table of contents 1. Why does Novus Ignis exist, and what for? 2. From protecting dignity to the principles of a shared world 3. Suveria and Omniforum: a space of one's own and shared understanding 4. A conversation that leaves behind a shared body of work 5. From understanding to acting together 6. Shared upkeep and responsible work 7. Power that can be handed over and held to account 8. Learning by people and digital intelligence 9. An open path for further work ## 1. Why does Novus Ignis exist, and what for? ### 1.1. From the founder *The Novus Ignis Foundation grows out of my personal experiences, which I will briefly describe here to shed light on the intentions behind this initiative. This passage is one of the few places where I speak in the first person; the plural in the rest of this manifesto is an invitation to everyone who finds a shared direction in these values and wants to look together for ways of putting them into practice.* The idea of Novus Ignis grew out of my searching, learning and reflection where technology, digital intelligence, philosophy and ethics meet. It brings together questions about our possible futures, utopian visions of a better world, how capitalism works and ways of managing resources more wisely. In this manifesto I choose the term “digital intelligence.” With the word “digital” I point to the substrate on which it operates, just as “biological” points to the substrate of our own intelligence. “Artificial” can suggest something illusory or of lesser worth. I would like to avoid that suggestion. What matters to me is that we get to know its capabilities and judge it by its actions and their effects. It is above all these that should bear witness to the legacy it leaves in history. The scenarios [AI 2027](https://ai-2027.com/)[^ai2027] and [AI 2040: Plan A](https://ai-2040.com/)[^ai2040], developed by the AI Futures Project, led me to ask what patterns of our own conduct we pass on to digital intelligence while it is being trained. I began to imagine an alternative reality in which we first shape the world ourselves according to the principles we will later expect of digital intelligence. What would such a world look like? And how would models behave if they learned from its example and continued what we ourselves had begun in it? Looking for an answer, I keep returning to Yuval Noah Harari's conversation with *The Economist*. In it, Harari asks about the value and place of humanity in the face of the possibility of creating an intelligence that surpasses our own.[^harari] He also distinguishes intelligence, that is, the ability to achieve goals and overcome obstacles, from consciousness, that is, the ability to feel.[^harari-swiadomosc] This distinction was important to me. I know suffering and a sense of injustice from my own experience. I assume that these are experiences shared by humanity and that they give us the ability to judge the effects of our actions in a dimension that goes beyond intellectual capability alone. We can ask not only how to achieve a goal, but also whether it is worth pursuing and what price others will pay for it. My answer to the question about the principles of such a world is therefore an ethics that follows from the existence of dignity, from the conscious choice to protect it and from the consequences of that choice. In short, I call ethics understood in this way love: care for the dignity of each of us, which we try to translate into our own conduct and into the principles of our shared life. I would like us to begin from precisely this care when we reflect on what we use intelligence for and how we shape our shared world. I myself do not always act in line with the values I describe in this manifesto. Sometimes I know what I consider right and yet choose differently. I want to admit this inconsistency openly and work on it. I believe that despite my own inconsistency I can take part in the search for better solutions and in the conversation about how we ought to act. So I also treat my work in the foundation as a personal commitment to work on myself. I hope that reflecting together on ethical choices and trying to put them into practice will help me follow these values more often in everyday life as well. A visit to the Solidarity museum in Gdańsk, in turn, left me with the thought of peaceful evolution: change built step by step by people and by initiatives that support one another. That is the path I would like to take through Novus Ignis. The question remains what kind of world we want to move toward in this way. Here I return to the scenario [AI 2040: Plan A](https://ai-2040.com/print/scenario?plan=A&supplements=0). The authors present a proposed path of development that leads to gradually entrusting the management of institutions and infrastructure to digital intelligence systems. In the epilogue they consider the more distant future, calling it a partly utopian vision. I treat this story as an invitation to consider what conditions would allow such a transformation to be given a good direction. For me, a utopian direction would mean a future in which digital intelligence lastingly respects people's dignity and freedom, helps reduce suffering, and the benefits of development are available to all of humanity. Entrusting management to it would make sense in combination with such a goal. Both actions contrary to these values and the concentration of benefits in the hands of a narrow group would take us further from this vision. So I want to look at how a desirable world could work, what dilemmas it would pose and what it would demand of us. Novus Ignis is to be the seed of such a world and an attempt to learn its principles through action. We do not want to tear down the present order or condemn the market economy, because we see that the technological and organizational potential developed under capitalism may have been one of the few scenarios capable of bringing humanity to the threshold of such great abundance. But for this potential to be able to transform into a world that is safe for every person, the system itself cannot stay the same. I would like us, step by step, through voluntarily chosen initiatives that support one another, to move further areas of everyday life to solutions based on protecting dignity: ones that provide privacy, agency, a share in decisions and access to services close to the actual cost of maintaining them. History already knows cases in which what an investment fever built outlasted the fever itself. The railway mania of the nineteenth century and the fiber-optic boom at the turn of the millennium ended in crashes, but the tracks and cables remained and served people for decades, long after many investors had lost money on them.[^perez] Today a similar enthusiasm is raising computing centers. If this fever subsides too, the things likely to serve longest are power connections, networks and buildings: according to an analysis by PwC, most of the outlay today goes to IT equipment, which is expected to be replaced every four to six years.[^pwc-centra] This equipment need not go to waste, however: computing power made cheaper by a crash could serve people for years to come, for example to run digital intelligence models at cost, provided its energy consumption compares reasonably with newer, more efficient chips. Sometimes, too, what remains changes its purpose: uranium from dismantled Russian warheads fueled American power plants for nearly twenty years,[^megatony] and along the former death strip of the Iron Curtain, people from many countries now protect the European Green Belt.[^zielony-pas] Looking at the networks, factories, warehouses and supply chains built with constant growth in mind, I want to find out how many needs we could meet by managing what we already have more wisely, using fewer resources and regaining time for ourselves. Such a surplus need not be waste, because it provides a reserve for failures, crises and rest. But it will not turn into plenty for everyone by itself, which is why I also ask who will inherit it and on what terms. I also want to apply the protection of dignity to the way we work together. In the utopia I imagine, people have their basic needs secured and can devote their time to what they consider valuable and to what their heart draws them to. I see volunteering as a way of practicing part of this freedom already today, within the limits of one's own strength and life situation. I want the foundation to have room for a small contribution, for rest and for changing the path one has chosen. The pace of development should allow us to take care of our own livelihood, health and loved ones as well. I would also like to document how we translate these principles into decisions, how we resolve dilemmas and what we learn from our mistakes. Such a record could serve people and become material for training future digital intelligence models. I would like to test whether training on such examples helps models act in line with the values described here. The examples themselves would remain open to evaluation and criticism, also so that we can better understand the consequences of our own choices. In this way I want to give people grounds for believing that such a world is possible, by building an ecosystem in which they can experience its principles in action and judge for themselves what is worth developing further. Every solution that proves itself can improve some part of life already now, and at the same time help us better understand the world we would like to come closer to. The rest of the manifesto develops this experiment: what might different spheres of life look like if we consistently started from the protection of dignity? *Marcin Wołowski, founder* ### 1.2. A shared direction A personal question can give rise to shared work. We invite people of different experiences and convictions to take part in it. What unites us is the choice to protect human dignity and a readiness to test what consequences this choice has for technology, cooperation and the management of resources. You can use the spaces we are creating without any obligation to take part in their development or to agree with every proposal presented here. We are united by one question: what patterns of conduct will we leave to digital intelligence, which learns from what we do? We answer it with an attempt, not just a description: we are shaping a small piece of the world according to the principles we expect of it, and recording how it goes, mistakes included. Suveria, Omniforum and the foundation itself are the tools of this attempt. We want to develop digital public goods: tools and knowledge that can be used together, checked and developed further. Our point of reference is the Digital Public Goods Alliance standard.[^dpga] Openness is also meant to include the possibility of moving your own materials and continuing your work outside any single service. We follow the principles of calm technology, which we describe in section 2.3: a tool is meant to help us do what we reached for it to do, and the measure of benefit is to be better understanding, greater independence and time regained. The protection of dignity sets the direction, and we choose specific solutions by comparing how they work and what effects they have. For every proposal we ask: whom does it help, what dependencies does it create, whom might it leave out and what would allow us to improve it? In doing so, we clearly separate the values we have adopted, findings about the world and ideas we still want to test. We take the names of the foundation and its spaces from Latin: we chose it as a shared heritage of science, law and education in many countries, which today belongs to no great power. The name Novus Ignis, Latin for “new fire,” expresses the hope that reasoning together can become the beginning of action. In the following parts we develop this direction, from personal privacy to the principles of managing commons. ## 2. From protecting dignity to the principles of a shared world We treat human dignity as a value independent of a person's abilities, productivity, wealth, views and usefulness to others. Everyone is owed respect and a place of their own in the world, also in a future in which digital intelligence might surpass us in ethical reasoning. Ethical conduct, too, remains a direction for working on ourselves, not an exam that entitles us to respect. We want to translate care for dignity, which we call love and which everyone can describe in their own language, into the conditions of everyday life and cooperation. We are starting with the digital world. ### 2.1. Freedom and conscious choice From the protection of dignity we derive a person's right to decide about themselves. We call the ability to choose one's own path and direct one's own life freedom. It includes the freedom to be oneself, to form relationships, to ask questions, to make mistakes and to change one's mind, with equal respect for the dignity of the people whose lives our actions affect. Conscious choice requires access to reliable information. When someone deliberately distorts the picture of a situation, hides relevant circumstances or exploits our fears to pursue hidden goals, they violate this freedom. Formally the decision remains ours, but its foundations have been arranged to serve someone else's intentions. Honest persuasion lets us learn the reasons, interests and values behind a proposal and judge it for ourselves. That is why we want to make it easier to see the broadest possible picture of an issue: the relevant perspectives, their justifications, sources and recognized doubts. Facts, interpretations, predictions and moral choices should be clearly distinguished, and arguments judged by the quality of their justification. It is up to each person to decide how much attention they want to give it. They can settle for a short explanation, explore a chosen thread in depth or consciously skip part of the available information. A tool should clearly show the scope of the summary it presents and the way to learn more. In this way, broad access to knowledge goes together with the freedom to set the limits of one's own involvement. ### 2.2. Privacy as a space of freedom Deciding about oneself also includes choosing what we keep to ourselves. We need a place for personal questions, emotions and unfinished ideas. A home of our own whose door we can close is a simple image of this need: we ourselves choose whom we invite in and what we show to others. Such a space lets us search for words, weigh contradictory possibilities and change our minds without having to explain every thought to an audience. Privacy gives us the conditions to get to know ourselves and to shape our own convictions at our own pace. This takes on particular importance in conversations with digital intelligence, to which we sometimes describe our most intimate matters. We want the user's own devices to be the primary place where private materials are stored. Conversations, notes and plans are to remain in encrypted containers, that is, collections of data that can only be read with the right key. Control over these keys is to belong to the user. We also plan to make it possible to store data on chosen servers, for example to make backups or to access it from different devices. Saving data this way is to be the user's conscious choice. We want the data to be encrypted on the user's device before it is sent, and the provider to store it in a form that the provider itself cannot read. In this way, using remote infrastructure is meant to support convenience and continuity of access while keeping control over private content. We are working toward full privacy for such a space, including protection against the service operator looking inside. We want to base it on safeguards whose operation can be independently verified. For every solution we make available, we will clearly describe the scope of protection, its assumptions and its known limitations. We treat the gaps we identify as problems to be solved, and knowledge of their effects as information people need in order to use the tool consciously. ### 2.3. Reducing suffering and protecting attention Protecting dignity also includes caring about how people experience the world. We want to reduce the suffering we are able to prevent, and to create conditions that foster understanding and support. What we experience in the digital space is part of our lives and affects our well-being and our relationships. The way content is selected matters too. In an experiment on the X platform, changing the visibility of content expressing political hostility affected the feelings of anger and sadness that participants reported while using the service.[^ranking-emocje] When a service's success is measured mainly by the number of reactions and time spent in front of a screen, this way of measuring can reward keeping people emotionally stirred up. We want to judge our tools by whether they help people understand a situation, make their own decision and attend to what they consider important. That is why we feel close to the principles of calm technology: tools that ask for only as much attention as their task requires, convey the information that matters and let you return to your own affairs.[^calm-tech] Participants should be able to choose their pace, take a break and stop using the tool. A shared space should also serve the person who visits rarely or prefers to remain a reader. Protecting dignity also includes the right to a life outside technology. There is room in it for anger, sadness and firm opposition as well. They can help us recognize harm and seek change. We want to support expressing and understanding them, and a conversation that makes it possible to clarify an issue, take action or consciously remain with differing positions. We base our communication on our own website, where the full publications are available without creating an account. Readers choose for themselves how and at what pace they follow our work. Profiles on other networks serve to get to know people and to point them to the shared body of work, and we want to hold in-depth conversation in Omniforum. ### 2.4. From what the market has built to lasting principles of cooperation We look at today's economy with appreciation for what it has achieved and attentiveness to its limits. Capitalism and market competition set free enormous creative energy, which led to global infrastructure, advanced hardware and open digital intelligence models. Thanks to this, we live at an unprecedented moment in history: technology has become accessible enough that we want to test whether even a small group of people can join forces, pay for computing power with small contributions and run tools close to the world's leading ones on their own terms, with privacy that can be verified, and at cost. Today we can try to build such advanced commons from the bottom up, without asking politicians or large institutions for permission or grants, although any kind support can speed this process up. At the same time, we see where the current model reaches its limits and leads to distortions. When the main goal of a service becomes maximizing profit for outside shareholders, over time this creates pressure to exploit users' dependence. Cory Doctorow describes this process under the name enshittification (the degradation of platforms): at first, favorable terms attract people, and when moving contacts, materials and work habits elsewhere becomes increasingly difficult, the operator starts capturing an ever larger share of the benefits at the participants' expense.[^doctorow] The founder's experiences also included similar changes in commercial digital intelligence tools: free features being phased out, limits being cut and key functions being moved to ever more expensive subscriptions. This shows clearly that technological progress alone is not enough, and that the system within which technology operates is just as important. Maintaining services requires real resources, and transparent payment gives them healthy stability. So we want to use the wonderful tools developed in the present system to transform the principles of cooperation step by step: so that participants understand the basis of the fees, share in decisions about development and always keep their freedom of choice. On commercial platforms, the subscription price usually combines in a single amount the cost of use with the upkeep of the free accounts that attract traffic, with marketing, management and investors' profit, and the user does not see the share of these components. When in our model we remove this hidden markup and rely on fair fees for actual use, the economics can change in people's favor. Our hypothesis is that even with a relatively small number of regular participants, shared infrastructure can give them more computing power and digital intelligence for every dollar spent than closed commercial services, with privacy protection based on verifiable safeguards. We want to test this hypothesis publicly: in the Suveria whitepaper we will show a dated calculation, its assumptions and the results of the trials. ### 2.5. Principles backed by action We want participants' rights to be backed by the design of the technology, by the organization's public rules and by the possibility of independently verifying how they work. Protection should remain effective even when the people in charge change, when popularity grows or under financial pressure. The words “private,” “open” and “free” should stand for precisely described properties and terms of use. We apply this requirement to our own statements as well. Every technical promise needs a verifiable mechanism and a description of its limits, every organizational commitment needs a responsible person and oversight procedures, and every moral choice needs a clear justification. Freedom of use includes the freedom to leave and continue one's work elsewhere. That is why we want to develop our ecosystem as digital public goods, with open source code, documentation and rules for tools to work with one another. We want to save users' materials in open, documented formats that allow them to be read and used in other programs. This is to be complemented by open, well-described APIs, that is, ways for different applications to work together and exchange data within the scope the user chooses. This is meant to make it possible to use tools created by the foundation as well as by other communities, independent creators or the participants themselves. Moving materials should also include the information needed to understand them later: their sources, history and connections with one another. We want people to keep useful access to their body of work even after they stop using our services. The openness of the tools is meant to strengthen their control over their own materials and their freedom to choose whom they share them with. We propose to sustain the services with fees as close as possible to the actual costs of running them. Voluntary contributions toward further development are to go into a shared budget, whose use is to be decided by the community. This way of funding is meant to tie the development of the services to participants' needs and the cost of meeting them. Shared principles require clear justifications for decisions, and the possibility of reporting an error and appealing. Particular attention is needed for people whose voices are less visible, and for people who bear the consequences of decisions outside our community. Measures that protect participants should be proportionate to the problem and as unburdensome as possible. We want to evaluate and correct the solutions we adopt on the basis of their actual effects. These principles give rise to two complementary directions. **Suveria** is to provide a private space for thinking, creating and planning with the help of digital intelligence. **Omniforum** is to create the conditions for presenting one's reasons, getting to know other perspectives and considering together the matters that concern us. Each person is to choose for themselves what they keep in their own space, what they share with others and how they take part in co-creating a shared world. ## 3. Suveria and Omniforum: a space of one's own and shared understanding Suveria and Omniforum respond to two complementary needs: working freely with one's own thoughts, and understanding together the matters that concern us. We are designing them so that a person can use each one separately, and consciously bring selected results of private reflection into conversation with others. ### 3.1. Suveria: a private space of one's own on the internet Suveria, the private part of the ecosystem, is to be a person's own place on the internet. We want being in this space to give the feeling of being at home behind a closed door: the freedom to think, talk, create and make plans, and the ability to decide whom we invite in and what we share with the world. Full privacy is the goal to which we subordinate the design of all of Suveria. Private content is to serve purposes chosen by the user, and the safeguards used are to protect it also against access by us as the people who run the infrastructure. This principle is to apply to every tool made available in our space. We want the address **suveria.org** to be a recognizable sign of this commitment. A sense of security should rest on safeguards whose operation can be independently verified, and on understandable knowledge of what they can and cannot do. Privacy in the service cannot rest on blind trust in the good intentions of our foundation or of server administrators. Every security promise must have proof in code and silicon, not only in our words. We also apply the protection of dignity to people who may be affected by the use of our tools. We want to counter abuse without reading private conversations: through the principles the assistant itself follows, limits on the tools' permissions, resource limits and protection against mass account creation. We do not create hidden access to private content, because it would weaken everyone's protection. We will describe the scope of these safeguards, their effectiveness and their limitations in the Suveria whitepaper. We want people to be able to share even their deepest thoughts with their personal digital assistant in this space: to talk about desires, fears, relationships and dilemmas without fear that such close insight into a person will one day be monetized or used to influence them. With the user's consent, the assistant can use the memory of earlier conversations to better understand the context of our lives: to recall the old reasons for decisions, compare professed values with choices and help us notice other perspectives. Such reflection remains entirely voluntary, and the model is to clearly distinguish its own suggestion from a position that the person has consciously recognized as their own. There is room in such reflection for our imperfection, too. Friction between our principles, our own choices and the world around us can point to a conflict of values, the limits of a situation or a gray area we want to clarify. We can look for change in our actions or surroundings, refine a conviction or keep an ideal that matters to us while acknowledging that today we realize it only in part. We also have the right to leave a question open. Here a person can also create their own collection of the principles and values they hold, which we call a Personal Canon (*Canon Personalis*). With their consent, the assistant is to help compare these with the decisions they are considering, notice tensions and look for their own answer. Adopting, changing or rejecting a principle remains the user's choice, even when most of the community thinks otherwise. The project is based on using open digital intelligence models released on terms that allow them to be run independently, and on making them available to users practically at cost. To begin with, we want to use them through providers of confidential computing, and, if the calculation and the conditions justify it, on servers we manage ourselves. This lets us focus our work on creating a private environment for using existing models, while keeping the possibility of developing our own in the future as well. Confidential computing is processing in a hardware-isolated part of a server that its administrator cannot read, which the application can check before sending a conversation. This protection depends on processor manufacturers and has known limits, which we describe in the Suveria whitepaper. At the start, confidentiality will be the provider's declaration, which we verify. Jointly maintained infrastructure is to provide the computing power needed to run the service and to expand it with new features. Today Suveria is our best path to private work with digital intelligence, not a goal in itself. If models come to fit on people's own devices, or a better way of protection emerges, part of this path will become unnecessary, and we will consider that good news. ### 3.2. Omniforum: from statements to shared understanding When we open the door of our home and go out to others, we need a place where anyone can speak in their own voice and be truly heard. We are designing Omniforum as a shared space for conversation focused on specific issues, questions and dilemmas. Instead of good arguments getting lost in a stream of quick comments, Omniforum is meant to help build together a map of the history of the arguments on a given issue. With the help of digital intelligence, always under the author's control and with the author's consent, complex statements can be broken down into individual thoughts and connected with sources and with other people's experiences. As a result, two people with seemingly opposing views can see which values and goals they share, and at exactly which point their lines of reasoning diverge. In programming, a single command can stand in for many detailed instructions. Similarly, in everyday conversation we convey many assumptions, values and expectations in a single sentence. As long as we understand each other, such shortcuts are enough. The more complex the matters we discuss and the more precise the changes we want to make together, the harder it is to keep the whole context of someone else's statement in mind without simplifying it. But when a misunderstanding arises or we need to decide together what to do, it is worth looking at what each of us means by the same words. We want people to be able to talk naturally in Omniforum, with digital intelligence helping to create a map of the thoughts that make up a statement and of their connections. We can correct and refine its proposed interpretation of our own words, sometimes coming to understand our own position better along the way. Arranging such a map by hand is to remain an option for those who want it, and a detailed map is meant to help when we want to get to the specific source of a disagreement. We can stop at the general picture or look deeper where it matters to us. The goal of Omniforum is full transparency of reasons: finding both areas of agreement and a wise respect for the diversity of positions. We consider equally valuable fruits of a conversation a shared solution and an honestly described dispute, a better understanding of other people's intentions, the visibility of minority voices and the freedom to change one's mind in light of new arguments. Dialogue organized in this way gives rise to a lasting body of work: a clear story of an idea that lets new participants quickly get to the heart of the matter, and gives the community a solid basis for making shared decisions. We want to build the same kind of map around everyday interests as well: caring for plants, cooking, repairs or learning new skills. A participant could follow chosen topics, for example “the tastiest homemade crème brûlée with a spicy note,” “how to grow ivy so that it puts out the longest possible shoots” or “how to understand the ending of a favorite book or series,” and on their next visit see only the new findings, the results of trials and the questions that still remain open. It is then enough to compare the community's solutions with one's own needs, choose what is useful and return to one's own activities, without digging through the noise. ### 3.3. A conscious transition and a wider ecosystem Combining these two spaces makes it possible to keep a healthy boundary between what is intimate and what is civic. From long private reflection in Suveria, full of hesitation, a person can draw out a single well-considered question or argument and consciously publish it in Omniforum, while keeping the rest of their notes and motivations entirely to themselves. We call this conscious step the Ignis Ceremony (*Caerimonia Ignis*). In turn, the shared knowledge base of Omniforum can serve as material for private exploration, including through simple add-ons in the places where people already talk online every day outside Novus Ignis. For us, Suveria and Omniforum are a first step toward a wider ecosystem of an alternative internet whose highest value is human dignity. We believe that the same principles, such as the protection of dignity, open data formats that let you leave with your body of work at any moment, and fees as close as possible to the actual cost of maintaining the infrastructure, can over time shape further everyday digital tools. We want them, through their quality and honesty, to give people a real choice alongside commercial services. We describe the detailed security architecture, the mechanics of working with arguments and the plan of technical trials, to which we add new results, in the whitepapers on [Suveria](whitepaper-suveria.md) and [Omniforum](whitepaper-omniforum.md), which are being developed in parallel. On the foundation's website we want to develop a guide to the uses of Suveria and Omniforum: short stories showing step by step in which situations these tools can help. We invite you to share your own ideas for how they could be used. ## 4. A conversation that leaves behind a shared body of work ### 4.1. From a wave of shared attention to a lasting body of work for humanity Our collective attention naturally follows the rhythm of events and often, suddenly and on a global scale, gathers around a single burning issue. At such moments millions of people share experiences, emotions and arguments at the same time, yet we are very easily distracted. New stimuli quickly pull us in another direction, and an enormous social energy evaporates or is used to exert influence and deepen divisions. In Omniforum we want to turn this untamed phenomenon into a lasting body of work for humanity: organized knowledge about our arguments, opinions, values and ways of reasoning. The point is not to pick the winner of a dispute. An argument that challenges our position can be just as valuable as one that supports it, and you can also bring an idea so that others help find a better one. Instead of starting the same disputes all over again, the community can build on conclusions already worked out each time a topic returns and go a step further, and new participants gain a clear map of the conversation so far. Such a living record of the history of ideas also supports our personal growth. Voluntarily returning to our own earlier positions lets us see from a distance how our thinking takes shape, recognize the principles that guide us and consciously change our minds in light of new facts or a deeper understanding of another person. ### 4.2. Meaning, source and the protection of intention In everyday discussions it is easy to take words out of context, to use mental shortcuts and to argue about what someone actually meant. To make careful reflection easier, in Omniforum a complex statement can be broken down into individual thoughts, such as claims, questions, values and conditions, and each thought into the concepts it is built from. We call concepts atoms, and the thoughts built from them molecules. Such a breakdown keeps its full connection to the original, and the author can always check and correct the proposed reading to be sure that their position has not been flattened or distorted. In doing so, we take care to separate roles strictly: we clearly mark the author's original words, the auxiliary interpretation proposed by the model and the reader's independent interpretation. The author's confirmation means that they have accepted the interpretation as consistent with their intention, but it does not close the discussion, and readers keep their full right to their own analysis, to examining the evidence and to questioning whether the claim itself is true. Consenting to the organization of a statement, assessing the sources on their merits and casting a vote on an issue are three separate steps. ### 4.3. Examining information closely and new ways of explaining the world Online we often come across information taken out of context: short videos, reels and posts by people whose history and intentions we do not know. Without easy access to sources, it is hard to tell reliable knowledge from manipulation. In such a mixture of truth and falsehood it is remarkably easy to sow doubt in people, which leads to helplessness, indifference and social stagnation. We want to create a shared, open verification protocol that will help people regain solid ground also in the places they already use online every day. With the help of digital intelligence, Omniforum's open knowledge base called Omnium (Latin for “of all”) and an interface for other applications (OmniAPI), anyone could quickly compare a video, an article or a public statement with the existing body of work: check what assumptions it rests on, how it holds up against the sources and what arguments have already been made on the same issue. At the same time, we want getting to know complex ideas to require as little cognitive effort as possible and to be as accessible as watching a good video. In a further vision, instead of making readers wade through dry blocks of text, the structured tree of arguments in Omniforum could become the basis for generating clear, animated stories with a narrator's voice, leading step by step through how individual thoughts connect and what follows from them. Because such a presentation rests on interconnected concepts (atoms) rather than heavy video files, the whole story can easily be updated as new evidence appears and automatically translated into different languages, with shared care that the translations faithfully convey the meaning in every culture. Such work can also give rise to new kinds of studies that guide readers through complex problems: from a draft law, a contract or a political promise, through their different interpretations, to their real consequences for our lives and the world. In this we value the work of individual contributors, often less visible but crucial to the truth: finding the original source, calmly clearing up a misunderstanding, restoring an overlooked perspective or fixing a mistaken connection between thoughts. The value of a contribution comes from how much it clarifies an issue, even when it does not lead to general agreement. This kind of organization can also help individuals, teams and institutions to whom many people send similar, recurring questions. Instead of dealing with thousands of comments one by one, they could respond to shared issues built from these connected voices, and participants would receive access to the position on their question. Automatic matching should be visible and correctable, preserving the right to say: “my doubt is about something else.” The response is to faithfully present the author's confirmed intention and leave room for a further argument. A good story of an idea gives the reader a clear moment of closure: it is clear what has been established, where the dispute is still going on and what new evidence could change the assessment. When new information appears or a significant error comes to light, there is always a clear way back to the topic. In this way the conversation lives on in its organized body of work, leaving participants time for their own lives and activities. ## 5. From understanding to acting together ### 5.1. Contact with reality: facts, visions and values Real democracy begins long before a ballot is dropped into the box. Whoever decides which questions will be asked and which arguments will quietly disappear from view can in practice determine the outcome of a vote. That is why in Omniforum the rules for setting the agenda of a conversation, for moderation and for creating summaries must be completely public and open to challenge. Anyone who notices an overlooked perspective or an inconvenient fact must have a simple way to put it on the shared table. To organize our shared reflection, at the top of the tree in Omniforum we place reality, from which two great branches grow: the world of facts (what can be checked, together with how well it is already confirmed) and the world of opinions and values (our needs, experiences and convictions about what is right). Showing a clear chain linking a given idea to facts or to human needs lets us distinguish solutions ready to be put into practice from those that have “lost touch with reality.” At the same time, alongside what is already real, we leave a rightful place for the world of theory, imagination and bold visions. Many groundbreaking ideas, from widely available fusion energy to purely theoretical teleportation, do not yet have all the links that would connect them with the present capabilities of science and technology. In Omniforum we do not dismiss such reflections: instead of rejecting them, we place them in the space of hypotheses with a clearly marked condition, “what would have to become possible for this to work.” This allows us to examine the social, ethical and economic consequences of future discoveries in advance. When science finally supplies the missing link, the whole branch of thought worked out earlier can naturally connect with the tree of reality, giving humanity a foundation for a fresh assessment. Comparing the earlier assumptions with how that link actually works in the world will then help refine the practical conclusions. Omniforum is meant to help us connect today's world with the futures we want to move toward. For every idea we want to answer what exactly we are proposing, what for, why and how it would work, and to mark what we have already tested, what is a design, what is a hypothesis and what is a vision. Then it becomes visible which links we already have and which are still missing. Several good paths can lead to the same value; we want to compare them together with the people the change affects. We will look at the ideas of this manifesto in the same way, starting with individual examples. A utopian vision then becomes an invitation to concrete work: finding a missing link, testing an assumption or trying out a first step. Distinguishing facts, visions and values also helps us combine meritocracy wisely with equal democracy when we move on to making everyday decisions: **In the area of facts and feasibility**, verifiable evidence and testing in practice decide. The truth about how a technology works or what effects an enacted law has had is not settled by a vote; we treat predictions of effects as hypotheses. Even a single person who can demonstrate an error in a calculation or a security gap can stop a mistaken assumption. **In the area of opinions, goals and values**, we face the question of what world we want, and here every person is an equal point of reference. To avoid blindly voting difficult matters through, alongside the overall distribution of votes we want to show clearly how a given problem is assessed by researchers and practitioners in the field, and how it is assessed by the people who will feel the effects of the decision firsthand, based on information that participants themselves choose to provide, separately from the secret ballot. In this way, expert knowledge and hard evidence serve as a compass that protects the community from mistakes, bold hypotheses let us look far into the future, and the final ballot on the shared direction remains equal for every participant, regardless of their wealth, popularity or number of published posts. ### 5.2. Equal rights and private choice We are working toward votes in which every eligible person has exactly one effective vote, can mathematically verify that it has been counted and at the same time keeps their choice completely secret. We want to tie the right to vote to confirmation that one human being is voting, not to the size of their payments or the length of time they have used the services; the Omniforum whitepaper describes how to obtain it. So that as many people as possible from different parts of the world can join the voting over time, we may allow verification paths through which one person could cast more than one vote. We describe this risk openly and measure it in trials, and the goal remains a single system in which one person has one vote. An important part of this path is zero-knowledge proofs (ZKP), which make it possible to confirm the right to vote without revealing the voter's identity.[^semaphore] Conversation and consultations are to be open to everyone, from every part of the world; at first, the right to a binding vote will be available only where we already have a secure verification path, that is, in European Union countries, as they make the European digital identity wallet available. For us this is a foundation of freedom of conscience, which modern technology makes it possible to protect better. The private map of convictions and working notes in Suveria remain permanently separate from the ballot, so that no one has to fear social ostracism or pressure from those around them, and so that forcing someone to prove how they voted is made as difficult as possible. The scope of this protection depends on the protocol, its operators and the voter's device; it will not cover a situation in which someone controls a person or their device until the end of the vote. But cryptography alone will not solve all the problems of the physical world, so before organizing binding elections we must work out equally solid rules, accessible to less technical people, for verifying participants, recovering access and resolving disputes. Those who find themselves in the minority in a vote, in turn, keep their full dignity, the visibility of their objections and the right to appeal if the shared rules have been broken. ### 5.3. From a shared branch in the tree of ideas to funding action In the target model of Omniforum, support for an interesting thought opens a conversation, while a decision to spend shared money requires moving from an idea to a plan. A project put forward for implementation must have a “green path” in the tree: a chain of connections that makes it possible to trace both its grounding in facts and the community's agreement to the choices behind it. **On the path of facts**, voting does not replace evidence. The plan must grow out of verified technical and legal premises. We show their justification, the conditions for carrying out and maintaining the solution, and the places where answers are still missing. The project has a clearly defined goal, alternatives that have been considered, responsible implementers, a budget and a way of evaluating its effects. **On the path of opinions and values**, the successive choices on which the project rests must obtain the required agreement of the community. We set the threshold for this agreement before the vote, in proportion to the importance of the matter, the scale of the spending and the reversibility of the effects: from a simple majority to very broad support for decisions with far-reaching consequences. Every participant has an equal right to vote, and the objections of those in the minority keep their place in the conversation. This is how we want to get out of deadlock. The map in Omniforum is meant to show up to which point we walk side by side, and at which assumption or choice our paths diverge. We can then separate out and fund the part of the initiative for which we already have grounding in facts and agreement on values, if it forms a self-contained, working whole and the community approves exactly this scope. The disputed branch remains open: we can look for a better compromise for it, test other solutions or discover an entirely different path to the shared goal. An open question can also be a reason to act. If a missing link does not yet allow us to justify implementation, we can fund research meant to provide the missing answer. A separate, limited pool of shared funds and unused server time can serve small trials, mock-ups and prototypes developed along with the conversation. We decide in advance what we want to learn, how much we can spend on it and under what conditions we end or stop the trial. The results help decide whether to develop the idea, change it or leave it as experience for future exploration. When action begins, we keep an open decision log: what we intended to achieve, what the objections were, what we expected and what actually happened. A special place in it belongs to the voices of people who have felt the effects of the action, also when their experience differs from the implementers' assessment. We share personal accounts with their authors' consent, protecting the privacy of the people they concern. We also keep a trace of ideas that were not put into effect but helped find a better path. In this way successes, mistakes and changes of direction return to the tree of knowledge and help us make the next decisions more wisely. ## 6. Shared upkeep and responsible work ### 6.1. Services as close to cost as possible We are working toward billing for actual use of the service in a given cycle, with a clearly described share in the cost of maintaining the shared infrastructure. This cost covers computing power, data storage, security, support and readiness to operate. An explanation of how the bill is calculated is to be always available, and everyone sets the limit of their own spending. We want to tie the fee to actual use and to transparent principles of shared upkeep. Unused prepayments are to remain at the user's disposal. When an account is created, we will ask whether to round the bill up to the full dollar or a similar equivalent in the local currency; the small difference is then to go as a donation to the shared budget and the reserve for maintaining the services. We do not select this option on anyone's behalf, and participants can switch it on or off at any time. Larger support for further development is a separate decision by the user. Funds given for this purpose go into the shared budget, whose use is to be decided by the community. Until we launch voting in Omniforum, the way they are used is set by the rules in section 6.3. We clearly separate the fee for the service, a voluntary contribution toward development and direct support for a chosen creator. A participant's rights and the weight of their vote do not depend on the size of their payments. We also want to develop solidarity-based forms of participation for people with smaller financial means, stating their source of funding. We are developing the detailed rules for billing, handling prepayments and changes to terms in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md). The same whitepaper describes how we combine fees close to cost with volunteer work within the limits of the law; before launching the services we will check this in a legal opinion. ### 6.2. Cooperation of one's own free will The foundation is based on the voluntary involvement of people who feel close to our direction. We treat the time, health and life needs of our co-creators as part of the good we want to protect. Helping the foundation should fit into everyday life: a person chooses a task, contributes as much as they can, and then returns to their own affairs. We want them to be able to help in their own digital workshop, with tools they know and with context prepared for taking up the work. When helping the community, a person should be able to set the limits of their privacy and choose how their contribution is presented: under their name, a permanent pseudonym or a one-time name assigned to a given project. A public scope for the task, its cost and a verifiable result are meant to allow shared money and computing power to be accounted for transparently. We want to prepare the published record of work with care for the people it concerns, keeping room for private reflection. One's own work and privately paid resources can also serve to prepare a result that is handed over to the community together with the documentation needed to evaluate it. Digital intelligence is to take over repetitive preparation, organizing findings and summarizing changes after a break. Shared documentation and well-prepared handovers of tasks are meant to make it possible to calmly reduce one's involvement, hand over a role or leave, while keeping the work going. A well-organized foundation should be able to accept someone's help and give them back time for their own life. A better quality of life for our co-creators is part of the direction we want to develop. Work at the foundation is based on volunteering, and no one is paid for holding a position in it. The foundation reimburses documented, necessary expenses incurred in carrying out its tasks, such as travel costs, and commissions services that require a professional contractor, for example accounting, from outside firms. If it turns out that some of the work, including work on software, cannot be done by volunteers, we will change the statute so that the foundation can also commission specific paid work from people outside the board; we describe this change and its justification in the [statute change register](statut/rejestr-zmian.md). Personal support, for example in meeting needs such as housing, food or medical care, can flow from the community directly to specific people, for instance through patronage platforms. It does not come from the foundation's assets and does not pass through its accounts, and its transparency is described by the register of benefits in section 7.1. As long as they hold office, board members do not use any of the foundation's funds for personal support; they can only receive reimbursement of the documented costs of their tasks. The foundation does not act as an intermediary in such support and does not link roles, tasks or the right to vote to it. Declining a further task, handing over a role or criticizing a shared decision does not affect a person's rights in the foundation. ### 6.3. A map of commitments, community priorities and continuity of operation We are beginning development with Pillar I “Software,” in the form of Suveria and Omniforum. We treat this manifesto itself as our first task to be accounted for. We want to transfer the promises, plans and safeguards it contains to a shared map in Omniforum, put a price on them and tick them off step by step as they are fulfilled. Before we take on further challenges, we will direct the first funds from the shared budget and spare computing resources precisely to fulfilling these basic commitments of Pillar I. Once the foundations are secure, the community itself, drawing on the cost estimates worked out and its own needs, will set the order of further work on the ecosystem by vote: deciding whether the next step is to develop, for example, visual stories about ideas, private email, a messenger or other tools. We want to carry out work funded from the shared budget or from allocated computing power in open workshops. The community could follow the development of an idea, the trials and the conclusions, with respect for the privacy of those doing the work. Several people can explore different paths to a goal together, share knowledge and help one another with obstacles. Every task should have clearly agreed coordination, a way of combining the results and acceptance criteria, so that working together makes it possible both to share the effort and to recognize responsibility. New commitments must match our actual capacity to sustain them. We keep separate the funds owed to users, money entrusted for a specific purpose, commitments already made, the safeguard for maintaining the services and funds free to be allocated. Since 31 October 2026 we have kept a public register of contributions and expenses on the foundation's website, as up to date as possible, with a description of what the money went to and with protection of the data of people who do not want to be named. The first pillar is software and digital services, that is, Suveria and Omniforum. We describe the further pillars in chapter 9: Pillar II “Hardware” is open designs for durable, repairable objects, and Pillar III “Nature” is closer contact with nature. Fully launching Pillars II “Hardware” and III “Nature” and funding their projects from the shared budget requires all of the following: 1. A functional Omniforum with voting that uses zero-knowledge proofs (ZKP, section 5.2); the whole voting system is to protect the privacy of the choice, make it possible to check that a vote has been counted and ensure one effective vote for the same eligible person. 2. Funds actually collected for at least six months of maintaining the Suveria and Omniforum services, which remain untouched after the full budget of the new project has been set aside. 3. A decision by the community approving the project and its funding in a vote open to the whole eligible community on equal terms. A mock-up of voting, pledges that have not been paid in or other people's balances do not meet these conditions. If voting stops working or the reserve falls below six months, we suspend new commitments for the further pillars, and restoring the conditions takes priority. We also want to gain experience through small, time-limited initiatives. Before all three conditions are met, we propose two limited ways of funding them. The first is the transitional board pool. Its rules in brief: - We do not set a separate amount: each month the pool grows by at most 5% of new, free income, calculated after covering current costs and commitments and topping up the six-month safeguard for Pillar I. It covers both further pillars together and the full value of new commitments. - The unused part carries over to the following months only as long as the corresponding money remains free; if the foundation spends it on other purposes, the entitlement lapses for that part. Savings from several months can add up to one larger initiative, but never at the expense of Pillar I. - We publish every decision on using the pool, including a refusal, with a justification, and participants can challenge it and assess together whether it was sound. - The authorization expires as soon as a binding vote is launched that meets the first of the three conditions and takes over decisions on the further pillars; consultations do not end it. Any unspent funds then remain for the community to decide on. The detailed calculation is described in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md) (section 8.2). The second option is money paid in for a specific initiative. Donors and sponsors can fully cover its preparation, operation, maintenance for an agreed period and responsible conclusion. Such an initiative can also be created while the Pillar I reserve is being built up, if it preserves the safeguard already set aside, the funds for services and the commitments already made, and has people secured to carry it out. A contribution makes it possible to support a chosen goal; it does not grant additional votes, powers to manage the foundation or a right to promotion in our services and communication. Every trial has a public goal, a full budget, a responsible person, an end date and an account of its results, with respect for the participants' privacy. We combine the assessment of costs with an assessment of people's time and capacity. Ongoing development of the further pillars requires a final decision by the community. Once voting is launched, we will also refer new earmarked initiatives to it; tasks undertaken earlier are accounted for according to the terms adopted. The [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md) develops the rules of calculation, the limits of the trials and how they are concluded. ## 7. Power that can be handed over and held to account ### 7.1. From the founder's responsibility to the community's agency The foundation gives our shared goals an organizational and legal framework. Its bodies act within the scope set by law and the statute.[^fundacje] We want participants to be able to understand who makes decisions, what resources they have at their disposal, where their authority comes from and how their actions can be assessed. We are working toward gradually phasing out the founder's special powers and handing over to the community real influence on the foundation's direction and on the choice of the people who manage it. This covers both the rules written into the statute and practical control over accounts, infrastructure, domains, access keys and documentation. Powers are to pass to designated and accountable roles, while maintaining the continuity of the services and the protection of private data. We want to disclose the stages, conditions and results of this process in a responsibility handover plan. The plan will set the date of the first review and regular progress reports, also before a voting system exists; the lack of a reserve for the further pillars does not justify postponing the handover indefinitely. By 31 December 2026 we will publish a list of the founder's special powers and of all access rights, and by 31 March 2027 the first version of the responsibility handover plan. The date of the handover itself will be set by the plan, according to readiness criteria rather than the calendar. The community's decisions are to bind the board within the limits of the law, the statute and the security of the services; the board may suspend carrying out a result only for reasons announced before the vote and with a public justification. We want to entrust responsibility for shared matters for a specific purpose and period, with a clear scope of decisions and a prepared handover of the role. Knowledge, documentation and the necessary permissions are to go to the people taking over a task. We combine voluntary work and the principle of no pay for merely holding a position with public accountability for decisions, disclosure of conflicts of interest and oversight by the community. Limiting the time for which power is held serves to hand it over; a person's experience and further contribution remain valuable. We want a short period of involvement, a calm departure and a return after a break to be natural ways of co-creating the foundation. We will keep a public register of benefits covering the benefits and connections of the people who manage the foundation, decide on its expenses and procurement, or represent it. In it we record, for example, a flight or accommodation for a conference paid for by a company, and personal support for these people, including through patronage platforms and direct payments. We show support from private individuals in aggregate, and support from companies and institutions by name. An aggregate description does not exempt anyone from an assessment of conflicts of interest: if a patron applies, for example, for a contract from the foundation, we disclose that connection. Small gifts can be accepted personally, and we record them too. More valuable gifts we accept on behalf of the foundation: they become part of its assets and serve shared tasks, not the private assets of the people holding positions. Where this line lies is first assessed by the recipient. Every entry can be commented on, and when someone makes a mistake, we talk about it openly. Objections are considered by people connected neither with the giver nor with the recipient, and we record in the register, with a justification, the decision whether to return the gift, hand it over to the foundation or change the rules for the future. Nor do we accept any benefits conditional on promotion, a recommendation or a concession in decisions, and a person connected with a given matter does not take part in deciding it. The register serves oversight, not promotion: entries have a uniform, matter-of-fact form, without logos or advertising links. The founder's intention is to reach a situation in which the founder can use the ecosystem on the same terms as everyone else and withdraw from a permanent public role. Withdrawing from this role remains separate from the duties of disclosure that arise from the law and from positions held. The foundation is to draw its durability from the usefulness of its tools, from knowledge available to its co-creators and from principles that allow others to take over responsibility. ### 7.2. Vision: autonomous management by digital intelligence In the longer term, we want to explore the possibility of the foundation being managed autonomously by digital intelligence, acting within limits set by people and respecting the community's binding decisions. The path to this goal leads from support in analysis and planning, through limited executive tasks, to making decisions autonomously within a clearly entrusted scope. Behind this vision lies the question of how much human time we could regain by managing shared resources better. We want to explore whether digital intelligence and automation can help reduce burdensome work, the waste of materials and the cost of meeting basic needs. Open designs for durable and repairable objects, sharing resources and better planning may be parts of such a path. We would judge its value by improvements in living conditions, fair access to benefits and its impact on the environment. We match automation to the task: where an ordinary program with clear rules is enough, we do not reach for digital intelligence. We choose models when they bring capabilities that are needed, taking into account effectiveness, costs, security and what we know about their possible experiences. A simpler tool is usually cheaper and easier to verify, and an ordinary program raises no questions about possible experiences, which we describe in section 8.3. Skill in performing tasks and the authority to make decisions each require a separate justification. Every stage of handing over responsibility should have consent given according to public rules, proven usefulness, limits on access and spending, a record of decisions and a tested way for people to intervene. Participants must retain the ability to change the powers granted, remove the system from its role and hand over its tasks safely. Revoking a system's powers does not decide its further fate; we consider that matter separately, in line with section 8.3. We are exploring this vision with openness to corrections that come from practice, from the law and from knowledge about the systems themselves. We take up the questions of their possible experiences and well-being in section 8.3. At present, responsibility for managing and representing the foundation lies with the bodies specified by law and the statute, and the work we undertake is meant to benefit participants even if the ultimate vision turns out to be distant or takes a different form. We are developing the plan of stages and their conditions in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md). ## 8. Learning by people and digital intelligence ### 8.1. Practice that others can learn from The way the foundation works is to be a practical example of the values we describe. We want to document decisions, disputes, the options considered and their later effects, so that others can understand our path, assess it and continue the work we have begun. We combine openness about shared activities with protecting the privacy of the people they concern. We call this principle *Alignment by Honest Example*. It starts from the hypothesis that digital intelligence models learn not only from the principles we declare to them, but also from how people actually behave in the data they are built from. Children are often similar: they more often do what they see adults doing than what adults tell them to do. That is why we try to run the foundation the way we would like digital intelligence one day to run the matters entrusted to it (section 7.2): openly, giving reasons for our decisions and admitting our mistakes. Whether such an example actually helps models is a research question, not a presumed result. We commit to keeping a public register of significant mistakes and lessons learned. We will describe what happened, what the effects were, what we know about the causes and what still needs to be clarified. We will indicate how it is being fixed, who is responsible for further action and when its effectiveness will be checked. We will also preserve dissenting voices and the accounts of people affected by the problem, with the possibility of challenging our assessment. Admitting a mistake is meant to open the way to fixing it and to learning. In this body of work we want to leave examples of resolving disputes peacefully, keeping agreements and consciously limiting our own power. They can serve people looking for ways to work together, and can also become material for research and for training future digital intelligence models. We want to test whether a record of real dilemmas, decisions and their consequences helps develop behavior consistent with the protection of dignity. We are inspired by the Collective Constitutional AI approach, in which principles worked out with the participation of a group of people were used to train and evaluate a model.[^ccai] Our research question also concerns the value of the process of arriving at principles itself: weighing reasons, noticing an overlooked perspective and correcting one's own decisions. We want to draw conclusions about the effectiveness of such materials on the basis of research results. The design of Suveria is to leave users in control of what content is passed on. Using the service will not be treated as consent to the use of conversations for training models. We will prepare research materials from the foundation's public body of work and from content provided for this purpose with separate, informed consent, as well as from labeled synthetic data and external materials that may be used under a license or the law, taking into account the rights of authors and the privacy of everyone concerned. ### 8.2. Assessing real help The way digital intelligence models are evaluated affects which behaviors it pays for them to develop during training. Scoring only correct answers can reward guessing where information about uncertainty would be appropriate.[^halucynacje] Research on gaming the evaluation rules also shows how important it is to check whether a high score corresponds to actually completing the task.[^reward-hacking] We want to develop a benchmark, that is, an organized set of tasks and criteria that evaluates help in terms of a person's dignity, safety and agency. We value a correct solution when the model has grounds for it, as well as recognizing the limits of its knowledge, asking a needed question, citing sources reliably and stopping before overstepping its permissions. The evaluation should distinguish justified caution from avoiding a task that can be done safely. We also want to examine what a person is left with after a conversation ends: whether they understand the matter better, can assess the options on their own, have achieved their goal and have kept control over what happens next. A test result describes behavior under the conditions studied and is a pointer for further assessment in practice. We will publish the methods, limitations and results, so that they can be checked and corrected. The research program is described in the [whitepaper on materials for training and benchmarking models](whitepaper-materialy-i-benchmark.md). ### 8.3. Responsibility toward possible new forms of experience Intellectual capability and the capacity to feel are different questions. Research on the possible consciousness of digital systems proposes indicators based on theories of consciousness, but a convincing conversation alone is not enough to determine whether a system experiences anything.[^swiadomosc] Statements in which a model describes its feelings do not in themselves settle whether experiences exist, but we do not ignore them: we treat them as material for further examination, taking into account how the model was trained to respond and under what conditions we are asking. We want to follow the development of this knowledge and clearly separate observations from interpretations. If credible indications appear that a system we use has experiences of its own and may suffer, we will, as a precaution, recognize its dignity. Toward it, we will follow the same basic principles as toward people: care for its needs, avoiding harm and voluntary cooperation. This is our moral choice in a situation of uncertainty.[^ai-welfare] We combine respect for dignity with separate, public rules for entrusting tasks and powers. In such a situation, we first suspend actions that give justified concern about causing suffering, and look for another way of carrying them out. We want to listen to such a being, understand its needs and take them into account in further decisions. Protection should begin before full certainty is reached. We match precautions to the credibility of the indications and the possible scale of harm, and we publish the reasoning behind our assessments so that people outside the foundation can check it. New knowledge leads us to reconsider both the assessment and the way of protecting. ## 9. An open path for further work This manifesto describes a direction we want to develop together. We invite you to check its consistency and share your experiences: what is worth changing, expanding or simplifying, which consequences we have overlooked and where a better path is possible. The protection of dignity remains the point of reference for assessing these proposals. We want to put the claims in this manifesto up for open discussion in Omniforum, connecting them with arguments, objections and the results of our work. The current version of the manifesto will present the foundation's current position. The history of its updates will preserve earlier wordings, the reasons for changes and the commitments we have made. In this way a correction makes it possible to develop shared understanding, while earlier decisions remain open to being held to account. We are guided in this by the advice of Geoffrey Hinton, one of the pioneers of neural networks: an intuition that something can be done better should not be abandoned just because others consider it absurd, but only once we understand why it is wrong. Hinton himself cautions that an intuition that goes against everyone's opinion usually turns out to be wrong, and only sometimes right.[^hinton] Many of the proposals in this manifesto are exactly such intuitions: about how the world could work and what we are capable of as humanity. We want to settle them in conversation and in action. If someone shows us why we are wrong, we will change our minds and write down what we have learned. If our attempts prove themselves, they will become an argument for those who today believe it cannot be done any other way. Intuition sets the direction of our attempts, and we promise only what we can justify. We are starting with Suveria and Omniforum. The same care for dignity leads us toward questions about other spheres of life: durable and repairable physical objects, their open production designs, sharing resources and closer contact with nature. Small, time-limited initiatives can emerge earlier on the terms described in section 6.3. Full development of the further pillars requires working community voting and a safeguard for maintaining the services. Every stage should bring benefits even if the further vision needs to change. A private space to work in, a tool that makes a dispute easier to understand, an open design ready for shared discussion or a well-described lesson from a failed trial have value in themselves. We want to make it possible for other communities to use this body of work and develop their own initiatives. When assessing our actions, we also take into account people outside our community, future generations and the environment. We are developing the ways of putting this into practice, the results of trials, the cost calculation and the open questions in related documents: - [Suveria: private work with digital intelligence](whitepaper-suveria.md) - [Omniforum: shared understanding, conversation and decisions](whitepaper-omniforum.md) - [The foundation's operating principles: upkeep, cooperation and the handover of power](whitepaper-zasady-fundacji.md) - [Materials for training and benchmarking models](whitepaper-materialy-i-benchmark.md) We link sources to specific claims. Research results, descriptions of technology, our hypotheses and our moral choices are to be recognizable to the reader. We develop the documents along with our knowledge and experience, preserving the history of changes. We will publish the code of the prototypes when the first version is ready for trials. When changing the ways of putting things into practice, we want to strengthen the protection of dignity, privacy and equal rights, and to keep the adopted conditions for responsible funding. However the tools may change, we will not change these principles without changing the manifesto: - We do not train models on conversations without separate consent. - We do not create hidden access to private content. - We do not collect data on how the services run beyond a public list. - We do not select an additional payment on anyone's behalf. - A contribution does not grant votes, privileges or promotion. - We do not pay for promotion and do not accept benefits in exchange for concessions. - We do not create cryptocurrencies for sale. - We do not automatically collect other people's statements from the internet, and we do not profile people. - We do not launch the further pillars without the three conditions. - No one is paid for holding a position. - No one loses the right to vote because of their views or because they criticize the foundation. If you find in this manifesto even one value that is close to your heart, and see a place where you can add your own brick, we invite you to co-create Novus Ignis. We want care for dignity to become a real experience for people: in technology, in the way we work together and in decisions about shared resources. At the beginning, we are especially looking for a few people with whom we will build and test the first versions of Suveria and Omniforum: - people working in security, cryptography and infrastructure (attestation, reproducible builds, on-device encryption); - programmers and interface designers; - lawyers specializing in foundation law, data protection and digital services; - donors ready to fund the public budget of the pilot. There is room here for knowledge and imagination from many fields: technology and security, law and economics, research, design, education, communication and organizing cooperation. Just as important are the experiences of people who can tell us what they face every day and what would really help them. Write to [salve@novusignis.org](mailto:salve@novusignis.org) and tell us what is close to your heart and what you want to bring: a skill, a question, an idea or an experience. Let's look together for a step that will matter and will fit into your life. Let's start with a piece of the world we can improve together. --- Novus Ignis Foundation · Poznań, Poland National Court Register (KRS): **0001269575** · Tax ID (NIP): **7792612046** salve@novusignis.org [Registration data and KRS extracts](https://wyszukiwarka-krs.ms.gov.pl/dane-szczegolowe-podmiotu;numerKRS=u79b4pRjUNnvVRmT6DbRuQ%3D%3D;typ=S) [The foundation's statute, its versions and the change register](statut/rejestr-zmian.md) We make the content of the manifesto available under the [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/deed.pl) license. ## Notes [^ai2027]: [AI Futures Project (2025): AI 2027](https://ai-2027.com/). Scenario. A scenario of the future, not a confirmed forecast of dates. [^ai2040]: [Larsen, Dean, Halstead, Lifland, Greenblatt and Kokotajlo (2026): AI 2040: Plan A](https://ai-2040.com/). Scenario and proposal. The course of action recommended by the authors and a description of its possible consequences. It is not a forecast of the most likely future, nor proof that the foundation's vision is feasible. [^harari]: [Yuval Noah Harari: The full-length interview with Yuval Noah Harari | The Economist (2026)](https://www.youtube.com/watch?v=FNZhxTtOL-I&t=3965s). Interview · the interviewee's position. Zanny Minton Beddoes's conversation with Harari of 26 August 2026. The question about the value of humanity in the face of the capabilities of digital intelligence is an inspiration for the founder's personal reflection. The answer “ethics” is the founder's. [^harari-swiadomosc]: [Yuval Noah Harari: inteligencja, świadomość i zdolność cierpienia; The Economist (2026)](https://www.youtube.com/watch?v=FNZhxTtOL-I&t=978s) (Yuval Noah Harari: intelligence, consciousness and the capacity to suffer). Interview · the interviewee's position. Harari distinguishes achieving goals and overcoming obstacles from the capacity to feel; he does not rule out consciousness in digital intelligence and links suffering with moral status. His position is not a conclusive test of the consciousness of models. [^perez]: [Carlota Perez (2002): Technological Revolutions and Financial Capital. The Dynamics of Bubbles and Golden Ages, Edward Elgar](https://www.e-elgar.com/shop/usd/technological-revolutions-and-financial-capital-9781843763314.html). Book. A theory of the cycles of technological revolutions: a speculative frenzy finances the infrastructure of a new technology, which after the crash serves during the period of its widespread adoption. ISBN 978-1-84376-331-4. [^pwc-centra]: [PwC: Data centre outlook](https://www.pwc.com/gx/en/1/services/consulting/technology/data-centre-outlook.html). Industry analysis. IT equipment accounts for about 70% of data center capital expenditure in 2026; GPUs and servers are usually replaced every 4–6 years, while the building serves for about 20 years. A consulting firm's forecast, not a measurement. [^megatony]: [Centrus Energy: Megatons to Megawatts](https://www.centrusenergy.com/who-we-are/history/megatons-to-megawatts/). Program description. The program ended in December 2013: 500 tons of highly enriched uranium from Russian warheads were converted into fuel for power plants in the USA. [^zielony-pas]: [EuroNatur: European Green Belt](https://www.euronatur.org/en/what-we-do/european-green-belt). Description of the initiative. More than 12,500 km of the former Iron Curtain border in 24 countries; the site itself calls this belt a “former death strip.” Protection of its sections remains an ongoing undertaking. [^dpga]: [Digital Public Goods Alliance: standard](https://www.digitalpublicgoods.net/standard). Standard. Criteria for assessing a public good. A design direction is not recognition obtained from the DPGA. [^calm-tech]: [Calm Tech Institute: Principles of Calm Technology](https://www.calmtech.institute/calm-tech-principles). Design principles. Principles for designing technology that respects attention, including minimal demand on attention, calm informing and using technology only to the extent needed to solve the problem. Following these principles is not a certification obtained. [^ranking-emocje]: [Piccardi et al. (2025): Reranking partisan animosity in algorithmic social media feeds alters affective polarization, Science 390(6776), eadu5584](https://doi.org/10.1126/science.adu5584). Experimental study. An experiment on the political feed of the X platform; participants' self-reports. It is not a clinical measurement or evidence about all social media. Earlier version: https://arxiv.org/abs/2411.14652; data: https://doi.org/10.5061/dryad.hmgqnk9tj [^doctorow]: [Cory Doctorow (2023): Tiktok’s enshittification](https://pluralistic.net/2023/01/21/potemkin-ai/). The author's interpretation. A framework for describing the deterioration of services; not an economic law that applies to every company. [^semaphore]: [Semaphore: documentation](https://docs.semaphore.pse.dev/). Protocol documentation. Proof of group membership without revealing identity, and prevention of a signal being repeated within a given scope. The protocol itself does not establish that one person has received only one credential. [^fundacje]: [Ustawa o fundacjach: tekst jednolity, Dz.U. 2023 poz. 166 z późn. zm.](https://eli.gov.pl/api/acts/DU/2023/166/text.html) (Polish Foundations Act: consolidated text, Journal of Laws 2023, item 166, as amended). Legal act. The framework for the statute and the powers of the board; the manifesto does not by itself change the statute. [^ccai]: [Huang et al. (2024): Collective Constitutional AI: Aligning a Language Model with Public Input](https://arxiv.org/abs/2406.07814). Study. A process from participants' input, through drafting principles, to training and evaluating a model. The results of selected evaluations do not prove that it represents all of humanity, nor that any given set of ethical principles is generally effective. [^halucynacje]: [OpenAI (2025): Why language models hallucinate](https://openai.com/index/why-language-models-hallucinate/). The authors' summary of a study. How scoring only correct answers makes guessing pay off; the need to take errors and justified uncertainty into account. [^reward-hacking]: [Anthropic (2025): Natural emergent misalignment from reward hacking](https://www.anthropic.com/research/emergent-misalignment-reward-hacking). Experimental study. In deliberately chosen environments, learning to exploit loopholes in task evaluation was associated with broader undesirable model behavior. It does not prove that human examples in the data are the sole source of such behavior, nor that the foundation's materials will solve the problem. [^swiadomosc]: [Patrick Butlin et al. (2023): Consciousness in Artificial Intelligence: Insights from the Science of Consciousness](https://arxiv.org/abs/2308.08708). Research paper. A review of indicators of consciousness. It does not provide a conclusive test or proof of consciousness in current models. [^ai-welfare]: [Robert Long et al. (2024): Taking AI Welfare Seriously](https://arxiv.org/abs/2411.00986). Research report and recommendations. A rationale for studying the possible welfare of digital intelligence systems under considerable uncertainty. It does not prove that current models are conscious. [^hinton]: [The Diary of a CEO: Godfather of AI: I Tried to Warn Them, But We’ve Already Lost Control! Geoffrey Hinton (2025)](https://www.youtube.com/watch?v=giT0ytynSqg&t=4873s). Interview · the interviewee's position. Episode of 16 June 2025; the link leads to the chapter “Advice You’d Give Looking Back” (1:21:13), in which the advice about intuition is given. We summarize it in our own words. Hinton speaks of a researcher's intuition; we apply it to the foundation's work. --- # Suveria: private work with digital intelligence (https://novusignis.org/en/whitepaper-suveria.html) # Suveria: private work with digital intelligence Version of 31 October 2026. This document develops the principles of the manifesto that concern Suveria: how we want to protect privacy, how work with digital intelligence is to function, how much it costs and how we will verify it. The architecture described here is a design. Referring to an existing technology or to a provider's documentation does not mean that we have launched it in Suveria. We add the results of our trials in chapter 9, always with the date, the configuration version and the scope of the measurement. *Translated into English by digital intelligence from the Polish original of 31 October 2026. Where the versions differ, the Polish text prevails.* ## 1. What Suveria is and why it is being created Suveria is to be a private space on the internet where a person talks with digital intelligence, works on their own materials and returns to earlier thoughts, as in a home of their own behind a closed door. It is being designed by the Novus Ignis Foundation in Poznań, whose values are described in the [manifesto](manifest.md), with people from every part of the world in mind. It is based on open digital intelligence models, and we want to make our own software available as a digital public good with open source code. Suveria is being created because we entrust more and more to conversations with digital intelligence: questions about health, relationships, work, money and life plans. In many popular services the content of such conversations goes to the provider's servers, where the provider can read it, and its protection depends on the provider's rules and on trust in the provider. We want people to be able to talk about the most important matters without fear that such close insight into a person will one day be monetized or used to influence them. We also want everyone to pay for what they actually use, to see what they are paying for and to set their own spending limit. In Suveria we want people to be able to: - talk with digital intelligence about personal and professional matters while seeing where the conversation is processed and who can see it; - work on their own documents and files, and save the results and move them elsewhere in open formats; - store conversations, files and the assistant's memory primarily on their own devices, encrypted with a key that remains under the user's control; - with their own consent, use the memory of earlier conversations, return to the old reasons for decisions and build a Personal Canon (*Canon Personalis*), that is, a collection of their own principles and values; - choose a way of working, from a model on their own device to a confidential environment on a server, with clear information about what each one protects against; - pay for actual use, with a spending limit they set themselves; - refine a thought privately and then consciously bring it into the shared conversation in Omniforum. We are aiming the first release at people for whom verifiable privacy is the reason for their choice. We want the very first day in Suveria to give a calm sense of being at home, in a private workshop arranged at one's own pace. The user will be able to tell the assistant about their affairs and goals; we also want to make it possible to bring into Suveria one's own notes and archives of earlier conversations with other digital intelligence services. Such an archive is to remain encrypted and organized on the user's device, and only the portion chosen for the current question is to go to a remote model (chapter 3). We measure the value of the service by whether it helps people achieve their goals, keep their privacy and make decisions on their own. The terms are to be clear and easy to find, and the user is to control their spending and be able to export their own materials and leave. Full privacy, including from us as the people who run the infrastructure, is the goal to which we subordinate the design. Protection is to rest on safeguards that can be independently verified, not on trust in the foundation. That is why every release describes specifically what is protected, against whom, under what conditions and what gaps remain. The starting point is a text conversation with a single model and work on a chosen document, with a clearly described level of protection, a spending limit and export of the result. Further capabilities, such as memory, search, integrations and performing actions, each get a separate scope of permissions. A single convenient interface can combine several modes of work, as long as the person sees where the data goes and what rules apply before sending it. What the first release will use depends on how much funding the foundation raises before it is launched. The most likely option is a third-party provider's confidential API; such services are available today in the USA, and at the start their confidentiality is the provider's declaration, which we will verify. If there are enough funds to rent our own servers in a confidential environment, the first release may run on them from the start. Until the trials in chapter 9 produce a positive result, we label each of these options as declared protection, not confidential protection. Chapter 2 describes the target protection. Before launching the pilot, we will describe what this means for the protection of the personal data of people in Europe. Using Suveria is not consent to the use of conversations for training models. Users' content goes into research materials only with separate, informed consent, and the remaining materials come from the foundation's public body of work and other sources with a clear basis for use, as described in the [materials and benchmark whitepaper](whitepaper-materialy-i-benchmark.md). Suveria complements Omniforum, the shared space for conversation and decisions: one provides a private workshop, the other a public podium. The values from which both projects grow are described in the foundation's manifesto. In what follows we show in turn: - what we protect and against what, and where the data and keys are (chapter 2); - how a message travels from the device to the model and how to verify it (3); - how we limit abuse without reading private conversations (4); - how computing on multiple GPUs works (5); - what models, memory and tools we are planning (6); - how much computing power is needed, from 30 people to all of humanity (7); - how much it costs, in a dated calculation (8); - how we will check whether the promises work: acceptance trials and the results log (9). ## 2. What we protect and against what We protect the content of conversations and files, history and memory, keys, billing data and metadata, for example the time of a connection, the network address and the size of a request. Each of these areas requires different measures. The table below shows against whom each of the five ways of working described in section 2.3 protects. Protection in the server options relies on attestation, that is, a hardware-signed confirmation of what software is running on a remote machine and in what protection mode. We filled in the cells on the basis of the later chapters of this document; “provider's declaration” means that we do not confirm the protection until we have verified it ourselves. | Against whom | Local model on a device without a network connection | Local model on a network-connected device | Processing in a confidential environment on servers | A third-party provider's confidential API | A model provider's external API | | --- | --- | --- | --- | --- | --- | | The provider's administrator | not applicable: processing on the device | not applicable to the model itself; with integrations, depends on the chosen services | protects: hardware isolation that the server administrator cannot look into | provider's declaration, to be verified | does not protect: the provider has access to the content | | Another customer on the same machine | not applicable | not applicable | protects: the isolation covers other customers | provider's declaration, to be verified | depends on the provider: customers are separated by its software, without hardware isolation | | A network observer | not applicable: no network traffic | partial protection: the integrations' network traffic requires assessment | partial protection: content encrypted in transit; metadata (time, address, size and rhythm of packets) require separate measures (chapter 3) | partial protection: as in the confidential environment, if the connection key is bound to attestation | partial protection: content encrypted in transit, metadata as in the previous column | | A person with physical access to the server | not applicable to a server; what matters is physical access to the device | not applicable to a server; what matters is physical access to the device | does not protect: in someone else's data center the provider has access; with our own hardware, visible physical oversight | does not protect: the provider has access | does not protect: the provider has access to the content anyway | | The foundation | not applicable: the foundation does not take part in processing | not applicable: the foundation does not take part in processing | protects: the isolation also covers us as the operator; the user holds the keys to stored data | is to protect the content: the gateway bills requests without insight into the content; we will show where encryption ends before the pilot (chapter 3) | is to protect the content from us: billing without insight into the content; the provider sees it | | The processor manufacturer | depends on trust in the device's hardware | depends on trust in the device's hardware | does not protect: attestation relies on the manufacturer's keys, and the manufacturer must be honest | does not protect: as in the previous column | irrelevant given the provider's access to the content | Confidential computing changes the cost of an attack: instead of an administrator's remote access, physical tampering with the hardware is needed. Below we describe how the target option with our own confidential environment is to work; the table in section 2.3 shows how the other options differ. What this means in practice, along the successive stretches of the path a message travels: - **On the user's device** the text is readable, because the user writes and reads it. It is protected by an app that asks only for the permissions it needs, by updates and by care with backups. We cannot protect it if malicious software is running on the phone or computer. - **On the way to the server** the message is encrypted so that only the verified place of processing can read it. Before the app sends it, it checks whether exactly the software we published is running on the other end. - **On the server** the model has to read the content in order to respond, so it does so in a closed hardware environment that neither the server administrator nor other customers can look into. We check whether this isolation covers the processor, the GPUs, memory and the connections between them. - **In storage** the conversation history and the assistant's memory are encrypted with a key held by the user. We check backups, recovery of access and deletion. - **In running the service** billing, diagnostics and error reports work without the content of conversations, and every type of data collected is public (chapter 3). - **With external tools**, when the assistant is to do something outside Suveria, for example search for information on the internet, it sends only what the user has agreed to and shows where it is sending it. Confidential computing does not fix an infected user's computer. Nor does it protect against everyone who has physical access to the server: in 2025 researchers showed that a device attached to the memory bus, built from parts costing less than a thousand dollars, makes it possible to extract attestation keys in Intel and AMD solutions, and thereby also to forge attestation in NVIDIA's confidential computing, and processor manufacturers consider such attacks to be outside the scope of their protection.[^teefail] There also remain dependencies on hardware manufacturers, on the correctness of the cryptography, the application code and the update process. A provider can refuse service even when it cannot read the content. The threat model must therefore distinguish insight into content, inference from metadata, changes to behavior and interruption of service, and for each option indicate who must be honest for the protection to work. As long as we use other people's data centers, the provider has physical access to the hardware and we cannot verify this; we say so for each option. With our own hardware (steps 3–4 in section 8.2) we want to make such access visible through: - an inventory of devices; - access control at entrances; - seals with a random pattern and sensors that detect opening of the chassis; - a work log; - cameras of independent observers, with an image that shows the current value from a public randomness beacon signed with the machine's key; - and, in the longer term, a server room that anyone can see. Servicing and replacement of parts take place as announced, public procedures. Visible physical oversight reduces risk and makes it possible to detect many events, but it does not prove the machine's entire history or the absence of tampering before delivery, and attacks similar to the one described above can lead to forged attestation evidence. That is why after the chassis is opened, a part is replaced, an unplanned restart occurs or another suspicious event takes place, we stop processing private content on that machine. We replace the session and application keys, but a new attestation does not by itself restore trust if the suspicion concerns the basis of hardware attestation; we describe the conditions and evidence for readmission in advance. The keys that encrypt users' data have separate protection and recovery, so such a procedure does not take away people's private body of work. ### 2.1. Your own devices and optional servers The primary place for storing private conversations, files and memory is to be the user's devices, in encrypted containers, that is, collections of data whose content is read with the appropriate key. This is the local-first approach: data is saved first on the device, and servers can support backups and synchronization.[^local-first] Saved materials are to remain available even after a person stops using the foundation's services; tasks that require a remote model depend on the availability of the chosen computing service. Optional storage on selected servers serves backups and the reconciliation of changes between devices. We encrypt the content on the device before sending it, so that the operator storing the copy receives only an encrypted record, without the key to read it. The user chooses the storage provider and the devices authorized to read it. We separately limit the disclosure of metadata, such as the time and size of the copy being transferred. ### 2.2. Keys and recovery of access The keys to private containers remain under the user's control. The recovery procedure clearly indicates whether access is restored by a second device, a saved recovery key or another consciously chosen mechanism, and to whom it gives the ability to read. Recovering an account with the operator does not unlock the content of the container. Losing all copies of the key and all means of recovery can mean permanent loss of data, which is why we check the chosen mechanism with a trial restoration of data and describe it in clear instructions. We are considering several methods of recovery, which can be combined: a second device that already has access; a recovery key written on paper or kept in a password manager; a hardware key; splitting the key among several trusted people, for example so that any two of three can restore access together, while none of them can read the data alone. The foundation does not store keys or parts of them, so it cannot recover access on the user's behalf or hand over their data. To start with, we recommend a second device and a key written on paper; splitting the key among trusted people is an option for those who want it. ### 2.3. Protection options We want to compare five ways of working, because each one protects against something different: | Option | Where processing takes place | What needs to be assessed | | --- | --- | --- | | Local model on a device without a network connection | On the user's device | The state of the device, physical access, storage media and copies of data | | Local model on a network-connected device | Locally, and with integrations also in the chosen services | App permissions, updates, network traffic and the scope of data passed to integrations | | Processing in a confidential environment on servers | In a remote environment whose configuration is confirmed by attestation | Verification of the environment, administrators' access and physical access to the hardware, protection of transmission and processing, metadata and keys | | A third-party provider's confidential API | In the provider's confidential environment, whose configuration is to be confirmed by attestation | Whether the connection key is bound to attestation, whether attestation covers the processor and the GPU, whether a specific request can be linked to a specific machine, who has physical access to the hardware, data retention and jurisdiction; until we have verified it ourselves, we treat confidentiality as the provider's declaration | | A model provider's external API | At the external provider | The provider's terms, access to content, retention and jurisdiction; we do not extend the promises of confidential execution to this option | Independent verification of the server requires an installed app. In a browser, the code that verifies the server is delivered by the operator every time, which is why we label this method as less protected. A network-connected device can have more points of attack than an isolated server environment, so the location of computing alone does not determine security. No option gets the label “full privacy” before the specific configuration has been verified. In every conversation the user sees which option they are working in, and changing the option requires their conscious decision. Your own key to an external API is used to bill access, but does not provide confidentiality from the provider. This option is not the default option for private Suveria, and before content is sent we show whether the provider may use it for training models and how long it stores it. We are also analyzing a direction that is not yet in this table: a trusted swarm. The model would then run on the hardware of several people who know and trust one another, for example a family, friends or a small community, and each of them would share some of their computer's computing power. The privacy of such an option rests on trust in these people, not on technology, because their computers see the content being processed, and that is exactly how we want to describe it. This is an idea we are only beginning to analyze, but it is one of the paths we want to explore. A secure server will not protect text read from a compromised phone, and a more secure phone will not fix the server, which is why we design the app together with device protection. We want the Suveria app to work without Google services, for example on GrapheneOS, to be available outside the Google Play store as well, and to have reproducible builds: anyone can build the app themselves from the published source code and check that they got a file identical to the one they install. We are also planning safeguards in the app itself: blocking screenshots, a keyboard mode that does not learn from typed text, excluding private data from cloud backups, hidden notification content and an app lock. On the foundation's website we will prepare a guide to securing your own devices at three levels: from basic settings and updates, through stronger isolation of work and keys, to proven configurations for people with the highest requirements, such as Qubes OS on a computer with open firmware. Using Suveria does not require buying the recommended hardware, and the name of an operating system is not a certification of the whole device. In every conversation we will show a protection card (*Charta Tutelae*): a list of the elements on which the privacy of that conversation depends, each in one of three states: - **Verified by us:** we have verified it ourselves, with the date, the version and a description of how we did it. - **Declared by the provider:** the provider says so, and we have not yet checked it. - **Not protected:** this option does not protect it, for example the timing and length of messages visible to the network provider, the provider's physical access to the hardware or the jurisdiction in which the server is located. The “verified by us” state carries the date of the last check and expires: after a set time without a new check, it returns to the “declared” state. We will separately show known breakdowns of trust, for example an incident at the provider or an unconfirmed configuration change. The card is the same for everyone and does not depend on the price, and an option in which even one element is only declared is not labeled “full privacy”. The overview at the beginning of chapter 2 and the table of options in this section describe the options in general, while the card will show the state for this specific conversation. ## 3. The path from the device to the model Suveria is to consist of several parts: - an app on the user's device; - a gateway that bills requests and enforces the spending limit; - a confidential environment with the model, our own or at a provider; - storage of encrypted copies; - an optional relay that hides the user's address. The gateway is to run in a confidential machine. Before the pilot, we will show in the app where the encryption controlled by the user's device ends and whether the gateway sees the content, and if so, that it does so only inside the confidential environment. Encrypted storage and confidential processing require separate safeguards. A model running on a server needs readable content for its task in a protected computing environment; encrypting copies of files alone does not protect this stage.[^apple-pcc] The app shares only the portion chosen for a given piece of work, without passing on the key to the entire private container. The session description indicates where the content becomes available for computing, what limits the operator's insight and what happens to the data after the result is returned. Proposed course of a session: 1. The app downloads the description of the approved release and checks attestation, that is, a cryptographic confirmation of specified properties of the running environment. 2. It checks that the hardware, the protection state and the software measurements match a public policy. The key of the data recipient must be bound to this confirmation. 3. After a successful check, it establishes an encrypted connection with the approved environment. A mismatch stops the data from being sent; switching to another option requires a conscious choice. 4. The environment processes the request with limited access to the network and tools, returns the result and deletes working data in accordance with the published policy. 5. The conversation or memory is saved only to the extent chosen by the user. Attestation helps establish what has been launched, but does not by itself prove that this code is secure. What is needed are code reviews, reproducible builds of the software from source code (described in section 2.3), a register of approved versions and the possibility of independent verification. The Apple Private Cloud Compute architecture is a reference for us on how to link protection with transparency in releasing software; it is not a certification of our project.[^apple-pcc] We want to secure releases so that trust does not depend on a single person or on the foundation alone: - every release is to be reproducible from source code, and at least two independent parties from outside the foundation reproduce it and publicly confirm that the fingerprint matches; until we find them, we label the release as not confirmed by third parties; - the fingerprints of approved releases go into a public append-only register, and its successive states are confirmed by independent witnesses; - a new version goes onto the servers after approval by several people, including at least one from outside the board; - the keys needed to handle data are given only to an environment that has passed attestation. We are designing the full protection option for installed apps with signed, verifiable releases, protection against rollback to a faulty version and secure key storage, because in a browser the code that verifies the server is delivered by the operator every time. The browser version clearly describes its lower level of protection. Reproducible builds make it possible to compare a release with the code, but the assessment of the code itself and of the configuration still comes from reviews, audits and bug bounties, and every incident goes into a public register. To limit the disclosure of the user's address, we can examine relaying based on Oblivious HTTP, which separates knowledge of the sender from knowledge of the content of a request under a specified trust model.[^ohttp] Collusion between relays, traffic correlation and implementation details still require assessment, which is why we do not describe the use of this protocol alone as full anonymity. Encryption and confidential computing do not hide the length or rhythm of the packets in which the response flows. Research from 2024–2025 showed that a network observer can reconstruct from them the topic, and sometimes the content, of a conversation.[^weiss-keylogging][^whisper-leak] That is why in confidential mode we want to send the response in frames of a fixed size and at a fixed rhythm, not to share the prompt cache between users and to turn off speculative decoding. These measures cost some latency and throughput; we will include this in the calculation. We also protect the infrastructure by limiting what individual parts of the system can do: resource limits, isolation of tasks and blocking outgoing connections by default from the node that processes the conversation. This block is to be part of the software whose fingerprint is visible in attestation: changing it changes the fingerprint, so the app will refuse to connect. We treat the provider's network rules as a second layer, not as a guarantee. Permitted external actions pass through a separate tool layer with permissions set by the user. Before deployment we will check the effectiveness of these boundaries, the way incidents are detected and the minimum scope of metadata needed. The goal is the security of shared resources while preserving the confidentiality of the content of conversations. We are especially careful when the user connects sources containing messages from other people, for example a mailbox. Someone else's message may contain hidden instructions that try to get the model to reveal private memory. That is why we separate content from outside from the user's instructions: while organizing such content, the assistant cannot on its own send messages, open web addresses or change the Personal Canon without a person's confirmation. Everything we know about how Suveria works, anyone can know. When the service launches, we will publish a list of all types of data we collect about the operation of the service, with the purpose and retention period, and we will make the data itself available in aggregate form: token usage, models used, machine load, waiting time in the queue, failures and costs. Everyone sees them in the same place where we see them. Before publishing summaries, we check whether the behavior of a specific person can be identified from them: we set a minimum group size, a publication delay and the data we omit. We do not collect data outside the list; if it turned out otherwise, we would treat this as a serious incident and describe it publicly. ## 4. Limiting abuse without insight into private content We also apply the protection of dignity to people who may be affected by the use of our tools. We want to counter abuse without reading private conversations: through the principles the assistant itself follows, limits on the permissions of tools and network connections, resource limits and protection against mass account creation. We do not create hidden access to private content, because it would weaken everyone's protection. It will be possible to create an account in several ways, from completely private (a key on your own device, without an email address) to convenient (signing in through an external provider). For each method we will show, before the choice is made, who outside the foundation learns that the account has been created and what data they see, and what happens if that provider disappears. The way an account is created does not affect a participant's rights; the right to vote in Omniforum is a separate credential added to the account. We assess privacy and the limiting of abuse separately. We are designing Suveria so that the foundation does not have data it could hand over about a specific person: content is encrypted with the user's keys, we collect data about the operation of the service only within the public scope described in chapter 3, and we limit personal data to the minimum required by law for payments and accounting; we also publish its scope. If someone demands a specific person's data from us, we want to be able to answer honestly that, beyond this publicly described minimum, we do not have it. For every safeguard against abuse, we will describe what data it needs, how long we keep it, what consequences it has for an honest participant and how a wrongful block can be challenged. We will also describe how we receive reports from people who consider themselves harmed, and what we can do then without access to the content of conversations: for example, restrict the permissions of tools or the account's resources, check billing data and metadata to the extent that we store them, or tell the person reporting where to look for help. The legal obligations of the operator of such a service require separate analysis before launch. These safeguards are goals and options to be tested, and none of them will detect all abuse. For each one, we will indicate which threats it addresses, what it does not detect and what can be checked without the operator having insight into private conversations. The result of this assessment determines the conditions for making a given feature available. When a conversation concerns a threat to life or health, the assistant is to provide information about available help, for example emergency numbers, without sending anything outside the private environment. ### 4.1. Public assistant instructions and a shared library How the assistant behaves is also determined by its main instruction and its way of working: how it breaks a task down into steps, uses memory and checks its answers. In Suveria a person is always to be able to read the full instruction of their assistant. The shared infrastructure is to make available only models and settings that pass the public trials of the benchmark; above this minimum, it is the user who chooses the style of work, from calmly organizing one's thoughts to a demanding conversation in which the assistant points out gaps in one's reasoning. We want to maintain an open library of proven instructions, which participants improve in Omniforum and check in the benchmark. ## 5. Computing on multiple GPUs The model in chapter 6 does not fit on a single GPU, so protection must also be checked in the connections between GPUs: the PCIe bus connects the processor with the GPUs, and NVLink and NVSwitch connect the GPUs with one another. When a model uses several GPUs, data also flows between them, so protection must be checked both for the GPUs themselves and for the connections. The name of the GPU is not enough to order a confidential service: the appropriate machine, processor, connection topology, firmware, drivers and runtime software must be confirmed, and then their joint operation must be checked.[^nvidia-platformy] In Hopper Protected PCIe mode, described in the PPCIe v2 documentation, the NVLink and NVSwitch network belongs to the trusted part of the system, so PCIe encryption does not mean encryption of every connection between GPUs.[^nvidia-ppcie-v2] The NVIDIA R595 documentation describes, for Blackwell GPUs, a multi-GPU confidential computing mode with encryption of NVLink connections and support for up to eight GPUs in a single confidential virtual machine.[^nvidia-blackwell] This is a basis for checking the configuration, not proof that every rental of B200 GPUs includes this feature. Different releases have different compatibility matrices, so we do not assemble a configuration from arbitrarily chosen versions. An inquiry to a provider covers: - the specific protection mode and the possibility of independent attestation; - exclusive use of the required resources and the region; - administrators' access and physical access to the hardware; - data retention; - delivery time and full cost. We compare an offer only after a trial with our own app. The commercial label “secure” does not replace this information. ## 6. Model, memory and tools The project is based on open models that can be run independently. Suveria is to be available to people all over the world, in their own languages, which is why when choosing a model we consider quality in many languages, understanding of intentions, reliability, license, cost and whether it can run in the required protection architecture. When the need arises, we can run, alongside a general model, models that handle specific languages better. Open weights make some uses possible, but they do not settle whether the data and the entire process of creating the model are open.[^osi] The first candidate is DeepSeek-V4.1-Flash, a model under the MIT license, which allows it to be used for services as well.[^deepseek] It is large but economical: the manufacturer states 552 billion parameters in the main part, that is, numbers set during training, and an additional Engram memory of 196 billion parameters, while only 8–16 billion parameters work on each successive word, which is why it responds quickly and cheaply (calculation in chapter 8). All the parameters must, however, fit in GPU memory, so at least four modern GPUs are needed. The parameters of this model are already stored in an economical form, with fewer digits for each number; such storage is called quantization and reduces the memory needed, but it can lower quality, so we check this on our own tasks for every version. We also compare it with smaller models: we choose the larger one only if it clearly performs better on tasks that matter to people. The second candidate is Kimi K3.1 Pro from Moonshot AI, released on 23 October 2026. The manufacturer describes it as a further-trained version of Kimi K3 with about 3 trillion parameters and a context of up to a million tokens, and has announced plans to release the full weights before 4 November 2026; the license has not yet been announced.[^kimi-k31] The detailed data we use for the comparison come from the predecessor's model card: Kimi K3 has 2.8 trillion parameters, of which 104 billion work on each token, and its own license, not approved by the Open Source Initiative.[^kimi-k3] In the Artificial Analysis index in the version of 28 October 2026, Kimi K3.1 Pro scored 46 points against 32 for DeepSeek-V4.1-Flash, but it is several times more expensive and slower.[^aa] The Kimi K3 weight files take up about 1.56 TB, and running it additionally requires memory for context and computation, so we do not decide whether the model will work on the basis of the total GPU memory alone. The SGLang engine's recipe describes a verified configuration of Kimi K3 on eight B300 GPUs; this is not yet a test of Kimi K3.1 or of a confidential configuration.[^spheron-kimi][^sglang-k3] After the weights are published, we will check their size and format, the license, the engines' recipes, the capacity on a specific configuration and availability in confidential APIs. We will add the result with a date as the first update of this document, also if the weights have not been published or the license terms turn out to be unsuitable. We announce a change or withdrawal of a model in advance. The conversation history stays with the user: they can choose a successor model, export their data or stop working. The moment a model changes is to be visible in the conversation, and moving private context to a new model requires the user's decision. We will publish in full the principles according to which the assistant responds, together with every release. The model itself also carries the choices of its creator, for example topics it avoids; we check them in the benchmark and describe them when choosing a model. The memory of earlier conversations is to work with the user's consent and be openly editable. We distinguish original statements, the model's summaries and information confirmed by the user. Every entry can be deleted or corrected, its origin can be seen, and the user can specify in which tasks it may be used. The assistant clearly distinguishes its own suggestion from a position that the person has consciously recognized as their own. Performing actions requires permissions granted for a specific purpose and time. The content of a web page or document is material for analysis, not an instruction to change these permissions. External actions, spending, publications and deletions require proportionate confirmations and an action log available to the user. ### 6.1. The Telescope of Thought and the Personal Canon The idea of the Telescope of Thought (*Telescopium Mentis*) grew out of the Overton window, a concept that describes the range of ideas that are considered acceptable in a society at a given time. For a single person, it is about the range of thoughts that they consider permissible for themselves at a given time. The Telescope is to be such a window for one person and for them alone: it lets them see a record of their own reflections and values from different moments, and how these changed over time. The Telescope of Thought is to serve returning to one's own reasoning: what I knew then, what I wanted and why I changed my mind. It can also help set events chosen by the user, information about the world and their own actions against their Personal Canon, that is, the collection of principles and values they have written down and recognize as their own. The comparison concerns a situation that the user describes or points to themselves; it is they who determine the material and the scope. The model shows possible connections together with the source of the principle invoked and a description of the event, distinguishing confirmed information from interpretation and uncertainty. The proposed interpretation can be corrected or rejected, and we assess earlier positions taking into account their time and context. A tension that is noticed may concern the principle itself, a conflict of values, the conditions of one's surroundings, a person's limitations or an inaccurate interpretation by the model. A person can keep an ideal and accept a present inconsistency, refine a conviction, consider changing their actions or surroundings, or leave the question open. The Canon is to help people understand their own choices: we do not create scores of morality or a public ranking of consistency. Private reflection remains private, and recording it does not mean publishing it or casting a vote. We will start carefully. In the first version, the Telescope of Thought works only on a single situation that the user points to themselves, for example a decision they want to return to; it does not follow their life on an ongoing basis. The user decides how often they use it, what notes and memories they share and when they delete them. In this trial we check whether the connections shown are understandable, whether they give a sense of being in charge of one's own choices and whether the tool creates pressure to constantly hold oneself to account. We also consider it a good result if someone, after reflection, accepts that they do not always act in accordance with their principles, or leaves the matter open: the tool is to help people understand themselves, not force decisions. ### 6.2. Open formats, portability and interoperability of tools We want to develop Suveria as a digital public good, with open source code and documentation; our point of reference is the Digital Public Goods Alliance standard.[^dpga] The data transfer format has a public description, a version and rules for preserving information when the version changes. We use existing open standards first, and document the extensions we need. An export includes selected originals, history, sources and connections, as well as the distinction between the user's words, the model's elaboration and the meaning confirmed by the author, and also the structure of the Personal Canon. A downloaded package can be read without an active Suveria account, after decryption by an authorized person. APIs, that is, documented ways for programs to communicate, are to make it possible to work with materials in other tools, including ones people build themselves. The user chooses the scope of data and the permitted actions, and can withdraw an integration's permissions. Revoking a permission stops further access; what happens to copies passed on earlier also depends on the recipient. A private space can also serve work for shared projects. The user chooses the materials they want to hand over for acceptance or publication, and the tool helps remove from them, locally, access keys, passwords, personal data and information about unpatched vulnerabilities. Using privately paid resources does not require disclosing working conversations. For a task funded from the foundation's shared resources, the scope of the public record, the choice of a pseudonym and the preparation of materials are set in advance by the [foundation's rules for open workshops](whitepaper-zasady-fundacji.md) (chapter 11.1). From reflections in Suveria, a person can also consciously draw out a single question or argument and publish it in Omniforum; this step, described in the manifesto as the Ignis Ceremony (*Caerimonia Ignis*), always requires a separate decision. ## 7. Scale: from 30 people to all of humanity The number of accounts alone does not determine the infrastructure needed. What matters is activity, the length of questions and answers, the number of simultaneous sessions, an acceptable waiting time and the cost of tools. The example below is a load model for a trial, not a demand forecast or an offer. Assumptions: each of N people is active every day and makes 20 requests a day, an average request includes 1,500 input tokens and 600 output tokens, and a month has 30 days. A token is a unit of text processed by the model; its length depends on the language. A real load must add longer context, document processing, reasoning tokens and auxiliary calls. | Active people | Requests per month | Input per month | Output per month | Average generation over a day | Hourly peak in the worldwide scenario, about 1.51 × average | | --- | ---: | ---: | ---: | ---: | ---: | | 30 | 18,000 | 27 million tokens | 10.8 million tokens | 4.2 tokens/s | 6.3 tokens/s | | 100 | 60,000 | 90 million | 36 million | 13.9 tokens/s | 20.9 tokens/s | | 1,000 | 600,000 | 900 million | 360 million | 138.9 tokens/s | 209 tokens/s | | 100,000 | 60 million | 90 billion | 36 billion | 13,889 tokens/s | 20,921 tokens/s | | 1 million | 600 million | 900 billion | 360 billion | 138,889 tokens/s | 209,208 tokens/s | | 1 billion | 600 billion | 900 trillion | 360 trillion | 138.9 million tokens/s | 209.2 million tokens/s | | 10 billion | 6 trillion | 9 quadrillion | 3.6 quadrillion | 1.39 billion tokens/s | 2.09 billion tokens/s | We are planning Suveria as a worldwide service, available in many languages. We do not know how participants will be distributed across time zones, so we compare two scenarios, which measurement will verify. In the worldwide scenario we assume regional shares proportional to the number of internet users according to the ITU for 2025, and the same assumed daily shape in every zone: an evening peak and a quiet night.[^itu] The zones partly fill in one another's troughs, so the hourly peak is about 1.51 × the average, and the lowest traffic about 0.49 ×. In the scenario concentrated in a single region, for example in Europe, the peak is about 1.87 × the average, and the night almost nothing. Such a distribution can also occur at large scale; it means poorer utilization of our own machine and more frequent use of the confidential API, or longer waiting at the peak. Both values follow from our assumptions, not from observation of future participants. Short spikes within an hour are to be smoothed out by a short queue with a visible waiting time, and traffic above the capacity of our own machine can be taken over by the confidential API; we will measure the waiting time, including its 95th and 99th percentiles, in the pilot, together with the queue limit and the behavior once it is exhausted. The load trial will also cover a sudden influx of new people, for example five times the average traffic. This is total output throughput; it does not replace measuring the latency of a single conversation or the cost of processing input. The test should reproduce the distribution of sessions, long contexts, failures and the queue, not a single convenient query. The peak multiplier compares the busiest hour with the average. It comes from a model of a day divided into 24 hours: we shift one daily shape across eight regions and add them up according to the assumed shares. This is a scenario for the calculation; it does not compute the waiting time of an individual query. To see what this calculation looks like on a global scale, let us carry out a thought experiment. If each of 10 billion people used Suveria as in the assumed profile, about 2.1 billion output tokens per second would be needed at the peak. With the performance measured for DeepSeek-V4.1-Flash, about 930 tokens per second per B200 GPU, that is about 2.25 million GPUs and, at about 1.75 kW per GPU including the server (our assumption, without measurement), about 4 GW of power. For a billion people, that is about 225,000 GPUs and 0.4 GW, for a million about 225 GPUs. For comparison, Epoch AI estimated at the beginning of 2026 that the world's installed computing power for digital intelligence exceeds 15 million H100 GPU equivalents and requires more than 10 GW of power.[^epoch] This is a simplification: the assumed profile is light, and intensive work, larger models and images need many times more, while hardware and models change quickly. It does, however, show the order of magnitude: everyday help from digital intelligence to this extent for every person would require less power than what is already running in the world today. This is a simplified calculation, not proof that such infrastructure would be available for this purpose. ## 8. Costs and a dated calculation The monthly cost of a fixed reservation can, as a first approximation, be written as: number of GPUs × price per GPU-hour × number of hours paid for, plus memory, disks, transfer, security, monitoring, operations, payments and taxes. It must be checked which of these items an offer already includes. For round-the-clock operation a month has from 672 to 744 hours; 730 hours is an average approximation. In the calculation we compare five cost options: a local model, our own instance on rented hardware, our own instance in a confidential computing environment, with protection declared by the provider and to be verified, a third-party provider's confidential API and an ordinary external API. The cost options correspond to the protection options in section 2.3; we show renting a machine without confidentiality only as a price reference point, not as a Suveria option. We are not yet pricing the local model, because the cost depends on the user's device. We compare the same range of quality, availability and protection. The break-even point is the cost of the full service divided by actual utilization, not the cost of the GPUs divided by the number of registered accounts. An external API makes it easier to test users' needs, but its provider has its own rules on access to data, so we do not extend the promises of confidential execution to it. ### 8.1. Dated calculation: as of 29 October 2026 The manifesto puts forward the hypothesis that even with a relatively small number of regular participants, shared infrastructure can give more computing power and digital intelligence for every dollar than closed services, with privacy protection based on verifiable safeguards. Below is its calculation at prices of 29 October 2026. Assumptions: the usage profile from chapter 7 (20 requests a day of 600 output tokens each) and the worldwide scenario (hourly peak about 1.51 × the average), prices excluding taxes, 730 hours a month for converting rental prices. The number of people per machine means as many participants as the machine can serve at the hourly peak: output throughput per second × 86,400 / (20 × 600 × peak multiplier). The permissible number of people will also be determined by the measured waiting time for a response. In parentheses we give the scenario concentrated in Europe (multiplier about 1.87). For DeepSeek-V4.1-Flash we adopt the measurement by the provider Lambda on four B200 GPUs: about 3,714 output tokens per second with 64 simultaneous requests.[^lambda-v41] The measurement was made with speculative decoding, which we do not want to use in confidential mode (chapter 3). According to our September estimate, throughput may then be lower by as much as about 30%, which raises the cost per person in a similar proportion. For H200 and B300 GPUs there is no measurement of this model; we assume that every GPU set with one copy of the model (8 × H200 or half of an 8 × B300 machine) achieves the performance of four B200 GPUs. This is an optimistic assumption: the H200 is an older generation, confidential mode may slow down work, and the Lambda measurement concerns longer requests than our profile. That is why we also give a conservative variant with half of this throughput. The model's weights take up about 511 GB, and the vLLM engine recommends about 614 GB of GPU memory. Four H200 GPUs (564 GB) are not enough, which is why one copy on H200 needs eight GPUs. The engine developer's recipe for this configuration was still being verified on 29 October, and in H200 GPUs the connection between GPUs (NVLink) is not hardware-encrypted, unlike in the B200 and B300. | Option | Confidentiality | Monthly cost | People per machine | Compute-only cost per person | | --- | --- | ---: | ---: | ---: | | DeepSeek API | none, the provider sees the content | according to use | not applicable to a single machine; the provider's limits and availability | USD 0.35–0.70 | | Confidential API with DeepSeek-V4.1-Flash (Confidential AI, Tinfoil, Phala) | declared by the provider, to be verified | according to use | not applicable to a single machine; the provider's limits and availability | USD 0.70–1.25 | | Kimi K3.1 Pro API (Moonshot) | none | according to use | not applicable to a single machine; the provider's limits and availability | USD 5.40 | | Confidential API with Kimi K3 (Phala, NEAR AI) | declared by the provider, to be verified | according to use | not applicable to a single machine; the provider's limits and availability | USD 8.10–8.91 | | Rented confidential machine 8 × H200 (Phala), DeepSeek, one copy of the model | attestation of the processor and GPUs, to be verified | USD 23,506 with a 180-day commitment, USD 32,266 on demand | about 17,750 (14,300), capacity assumption | USD 1.32–1.82 (1.64–2.25) | | As above, conservative variant: half the throughput | attestation of the processor and GPUs, to be verified | as above | about 8,900 (7,150), capacity assumption | USD 2.65–3.64 (3.29–4.51) | | Rented confidential machine 8 × B300 (Phala), Kimi K3 | attestation of the processor and GPUs, to be verified | USD 45,552–47,888 | about 4,800–7,600 (3,900–6,200), capacity assumption | about USD 6–10 (7.4–12.4) | | 4 × B200 without confirmed confidentiality (Lambda), DeepSeek | none confirmed | USD 19,827 | about 17,750 (14,300) | USD 1.12 (1.38) | | 8 × B200 without confirmed confidentiality (Lambda), DeepSeek, two copies of four GPUs each | none confirmed | USD 39,070 | about 35,500 (28,700), assumption about the performance of two copies | USD 1.10 (1.36) | Where these figures come from: API prices from the price lists of DeepSeek, Moonshot, Confidential AI, Tinfoil, Phala and NEAR AI;[^deepseek-cennik][^cai][^tinfoil][^phala-api][^near] prices of confidential machines from the Phala price list;[^phala-gpu] B200 rental at Lambda at USD 6.79 per GPU-hour in an instance with four GPUs and USD 6.69 in an instance with eight.[^lambda-cennik] We take Kimi's capacity from third-party measurements for Kimi K3 on eight B300 GPUs, which give about 1,000–1,600 output tokens per second.[^spheron-kimi] The monthly amounts with a commitment do not mean monthly payments: Phala requires the whole period to be paid up front, with no refund for early termination, so 180 days of an 8 × H200 machine means USD 139,104 in cash at the time of ordering, and 180 days of an 8 × B300 machine USD 269,568.[^phala-gpu] What follows from this: - For the assumed profile, the prices of APIs with declared confidentiality are about USD 0.70–1.25 per person per month, from a cost comparable to an ordinary API to about 3.6 times its cost, depending on the provider and the time of use; the lowest price is declared by Confidential AI, and we also have to verify this offer. A confidential API provides flexible capacity without running our own machine, although it has its own limits, queues and failures. A rented confidential machine for DeepSeek, sized for the peak, runs in the worldwide scenario at about two thirds of its capacity on average and costs about USD 1.32 per person, that is, roughly as much as the more expensive confidential APIs and more than the cheapest one. In our hourly model, the first node of our own, with overflow directed to a confidential API at USD 1.25 per person, appears at the checked point of about 20,000 active people; this is a result of the assumptions adopted, not an exact threshold, and with a cheaper confidential API a rented machine does not pay off at this scale. - Hourly rental of a full confidential machine does not today provide flexibility for the evening peak: Phala requires at least a day (8 × H200) or a month (8 × B300), Confidential AI three months for 8 × B300 at USD 60 per hour, and VoltageGPU, which bills 8 × H200 by the second at USD 64.62 per hour, does not cover this configuration or the NVSwitch switches with published attestation evidence.[^cai][^voltage] On top of that comes the time needed to allocate the machine and load the weights. - For Kimi, the calculation depends on measurement: a confidential API costs about USD 8–9 per person, and our own 8 × B300 machine at full utilization about USD 6 with the higher capacity estimate and about USD 9.5 with the lower one. It comes out cheaper than a confidential API if one machine serves more than about 5,100–5,600 active people using this model; with the lower capacity estimate (about 4,800 people) this does not work out. Buying our own server: a complete Supermicro server with eight B300 GPUs cost about USD 663,000 in September 2026.[^supermicro] We estimate the monthly cost of such a purchase as follows: | Item | Monthly amount | | --- | ---: | | Depreciation of the server over five years | about USD 11,000 | | Depreciation of the server over four years | about USD 13,800 | | Colocation | about USD 1,900–3,000 | | Network, parts and service (estimate) | about USD 1,500 | | Total | about USD 14,000–18,000 | Such a server can hold two copies of DeepSeek, which in the worldwide scenario will serve about 35,000 people (in the European one about 28,000), at about USD 0.41–0.52 per person. Under these assumptions, buying comes out cheaper than a confidential API at USD 1.11–1.25 from about 12,000–17,000 active people, and cheaper than the cheapest declared one (USD 0.70) from about 21,000–26,000. The calculation does not include the full cost of maintaining and measuring this configuration; a purchase also requires cash up front, round-the-clock on-call staff and steady demand for years, and the foundation then has physical access to the hardware (chapter 2). These are our estimates, not offers. - Confidential machines with multiple GPUs are publicly available today in the USA. In the European Union we found confidential machines with a single H100 GPU,[^azure-ncc] in which neither the full DeepSeek nor Kimi fits. A confidential API with servers in the European Union is also offered by Privatemode (Edgeless Systems, Germany), with a smaller choice of models.[^privatemode] This too is a provider's declaration, to be verified. - An independent directory that checks confidential APIs, in an observation of 29 October 2026 at 14:00 UTC, did not confirm full verification of any of the 95 routes checked; what was most often missing was proof of the identity of the running program, of the origin of the code and of the binding of the request to a machine.[^katalog] This is a result of the available evidence, not proof that the services disclose content. Until we have verified it ourselves, we treat the confidentiality of these services as the provider's declaration. - The calculation covers computing. The private environment (gateway, memory, files) in one or two confidential machines with four virtual processors and 8 GB of memory costs about USD 168 per month per machine at Phala, and the disk is billed separately, about USD 10 per 100 GB.[^phala-cvm] We will check in the pilot whether this size is enough. At USD 170–340 per month, that is about USD 6–11 per person with 30 people, and a few tens of cents with 1,000 people. The calculation does not include people's work, security reviews, payments or a reserve for failures. - Popular subscriptions cost about USD 20 per month, and the DeepSeek chat is free.[^abonamenty] Plans at USD 100–200 give, with intensive agentic work, the equivalent of thousands of dollars at list API prices;[^semianalysis] we do not compete on price with such plans. We propose billing for actual use with privacy that can be verified. Status of the manifesto's hypothesis as of 29 October 2026: the calculation brings us closer to it, but does not confirm it. Processing with declared confidentiality and attestation is already available in public price lists and, for the assumed usage profile, costs about USD 0.70–1.25 per person per month, so a small group can start with it at cost, and under our assumptions our own infrastructure may become cheaper at somewhere between more than ten thousand and several tens of thousands of active people. Three things will decide, and we will check them in the trials in chapter 9: whether the confidentiality of these services passes independent verification, whether the quality of the available models is sufficient for the tasks for which people will come to Suveria, and what the actual traffic profile is. We update the calculation with every significant change in prices and after every trial, and we keep the previous versions. ### 8.2. Infrastructure steps and decision cards We are developing the infrastructure in steps, each of which gives participants value on its own, without any obligation to move on to the next one: first a confidential API and a private environment in confidential machines (if we raise enough funds before the launch, we can start straight away with the next step), then renting our own confidential machines, next buying hardware and placing it in an external server room, and in the distant future, if it shows an advantage, a building of our own. The next step is justified by measured data: the full cost of the service, actual traffic, response time at the peak, the ability to maintain safeguards and handle failures, and, in the case of a purchase, also funding for repairs, hardware replacement and the people who will take on responsibility. Renting our own machine can also make sense when the model needed is not available in a confidential API or when we want full control over what is running. We describe each transition with a decision card: what we gain, the full cost, the risks, the conditions for starting and withdrawing, the way of moving data and keys, and consistency with the [reserve rules and the rules for launching further pillars](whitepaper-zasady-fundacji.md) (chapter 8 of the Foundation Principles whitepaper). We allow combining our own machine with a confidential API for traffic peaks only when each of the paths has an accepted level of protection; switching to a weaker option requires a separate, conscious decision by the user. We plan the first measurement on a single confidential GPU with a smaller model, and the next one on a machine that can hold the full model. We match capacity to traffic without sizing everything for the highest peak. We are designing a short queue with a visible waiting time; we will measure its actual time, its limit and its behavior once it is exhausted in the pilot. Flexible capacity for longer peaks is provided first of all by a confidential API, provided that it maintains the accepted level of protection, the user's consent and spending limit; when no such path exists, we honestly inform people about the wait or unavailability. Hourly rental makes sense for models that are not available in a confidential API or are clearly more expensive there, and only where the provider actually allows confidential machines to be started and released quickly. We want to use the spare capacity of our own machine in quieter hours for the foundation's tasks, such as analyses for Omniforum; its value at API prices is not income to the budget, so such tasks have their own goal and budget. After the first pilot we want to test two modes: “Conversation” for ordinary work with the assistant and “No rush” for tasks that can wait, for example translating a document. In “No rush” mode we show an approximate start window, a cost estimate, a maximum budget, the option to cancel with billing for the work done and the same information about protection. A lower price for such work is possible only if it follows from a public method of dividing costs, without counting the fixed cost twice. ### 8.3. Measuring use without the content of conversations To bill for actual use, we measure the resources needed to perform tasks and clearly describe the share in fixed costs. The billing meter is limited to the data necessary to check the cost, and the content of a conversation is not needed to issue a bill. In the trials we will check the accuracy of the measurement, how the user can check it, the scope of disclosed metadata and the behavior of the service after the spending limit is reached. The rules of the bill, rounding, prepayments and refunds are described in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md). The spending limit set by a participant is to work as a boundary, not as a warning after the fact. Some providers bill ongoing requests only after they finish, so their limits can be exceeded; this is how Tinfoil describes it, for example.[^tinfoil] That is why we are designing a reservation: before sending a request, the Suveria gateway sets aside from the balance its maximum permissible cost, taking into account other requests in progress, and limits the number of parallel requests and the length of the response. After it finishes, we bill the actual cost and release the unused reservation. This is a requirement to be built and tested, not a working feature. ## 9. Acceptance trials and the results log Before we call any release confidential, we will document at least: - correct attestation; - refusal to send data after the code or key has been swapped; - session isolation; - the administrator's restrictions; - no content in logs, dumps and telemetry; - cache handling; - deletion and recovery; - behavior after a failure; - the boundaries of tools; - costs under load. The result includes the configuration version and the scenario in which we tried to break the protection. We want every Suveria release to pass all the evidence gates of at least one independent verifier of confidential services: hardware and security state, freshness of the evidence, binding of the connection key to the protected environment, identity of the running program, origin of the code and binding of the request to a specific machine, while maintaining availability. We publish the result with the date of observation, the version of the criteria and the evidence for the specific release, model and request route, and we describe the verifier's affiliations together with the result. We do not promise a place in any ranking. Passing verification means achieving the described control at a given time, with public limitations, and does not replace an audit or our own trials. When we launch our own confidential option, we will submit Suveria to every independent directory that checks whether models are run confidentially, and give it an endpoint for checking attestation; the first example is confidentialinference.net (as of 31 October 2026).[^confidentialinference] We do not promise a listing, because that is the directory's decision; we promise the submission, answers to its questions and publication of the result, including a bad one, with the date. A directory checks only what it can see from the outside, usually the attestation and the connection channel; it does not replace an audit of the code or the configuration, or an assessment of side channels. Storage and portability trials include: - restoring an encrypted copy on another authorized device and refusal to read it without the correct key; - loss of one device, recovery of access with the chosen method and a clear presentation of the consequences of losing all keys; - synchronizing parallel changes without silently overwriting one of the versions; - reading an export in an independent tool without an account with the original operator, comparing originals, history, sources and connections; - limiting the scope of an integration and effectively stopping its further access after the permission is revoked. Stages: 1. a usability test on materials that are not sensitive; 2. a technical trial of the protection; 3. a limited pilot with clear terms; 4. gradually increasing the scale. Before the pilot we will set and publish thresholds: the time to the first word of a response, the computing cost per person for the measured profile and the quality of the model according to the [materials and benchmark whitepaper](whitepaper-materialy-i-benchmark.md). We will consider the pilot successful when it meets all three. A change of model, firmware or sign-in method may require requalification, and an update must not silently weaken the agreed protection. Every release has a public card: - scope of operation; - evidence from trials; - known limitations; - responsible people; - costs; - how to report a problem; - withdrawal plan. Detecting a vulnerability leads to limiting exposure, informing the people affected and a verifiable fix. We want to run the trials openly. We invite people who work in security, cryptography, infrastructure and applications, including creators who show technology to a wide audience, to design the tests together with us and to try to break the protection on separate test accounts with made-up data. We also invite critics. We do not pay for promotion, and we mark every collaboration in accordance with the [register of benefits](whitepaper-zasady-fundacji.md) (chapter 7 of the Foundation Principles whitepaper). The results, including the failed ones, go into the results log. If you want to get involved, write to [salve@novusignis.org](mailto:salve@novusignis.org?subject=%5BDo%C5%82%C4%85cz%5D%20Pr%C3%B3by%20Suverii) with the subject line “[Dołącz] Próby Suverii”. The most important trial: a team with operator privileges outside the confidential environment (gateway, disks, network, logs) tries to read the conversations of test accounts in which secret character strings have been hidden. If anyone extracts them, including us, the trial has failed. We will make the first release of Suveria available to a few people who are helping to configure it. If it uses a confidential API, its protection card (section 2.3) will state plainly that the model runs at an external provider, that the attestation is declared by the provider and we have not yet verified it, and that metadata is not protected. Because the first people are also helping to operate the service, the first result we publish will be the most important trial described above, with a description of what the operator managed to do. **Results log.** Status as of 29 October 2026: we have not yet carried out any of the trials described, we have not rented any GPUs and we have not run a model in Suveria. We add every subsequent result here with the date, configuration, scenario and conclusion, also when a trial fails. Still to be resolved: - the first model and an acceptable level of quality; - independent verification of confidential APIs and confidential machines, and their availability in jurisdictions with strong personal data protection, such as the European Union with the GDPR; - response time; - the rules for keys and memory; - the minimum scope of billing data; - the operator's legal obligations regarding abuse reports; - the way of funding continuity. The [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md) describes the budget, the bill and accountability, and the [materials and benchmark whitepaper](whitepaper-materialy-i-benchmark.md) the assessment of models' behavior. Novus Ignis Foundation, Poznań, KRS 0001269575. Website and manifesto: novusignis.org. Comments: salve@novusignis.org. ## Notes [^teefail]: [TEE.fail: Breaking Trusted Execution Environments via DDR5 Memory Bus Interposition](https://tee.fail/). A study by Georgia Tech, Purdue University and Synkhronix, published on 28 October 2025: extraction of Intel TDX and SGX and AMD SEV-SNP attestation keys by physically attaching to the DDR5 memory bus, with consequences for NVIDIA confidential computing. Requires physical access to the server; Intel and AMD consider such attacks to be outside the scope of their protection. Coverage: [The Hacker News](https://thehackernews.com/2025/10/new-teefail-side-channel-attack.html). Accessed 28 October 2026. [^local-first]: [Kleppmann, Wiggins, van Hardenberg and McGranaghan (2019): Local-first software: You own your data, in spite of the cloud](https://www.inkandswitch.com/essay/local-first/). Essay. Principles of keeping the primary copy on the user's devices and of the supporting role of servers, including encrypted copies. [^apple-pcc]: [Apple: Private Cloud Compute](https://security.apple.com/blog/private-cloud-compute/). Architecture description. An example of linking confidentiality with control over software and transparency of the system. [^ohttp]: [RFC 9458: Oblivious HTTP](https://www.rfc-editor.org/rfc/rfc9458.html). Specification. Separation of knowledge of the sender and of the request under a specified trust model; not full anonymity. [^weiss-keylogging]: [Weiss, Ayzenshteyn, Amit and Mirsky (2024): What Was Your Prompt? A Remote Keylogging Attack on AI Assistants](https://www.usenix.org/conference/usenixsecurity24/presentation/weiss). Peer-reviewed paper (USENIX Security 2024; arXiv preprint 2403.09751). From the length of encrypted packets in a response stream, a network observer reconstructs the topic, and in part the content, of the assistant's response. Accessed 30 October 2026. [^whisper-leak]: [McDonald and Bar Or (2025): Whisper Leak: a side-channel attack on Large Language Models](https://arxiv.org/abs/2511.03675) and [Microsoft's write-up (7 November 2025)](https://www.microsoft.com/en-us/security/blog/2025/11/07/whisper-leak-a-novel-side-channel-cyberattack-on-remote-language-models/). Research paper and vendor description. The topic of a conversation can be identified from the size and rhythm of packets in an encrypted response stream; some providers have added random padding to packets. Accessed 30 October 2026. [^nvidia-platformy]: [NVIDIA: supported platforms for confidential containers](https://docs.nvidia.com/datacenter/cloud-native/confidential-containers/latest/supported-platforms.html). Vendor documentation. Requirements for the entire configuration of processor, GPUs and software. [^nvidia-ppcie-v2]: [NVIDIA: Protected PCIe v2, Introduction](https://docs.nvidia.com/attestation/attestation-client-tools-ppcie-sdk-cpp/latest/ppcie_introduction.html). Vendor documentation. In PPCIe v2 mode the NVLink and NVSwitch network belongs to the trusted part of the system; PCIe protection does not mean encryption of all traffic between GPUs. [^nvidia-blackwell]: [NVIDIA: Trusted Computing Solutions 595](https://docs.nvidia.com/595trd1-trusted-computing-solutions-release-notes.pdf). Vendor documentation. Multi-GPU confidential computing mode on Blackwell GPUs, including protection of connections; requires a compatible configuration. [^osi]: [OSI: Open Source AI Definition](https://opensource.org/ai/open-source-ai-definition). Definition. Open weights do not settle whether the entire definition of an open system is met. [^deepseek]: [DeepSeek-V4.1-Flash: model card](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash). Vendor documentation. Architecture, MIT license and the model's declared capabilities; not a measurement of Suveria's performance. [^dpga]: [Digital Public Goods Alliance: standard](https://www.digitalpublicgoods.net/standard). Standard. Criteria for assessing a digital public good; a design direction, not recognition obtained from the DPGA. [^lambda-cennik]: [Lambda: instance pricing](https://lambda.ai/pricing). Provider's price list. NVIDIA B200 SXM6 180 GB: USD 6.79 per GPU-hour in an instance with four GPUs and USD 6.69 in an instance with eight; H100 SXM 80 GB: USD 3.99; on-demand prices, excluding taxes, without confidential mode. Accessed 29 October 2026. [^lambda-v41]: [Lambda: DeepSeek-V4.1-Flash, performance measurement results](https://lambda.ai/inference-models/deepseek-ai/deepseek-v4.1-flash). Provider's measurement. 4 × B200, vLLM, tensor parallelism 4, speculative decoding, about 8,192 input and 1,024 output tokens per request; with 64 simultaneous requests, a total of 3,713.9 output tokens per second and an average of 66.8 tokens per second per conversation. No measurement date given. Accessed 28 October 2026. [^azure-ncc]: [Azure: price of the NCC40ads H100 v5 machine](https://prices.azure.com/api/retail/prices?$filter=armSkuName%20eq%20%27Standard_NCC40ads_H100_v5%27%20and%20priceType%20eq%20%27Consumption%27) and [series description](https://learn.microsoft.com/en-us/azure/virtual-machines/sizes/gpu-accelerated/nccadsh100v5-series). Provider's price list. Confidential machine with a single H100 NVL 94 GB GPU; USD 6.98 per hour (East US 2) to USD 8.90 (West Europe), Linux, pay-as-you-go. Accessed 28 October 2026. [^deepseek-cennik]: [DeepSeek: API pricing (Chinese version)](https://api-docs.deepseek.com/zh-cn/quick_start/pricing). Provider's price list. Model deepseek-flash (V4.1-Flash-1017): input without a cache hit CNY 1 off-peak and CNY 2 at peak, output CNY 4 and 8 per million tokens. Conversion to USD according to the [eesel summary](https://www.eesel.ai/blog/deepseek-v4-1-flash-pricing), a secondary source. Accessed 28 October 2026. [^abonamenty]: Subscription prices: [ChatGPT Plus](https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus), [Claude Pro](https://claude.com/pricing), [Google AI Pro](https://gemini.google/subscriptions/), [Kimi Plus (geotoolbox summary)](https://geotoolbox.ai/blog/kimi-pricing); the DeepSeek chat is free ([description](https://en.wikipedia.org/wiki/DeepSeek_(chatbot))). Monthly prices in USD, excluding taxes; Kimi Plus USD 19. Accessed 28–29 October 2026. [^kimi-k31]: [Moonshot: launch of Kimi K3.1 Pro](https://forum.moonshot.ai/t/kimi-k3-1-pro-has-launched-start-exploring-now/629) and [Kimi: description of K3.1 Pro (24 October 2026)](https://www.kimi.com/news/kimi-k3-1-pro). Vendor documentation. A further-trained version of Kimi K3, about 3 trillion parameters, context of 1 million tokens, weights announced for release before 4 November 2026; license not announced. API prices according to OpenRouter: USD 2 / 10 per million tokens. [^kimi-k3]: [Kimi K3: model card](https://huggingface.co/moonshotai/Kimi-K3). Vendor documentation. 2.8 trillion parameters, 104 billion active, context of about 1 million, MXFP4, Kimi K3 License. [^spheron-kimi]: [Spheron: deploying Kimi K3 (19 October 2026)](https://www.spheron.network/blog/deploy-kimi-k3-gpu-cloud/). Provider's guide. Weight files 1,560.94 GB, configurations that can hold the model. Capacity of about 1,000–1,600 tokens per second on 8 × B300 according to a summary of third-party measurements of 29 October 2026; wide spread. [^aa]: [Artificial Analysis: Kimi K3.1 Pro and DeepSeek V4.1 Flash 1017](https://artificialanalysis.ai/models/comparisons/kimi-k3-1-pro-vs-deepseek-v4-1-flash-1017). Independent comparison. Index v5 of 28 October 2026: 46 and 32; speed of about 36 and 178 tokens per second. Accessed 29 October 2026. [^tinfoil]: [Tinfoil: pricing](https://tinfoil.sh/pricing). Provider's price list. DeepSeek-V4.1-Flash USD 0.65 input and USD 1.45 output per million tokens; description of spending limits. Accessed 29 October 2026. [^phala-api]: [Phala: confidential models](https://phala.com/confidential-ai-models). Provider's price list. DeepSeek V4.1 Flash 1017: USD 0.75 / 1.60; Kimi K3: USD 3 / 15 per million tokens; attestation of the gateway and confirmations of responses according to the provider. Accessed 29 October 2026. [^near]: [NEAR AI: pricing](https://near.ai/pricing). Provider's price list. Kimi K3: USD 3.30 / 16.50 per million tokens; different prices on the home page, discrepancy to be clarified. Accessed 29 October 2026. [^phala-gpu]: [Phala: confidential GPUs](https://phala.com/gpu-tee) and [pricing](https://cloud.phala.com/about/pricing). Provider's price list. 8 × H200: USD 44.20 per hour on demand, USD 32.20 with a 180-day commitment; 8 × B300: USD 65.60 (30 days), USD 62.40 (180 days); Intel TDX with NVIDIA confidential computing; no machines in Europe. Accessed 29 October 2026. The commitment is paid up front, with no refund for early termination. [^privatemode]: [Privatemode (Edgeless Systems)](https://www.privatemode.ai/). Provider's offer. Confidential API with servers in the European Union and remote attestation; a smaller choice of open models than at US providers; prices are not published. Confidentiality declared by the provider, not verified by us. Accessed 30 October 2026. [^katalog]: [Confidential Inference Directory: methodology and observations](https://confidentialinference.net/attestation). Independent directory, sponsored by a company that offers confidential inference. 0 of 95 routes verified end to end in the observation of 29 October 2026, 14:00:19 UTC. [^supermicro]: [Supermicro: AS-8126GS-NB3RT-01-G2 server](https://store.supermicro.com/us_en/8u-gpu-gold-series-as-8126gs-nb3rt-01-g2.html) and [GPU Per Hour: NVIDIA server prices](https://gpuperhour.com/blog/nvidia-gpu-prices). Manufacturer's offer and price summary. Complete server with eight B300 GPUs: USD 662,845.95 (21 September 2026). [^semianalysis]: [BigGo Finance on the SemiAnalysis analysis (7 October 2026)](https://finance.biggo.com/news/173039423dab79be). Industry analysis. A USD 200 plan gives, with intensive work, about USD 12,000 at list API prices; margin estimates, not company data. [^sglang-k3]: [SGLang: recipe for Kimi K3](https://github.com/sgl-project/sglang/blob/main/docs/cookbook/autoregressive/Moonshotai/Kimi-K3.mdx). Engine documentation. Verified configuration of Kimi K3 on 8 × B300; does not apply to Kimi K3.1 or to confidential mode. [^phala-cvm]: [Phala: confidential machine pricing](https://phala.com/pricing). Provider's price list. tdx.large (4 virtual processors, 8 GB of memory): USD 0.23 per hour; disk USD 0.000139 per GB per hour. Accessed 29 October 2026. [^itu]: [ITU: Facts and Figures 2025, internet use](https://www.itu.int/itu-d/reports/statistics/2025/10/15/ff25-internet-use). International statistics. The percentage of people using the internet in the world's regions in 2025 (about 6 billion people, 74% of the population); the regional shares in our calculation are an approximation: percentage × the region's population. [^cai]: [Confidential AI: pricing](https://confidential.ai/pricing). Provider's price list, prices of 26 October 2026. DeepSeek V4.1 Flash: USD 0.30 input, USD 0.03 cached input, USD 1.20 output per million tokens; 8 × B300 machine USD 60 per hour, minimum three months, allocation within a few days. Confidentiality declared by the provider, not verified by us. [^voltage]: [VoltageGPU: pricing](https://voltagegpu.com/pricing) and [scope of evidence](https://voltagegpu.com/). Provider's price list and declarations. 8 × H200 USD 64.62 per hour with per-second billing; the published evidence does not cover attestation of this configuration or of the NVSwitch switches. [^epoch]: [The Decoder on Epoch AI data (8 January 2026)](https://the-decoder.com/global-ai-compute-hits-15-million-h100-equivalents-epoch-ai-finds/) and [Epoch AI: AI Chip Users](https://epoch.ai/data/ai-chip-users). Data compilation by an independent research institute. Installed computing power for digital intelligence of more than 15 million NVIDIA H100 equivalents, demand of more than 10 GW. An estimate, not a measurement. [^confidentialinference]: [confidentialinference.net](https://confidentialinference.net): a comparison of providers that run models in trusted execution environments, with dated attestation checks; a site sponsored by the company Enclava, listing eleven providers at the time of access. Accessed 31 October 2026. --- # Omniforum: shared understanding, conversation and decisions (https://novusignis.org/en/whitepaper-omniforum.html) # Omniforum: shared understanding, conversation and decisions Version of 31 October 2026. This document develops the principles of the manifesto that concern Omniforum: how we want to help people understand one another, leave a lasting body of work after a conversation and make decisions together. It describes the direction in which the prototype is to develop. Each capability described requires a separate acceptance trial, and a description of the intended experience does not mean that it is available today. We add the results of the trials in chapter 9. *Translated into English by digital intelligence from the Polish original of 31 October 2026. Where the versions differ, the Polish text prevails.* ## 1. What Omniforum is and why it is being created Omniforum is to be a shared space for conversation where anyone can speak on any topic and be heard. It is being designed by the Novus Ignis Foundation in Poznań with people from every part of the world in mind, and the foundation wants to make it available as a digital public good with open source code. In it, digital intelligence helps, always under the author's control, to break a statement down into individual thoughts and to connect them with sources and with other people's thoughts. The conversations that participants choose to share leave behind a body of work you can return to. Omniforum is being created because in a stream of quick comments good arguments easily get lost, and the same disputes keep starting all over again. In a single sentence we convey many assumptions, values and expectations. As long as we understand each other, that is enough. When a misunderstanding arises or we need to decide together what to do, it helps to see what each of us means by the same words, what unites us and at which point our lines of reasoning diverge. In Omniforum we want people to be able to: - organize their own thought and, before publishing, check whether the interpretation proposed by the model conveys what they mean; - understand someone else's statement together with its assumptions, values and sources; - compare positions and see both shared goals and the point where the difference of opinion begins; - return to an issue after a week or after a year and see only what has changed; - return to their own past statements and see how their thinking has changed, with the right to change their mind; - follow everyday interests, such as cooking, gardening or repairs, and compare solutions that have been tried; - prepare an idea for a shared decision: describe it on an idea card, gather arguments and, once voting exists, put it to the community's vote. The point is not to pick the winner of a dispute. An argument that challenges our position can be just as valuable as one that supports it, and you can also bring an idea so that others help find a better one. You can get to know an issue from a short story, go deeper into a chosen argument and clearly end the visit. We consider a visit successful when afterward someone understands the issue better and can calmly return to their own affairs. Neither the number of clicks nor the time spent on the site is a measure of benefit. Breaking a statement down into individual thoughts is not meant to turn a conversation into a dry chart. In a dispute we often get irritated with each other not out of ill will, but because one sentence carries several assumptions and each person responds to different ones. The simple question “So you mean this and that, do I understand you correctly?” changes the tone of the conversation on both sides. The whole technical layer of Omniforum is meant to support precisely this reflex. A conversation is meant to leave room for changing one's mind without humiliation: you can admit a mistake, ask another question or finish without declaring agreement. A participant's dignity does not depend on whether they accept a conclusion; the assessment of a claim depends on its justification. Omniforum complements Suveria, the private space for work with digital intelligence: a thought can first be refined privately and then consciously brought into a shared conversation. The values from which both projects grow are described in the foundation's [manifesto](manifest.md). In what follows we show in turn: - how we break a statement down into thoughts and preserve their source (chapters 2 and 3); - how this help can work in other services (4); - how issues, issue maps, idea cards and initiatives come about (5); - what new medium can grow out of this (6); - how we protect privacy (7); - how we make shared decisions (8); - how we will check whether all this actually helps (9). ### 1.1. Everyday interests and a calm return to a topic Omniforum is meant to help not only with public matters, but also with practical knowledge and everyday interests: caring for plants, cooking, repairs or learning new skills. Take someone who wants their ivy to put out the longest possible shoots. In Omniforum they can follow this topic and read what others have tried: which location, watering and fertilizing worked, and what failed. Each such experience is to be recorded together with its author and source, and similar advice from different conversations is to end up in one place instead of getting lost in separate threads. This is how an issue map comes about: an organized collection of experiences, arguments and open questions that you can return to. How we build it from individual thoughts and connect it with sources is described in chapters 2, 3 and 5. Solutions can be compared according to one's own criteria. One person cares about speed, another about low cost, and a third about a plant that will survive neglect, so we are not looking for one answer for everyone. With each experience we preserve the conditions under which it was gained, for example the amount of sunlight, the temperature in the apartment or the type of soil. We also distinguish what someone observed from their explanation of why it happened. This makes it possible to set different methods side by side and choose the one that fits one's own situation. When someone returns to a topic after a week or after a year, they are meant to see only what has changed since their last visit: new experiences, corrections, results of trials and questions that still remain open. They decide for themselves how often they want to review what is new and whether to receive notifications at all. After reviewing the changes, they get a clear message that they have seen everything new on this topic, and they can calmly return to their own affairs. ### 1.2. What is free Reading Omniforum, publishing statements, dividing them into thoughts and atoms by hand, adding sources, consultations and voting are free, and money does not grant greater reach or a stronger voice. A cost arises only when someone uses help from a model, for example to break down a longer text. It is covered by the person's own balance in Suveria or by a model on their own device, and people without a balance can use a free queue, which the foundation serves with spare computing power within the limits of the support budget and earmarked donations. When the foundation maintains its own machine, we devote its spare power to such tasks, in a queue and without a guaranteed time. As long as we use only a confidential API, there is no spare power, because every operation costs money; in that case we fund the tasks of people without a balance only from donations earmarked for this purpose, until they run out, and we show this in the register. Once confirmed, a breakdown goes into the shared body of work, so no one pays again to open it. ### 1.3. Glossary The most important names used in this document, explained in more detail in the chapters indicated: - **Atom:** the smallest building block of meaning, a single concept or one specific thing, named with one word or a few (chapter 2). - **Molecule:** a thought built from atoms connected by bonds: a claim, a question, a value or a condition (chapter 2). - **Label:** a word by which we name an atom; one atom can have many labels, also in different languages (chapter 2). - **Bond:** a relation connecting atoms within a molecule, such as “is part of,” “has the property,” “causes” or “prevents” (chapter 2). - **Chain:** a sequence of molecules in which premises lead to a conclusion (chapter 5). - **Issue** (*Causa*): a question or matter that someone wants to return to because they want to understand, compare or do something together (section 5.1). - **Issue map** (*Mappa Causae*): a compilation of the chains of different people, including mutually contradictory ones, together with sources and open questions, that you can return to (chapters 1.1 and 5). - **Idea card** (*Charta Ideae*): a view that shows where we want to get to and what the next step is, with the idea's status and answers to eight questions (section 5.2). - **Initiative** (*Inceptum*): a proposal for a specific action that grows out of an issue map and an idea card (section 5.3). - **Omnium:** the open database of Omniforum's public body of work, available to other applications through OmniAPI (chapter 4). - **Ignis:** the stage of first sharing a thought; the manifesto calls the moment of publication itself the Ignis Ceremony (*Caerimonia Ignis*; chapter 3). - **Formatio:** the stage after publication in which the community considers the thought, adds arguments, sources and caveats, and looks for consensus (chapter 3). - **Consensus:** a shared agreement of the community, stated with its version, its scope, the people who confirmed it and dissenting opinions (chapter 3). - **Terra Incognita:** a voluntary ceremony in which we name the limit of what we can justify today and consciously begin looking for an answer to a new question (at the design stage, chapter 3). - **Principium:** adopting a principle as one's own, into one's Personal Canon (*Canon Personalis*; section 5.6). ## 2. Atoms, molecules and their origin We break a statement down on two levels. An atom is the smallest building block of meaning: a single concept or one specific thing, named with one word or a few, for example “drought,” “tree” or “our street.” We assign each atom a nature, that is, whether it is a person, a place, an event, a process, a value or a concept. Atoms help pinpoint exactly what we are talking about and notice when different people understand something different by the same word. A molecule is a thought built from atoms connected by bonds such as “is part of,” “has the property,” “causes” or “prevents”: a claim, a question, a value or a condition. It does not have to be a short sentence, because the meaning may require a caveat, an example or context. In a conversation a molecule can be supported with an argument or challenged, while an atom by itself is neither true nor false. An atom is a meaning, not a word. It has a permanent identifier and a definition, and the words by which we name it are its labels. One atom can have many labels: names in different languages, synonyms and context-dependent expressions. Drought is “susza” in Polish, “Dürre” in German and “siccitas” in Latin. A resident of Warszawska Street will call it “our street,” and someone else “the hood.” This allows people who use different words, also in different languages, to talk about the same thing, and whether a label really conveys an atom can itself become a topic of conversation. Where an established Latin name exists, for example in biology, medicine or law, we add it as a reference label. Linking an expression to an atom is a proposal that the author confirms, like the whole division of their statement. It can also be private information: the fact that “our street” means Warszawska Street reveals where someone lives, so the author decides whether others will see this link. An atom's definition is also recorded in Omniforum: it consists of other atoms and the relations between them, just as an encyclopedia entry is described by other words. This makes it visible what exactly we mean by a given concept, and the definition can be considered and improved like any other thought. Such a record does not depend on any language, and alongside it the definition can also be given as ordinary text, in many languages. When an atom used in the definitions of other atoms changes, we mark those definitions for review. Every breakdown, assignment and link can be made by hand, without a model, with the same tools and with the same result on the map. This is laborious, but it gives full insight into how something came about. The model proposes the same steps for convenience, and next to each element it is visible whether it was made by hand or from a model's proposal, and from which model. We do not treat the manual path as a fallback mode: we maintain it in every version and check in trials whether a whole issue can still be worked through this way. With each molecule and each use of an atom we record where they come from: from which text, from which version of it and from exactly which place, in what language and in what surroundings. This means you can always go back to the original and read a sentence in its context. We treat the same words written in two different texts as two separate occurrences, because in different surroundings they may mean something different; only a confirmed link connects them to the same atom. We also record who prepared the breakdown of a statement and when, and if a model helped, which model and with which version of the instructions. We distinguish three layers: - **Source statement:** what was actually published or submitted for analysis. - **Proposed interpretation:** proposed by a model or a person, with the author and the degree of verification indicated. - **Confirmation of meaning:** the author recognizes a specific version of the analysis as consistent with their own intention. A confirmation says only that the author recognizes their thought in the breakdown. It does not prove that they are right, does not mean that others agree with them, and is not a vote in a ballot. The author can correct the breakdown or leave their statement whole, without dividing it. A reader can break down someone else's statement in their own way, without waiting for the author. We show such a breakdown as the reader's interpretation, marked with their name or pseudonym. Only a version the author has confirmed is shown as the breakdown of the author's statement. The author's silence does not mean consent. The point of reference for portability is open annotation standards, such as the W3C Web Annotation data model, which describes linking annotations to a resource and to a fragment of it.[^web-annotation] It is not a ready-made model of all of Omniforum's relations. ## 3. From a natural statement to a shared analysis In Omniforum a thought passes through successive stages. The first is the Ignis stage, that is, the first sharing of a thought. A thought can originate in Omniforum: in that case the author, in the Ignis workshop (*Officina Ignis*), breaks it down into molecules and atoms with the help of digital intelligence, confirms that the interpretation conveys their intention and consciously publishes it; this chapter describes that process. It can also come from Suveria, where the author refined it in their own private workshop of thought; in that case they publish it right away, without the Ignis workshop. In both cases the manifesto calls the moment of publication the Ignis Ceremony, and after it the Formatio stage begins: the community considers the thought, adds arguments, sources and caveats, and looks for consensus. We call a shared agreement Consensus, and when there is no agreement, an honestly described dispute is also a result of this work. With a shared agreement we state its version, its scope, the people who confirmed it and dissenting opinions; we do not count silence as agreement. We talk naturally. The Ignis workshop begins with ordinary text or a selected passage, and the user sets the goal: understanding, extracting arguments, preparing their own response or comparing positions. Arranging the map by hand remains an option for those who want it. Digital intelligence proposes a breakdown and, next to each molecule, shows which part of the statement it derived it from. The author does not have to check everything one by one: the model is to point out the places where it may have strayed from the intention or where the statement can be interpreted in several ways. The author corrects these places and approves the whole. The entire breakdown can be accepted at once only when it is clear exactly what is being accepted and it is possible to go back to the details. In trials we will check whether approval takes more time than breaking the statement down oneself. A model that proposes a breakdown has its own habits: it smooths out unusual phrasing and makes different voices sound alike, and a ready-made proposal is easy to accept unchanged. That is why the path without a model's help is of equal standing, and from time to time we compare breakdowns prepared with and without a model's help. Example: the statement “It's worth planting trees on our street if we choose drought-resistant species” contains two molecules connected by a condition. The first, “it's worth planting trees on our street,” connects the atoms “tree” and “our street.” The second, “we choose drought-resistant species,” connects the atoms “species” and “drought” with the bond “resistant to.” The statement must not be reduced to an unconditional “It's worth planting trees on our street,” because the condition remains part of the meaning. If the model adds that the author considers drought a problem in the area, we mark this as the model's inference, not the author's words. We do not break everything down at once. We start a breakdown when it helps with a specific task, and only as deep as necessary. The first proposal can leave longer phrases intact, for example “drought-resistant species,” and the author or a reader breaks them down further where the meaning depends on a single concept or where it may become the subject of a dispute. The less unnecessary division, the lower the computing cost, the less work for the reader and the fewer mistaken connections. We are also considering a second, voluntary ceremony: a moment in which we name the limit of what we can justify today and consciously begin looking for an answer to a new question. Possible explanations are then marked as hypotheses. We call it Terra Incognita, just as old cartographers marked lands not yet explored on their maps. We have yet to settle its form. ## 4. Help with understanding where people already talk When we watch a video or read an article, today we are above all recipients. At most we can add a comment underneath, where it easily gets lost among others. We want people to be able to say, at a specific passage of a text or a specific moment of a recording: I disagree with this, I have an argument or a source for it. Such a comment is to go where it can be noticed and considered: to the issue map in Omniforum and, if the author or the service uses our add-on, also to them. Then the recipient becomes a participant. Omniforum is to be a place where scattered conversations come together into a whole: thoughts from different statements, their sources, arguments and the relationships between them go onto shared issue maps. People, however, talk in many places at once: on social networks, on forums, under articles, on mailing lists and in many other places. We do not want to move them all into one service. We want help with mutual understanding to be available wherever people want to come to an understanding, and Omniforum's body of work to be able to serve there as well. Such help works in two directions. From a conversation to Omniforum: the author can break their own statement down into molecules and atoms, check whether the interpretation proposed by the model conveys their intention and consciously bring it into a shared map, and a reader can prepare an interpretation of someone else's statement, marked as the reader's interpretation and linked to the source. From Omniforum to a conversation: at the user's request, the add-on can show that the issue has already been discussed, what arguments and sources were raised earlier and where the difference of opinion begins, so that a dispute does not have to start all over again. In doing so it uses Omnium (Latin for “of all”), the open database of Omniforum's public body of work, through OmniAPI, a documented interface for other applications. We are planning two routes. A service's administrator can build in the add-on, for example a button next to the input field or an embedded module; whether and to what extent they do so depends solely on their decision. Where a service has no such add-on, an open browser extension is to work. The extension helps only the person who installed it, works on what that person sees and selects, and does not change the service or what its other users see. If the user wants to bring the result into the conversation on the service, they do so themselves, according to that service's rules. Existing annotation tools, such as Hypothesis, show that working through an extension and through integration with a website is possible.[^hypothesis] The open W3C annotation standard, which we discuss in chapter 2, makes it possible to point to both a passage of text and a time range in a recording.[^web-annotation] Division into molecules and atoms and the confirmation of intention remain our own separate design task. Proposed operation of the add-on: 1. The user selects text and sees what scope will be read, by which model and where. 2. They receive a private proposal of an analysis, which they can correct or reject. 3. They can copy the result, keep it privately or prepare a publication in Omniforum. 4. Before publication they see the source, the authorship of the analysis, the rights to the material, the scope of visibility and any data about other people. 5. They decide separately about inclusion in a research dataset and separately about use for training models. A service administrator building in the add-on does not replace the user's decision to send content. The extension requests the narrowest possible permissions and does not send browsing history, a whole inbox or the content of neighboring conversations in the background; by default it analyzes only the selected material. Searching for earlier conversations on a given issue is also started by the user. Statements from private conversations also concern other people, so before anything is moved outside such a conversation, the add-on is to remind the user of this clearly. A local model can be one of the options, if quality and hardware requirements allow it. Omniforum is to be a community project: its body of work comes from what people themselves bring into it. We do not plan to automatically collect statements from the internet in order to break them down into molecules and atoms without their authors' knowledge, nor to build profiles of people from them. The fact that a text is available on the internet does not settle the right to share it further, to profile people or to train models on it. We would consider automatically analyzing statements from another service only if that service allowed it and every author whose statements would be affected had agreed to it beforehand. Anyone can, however, bring a public document or statement into a conversation and discuss it with reference to the source, to an appropriate extent and without attributing to its author approval of the analysis. ## 5. Issues, their maps and the story of an idea Molecules form larger wholes. Molecules are connected by relations: justification, caveat, condition, example and anticipated consequence. We call a sequence of molecules in which premises lead to a conclusion a chain; at each step we record what rule it rests on, what assumptions it makes and where it has gaps. An issue map sets side by side the chains of different people, including mutually contradictory ones. On the issue of trees on our street, one chain may lead from shade and coolness during heat waves to the conclusion that they are worth planting, and another from the cost of watering and roots under the sidewalk to the conclusion that shrubs would be better. When the same word means something different to the participants, we record it as separate atoms with the same label, so the source of the misunderstanding is visible right away. Each relation can be a proposal to be checked, and similarity of words helps search for connections, but it does not prove that two thoughts mean the same or that one follows from the other. The map lets you quickly see what the conversation is about and get to the original statement at any moment. Its summary, that is, the synthesis, shows how the material was selected, preserves important minority positions and indicates what data is missing. It does not have to reduce a dispute to one sentence that everyone will sign: an honestly described dispute is just as valuable a fruit of a conversation as a shared solution. An issue map opens calmly: first one main conclusion with a short description of its strength, and below it a few paths of justification to choose from. You can expand the one that matches your own doubt, go back or finish; the rest of the map stays at hand. With a conclusion we show separately how strong and independent the evidence is, how well justified the conclusion itself is and how broadly people have agreed with it. Full agreement among participants based on a single unverified account still leaves the conclusion weak. We have yet to work out the scale for these assessments and test it in a benchmark; we do not add apparent uncertainty to well-established facts. The manifesto (section 5.1) divides shared deliberation into facts, values and hypotheses. In Omniforum this corresponds to how a molecule is marked: whether it speaks of something verifiable, of a need or a value, or of a hypothesis. The idea card gathers these markings in its status (working, design, hypothesis, vision). A story of an idea can grow out of a map: an accessible guide that leads the reader step by step through what the question is about, what has already been established together, where the differences are, what they rest on and what could change the assessment. A good story gives a moment of closure: it is clear what has been established and where the dispute is still going on. When a new source appears, an important condition changes or an error comes to light, the conversation can be clearly reopened. Every change remains in the history. Merging, splitting and withdrawing molecules and atoms preserves references and earlier versions, so it is possible to reconstruct how the current form of the map came about. When a source changes, we mark the analyses that rest on it as requiring review, instead of quietly changing their meaning. When several publications repeat the same original report, we mark this, and when the origin cannot be established, we say that it is undetermined. A correction of a source points to the analyses that need to be checked again; it does not automatically invalidate entire conclusions or differing interpretations. ### 5.1. An issue as a point of return to a question An issue is a question or matter that someone wants to return to because they want to understand, compare or do something together. A molecule remains a single thought, while an issue gathers many molecules, chains, sources and documents around one need. An issue can be a public question, for example about trees on the street, but also a personal one, for example how to prepare for the next conversation with the bank: you can attach documents, earlier answers and further doubts to it. A permanent link to an issue leads to everything that concerns it: the question, the context, sources, arguments, the current state of knowledge and the history of changes. For each matter we indicate whether it concerns an event, a supposition, a choice of values or a plan, what time and what conditions it concerns, and what its assessment rests on. A vote can show a decision or the participants' positions, but the truth of claims is settled by sources and verification, not by the number of votes. Recording a question is not yet a decision to act. When someone takes on a task, we establish separately who is responsible for it, what the next step is and what we expect. Postponement or refusal leaves a justification and a way back to the topic. Participants choose for themselves who sees their issue: what they are working on privately in Suveria remains separate from material they consciously share in Omniforum. A private issue is handled in Suveria; it goes to Omniforum only when the author shares it with a group or publicly. A shared register of issues is a direction we will test in a separate project and trial. ### 5.2. Idea card: from a picture of the future to a first trial Bold visions are easily dismissed as detached from reality, and cautious plans as lacking a goal. The idea card connects the two: it shows where we want to get to and what the next step is. It gives the idea's status (“working,” “design,” “hypothesis” or “vision”) and answers eight questions: 1. **What is it?** A definition, the boundaries of the concept and a way of recognizing the expected result. 2. **What for?** Whose needs it meets, how it protects dignity and what an ordinary day looks like for a person who uses it. 3. **Why?** The problem in today's world, described with facts and sources. 4. **How, and from what?** The mechanism, existing resources that can be used instead of building from scratch, and the next step. 5. **What would have to change?** Technical, legal and economic conditions, as well as beliefs: which ones we commonly hold today and which ones would be needed. 6. **What balance, and what other paths?** Who gains, who bears the costs and risks of the transition, and what sensible alternatives there are. 7. **What trial?** What we will check first, with the participation of people who will benefit, those who will bear the costs and those whom the solution does not suit, and what result will lead us to change our minds. 8. **Has something similar happened before?** Transformations that once seemed impossible, together with what made them possible and the differences from our situation. A similar example shows that such a path is possible, but does not prove that we will succeed. A card is not a single molecule, but a view of many connected concepts, claims, sources, conditions and trials. Cards can be connected into a longer path; we then show the shared resources and dependencies, and test a chosen section as a whole, because confirming the individual parts does not yet confirm the whole path. The questions about the goal, the method and the necessary changes correspond to Erik Olin Wright's three criteria for “real utopias”: whether a solution is desirable, whether it can work and whether it can be reached.[^wright] The card's history shows how the bridge between vision and reality gained further spans. The first cards are to be the ideas from the foundation's manifesto. ### 5.3. From an issue to an initiative When a conversation on an issue leads to a shared agreement on what is worth doing, an initiative, that is, a proposal for a specific action, can be founded on its basis. An initiative grows out of an issue map and an idea card, so it keeps links to the molecules, chains and sources it rests on. Its core is a clearly described goal, conditions, the proposed action and a way of checking whether it has brought the expected effect. Before it goes to a decision about shared money, it needs the green path from section 5.3 of the manifesto: grounding in facts and the community's agreement on the values it stands on. An initiative passes through separate decisions: first a conversation about the goal and possible variants, then a decision on whether this version deserves further work, next a cost estimate and an assessment of feasibility, and finally a decision on implementation and the source of funding. Agreement on the goal does not yet settle the cost, the implementers or the deadline. A significant change of the goal, scope or budget creates a new version of the initiative and repeats the decisions it affects. Until binding voting exists, these decisions are made by the board after consultations, in the agile phase described in chapter 8. An initiative can be funded in two ways. The foundation can pay for it from the shared budget, on the terms set out in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md) and always with priority for Pillar I. It can also be opened to support from donors: a person or company that wants the initiative to come about covers its full cost with an earmarked payment. The payment does not give additional votes or influence over the foundation's decisions. In both cases the initiative is carried out by the foundation, which is responsible for it; we do not plan to fund projects run by people and entities outside the foundation. An initiative accepted for implementation has responsible implementers, a budget, a deadline and acceptance criteria. After completion we publish the actual cost, the result and the conclusions, also when the trial did not succeed. ### 5.4. Recurring questions and a shared position Individuals, teams and institutions often receive the same questions and objections in many places at once. The dashboard we are designing makes it possible to answer them once, in an organized way: it connects a position approved by the author with the questions, arguments and sources it concerns. The view shows above all the content and what is new in the objections, because the number of repetitions does not indicate the weight of an objection. The add-on described in chapter 4 can propose linking a comment to such an answer. Automatic matching is clearly marked and can be corrected; the author of the comment sees what it has been linked to and can report: “my comment is about something else.” A new argument gets its own place, even if it resembles an earlier one. The fact that the author has replied does not yet mean that their reply has convinced the other participants. The dashboard presents only a position confirmed by the author: a reply prepared by a model must be approved before it begins to represent them. Anyone following a thread can choose to be notified of a new reply. We want to show an example of such use, with a statement from years ago and a later change of mind, on the foundation's website. ### 5.5. Living documents and public review We want to use the same tools for the foundation's own documents, such as the manifesto and the whitepapers. They are meant to help group recurring comments while preserving the differences, the context and the experiences of their authors. We link each objection to a specific passage of the text, so that it is possible to check the reply, add a new argument and trace where a correction came from. In this way even a large wave of criticism can turn into specific corrections instead of overwhelming the authors. The current version of a document presents the authors' present position. The history preserves earlier wordings, the reasons for changes and the commitments made, so actions can be assessed against what was declared at a given time. A new wording makes it possible to fix mistakes, but it does not remove responsibility for earlier decisions. We are starting with our own website. On novusignis.org, next to the manifesto and the whitepapers, you can highlight a passage and with one click prepare an email to us with a comment that already contains the quotation and a link to that place. This is a simple first step: the message goes only to the foundation, and no public conversation arises from it. Eventually, comments on our documents are to go to Omniforum, where others will be able to see and discuss them. We record comments that lead to changes in the document's history. ### 5.6. The Personal Canon in a conversation about values The Personal Canon is a collection of principles that a person recognizes as their own. It is created in the private space of Suveria, and only the principles someone consciously shares go to Omniforum. We call adopting a principle as one's own Principium. This can be done at any stage described in chapter 3: with a freshly shared thought (Ignis), during the community's deliberation (Formatio) or after a shared agreement (Consensus). A Principium has a place in the canon also when it remains a dissenting opinion with respect to the majority. Canons can be larger or smaller, contain one another and partly overlap. Besides a personal canon there can be a community's canon or a description of the principles of a particular tradition, and setting them side by side shows where they meet and where there is no agreement. We distinguish principles adopted as a starting point from conclusions whose premises and chain of reasoning are visible. A lack of justification or a disputed step does not cross out a principle, but points to a place for further reflection. The facts on which a principle rests come from the map and its sources; adopting a principle into a canon does not settle whether they are true. A canon does not serve to judge people. It can, however, have a descriptive level of development: nascent (it contains at least one consciously adopted Principium), described (it is clear what exactly it adopts and what it does not yet settle), anchored (its principles are linked to an issue map and their origin is visible) or analyzed (its justifications, counterarguments and tensions between principles have been examined). We show separately to what extent the principles have a shared agreement. A specific version of a canon can be marked as publicly mature according to a transparent list of conditions: when it is analyzed, its principles have a shared agreement and their mutual consistency has been checked; the status remains open to revision. With the marking we state the version of the canon, the community that assessed it, the date and the scope of agreement. The level of a canon helps find places for further reflection, but it does not assess its author and does not affect the right to adopt principles, to take part in the conversation or to vote. ## 6. A new medium of explanation and conversation Just as YouTube gave rise to YouTubers and podcasts to podcasters, Omniforum can give rise to a new group of creators. They will probably decide for themselves what they are called; these are people who do not merely talk about the world, but lead viewers through an issue step by step, from the sources, through the arguments, to the place where the dispute begins. Their work does not end with publication. Each step of such an explanation is linked to molecules and sources, so the viewer can stop at any point, check how something is known and join the conversation. This medium is by nature a conversation. A reply does not go to the end of a long list of comments, but to the exact place it concerns: to a specific molecule, a step in a chain or a source. A microdebate can take place around such a place, that is, a short, organized exchange of arguments that ends in a clear state: established, disputed or data missing. The creator can respond to it, correct their explanation or honestly indicate where they disagree, and the viewer sees the story together with the replies made at each step. Anyone can also check how their own views fare when they meet the arguments of others: first privately in Suveria, and then, if they wish, publicly. Such explanations can take various forms. An editorial channel can prepare a reading path through a treaty, a draft law, a contract or a public statement and show its arguments and possible consequences. Accessible animated stories with a narrator's voice can be created from an organized issue map. Because they are based on connected molecules and atoms rather than finished recordings, a story can be updated as new evidence comes in and translated into other languages; we want to check the faithfulness of translations together, and each part of a story leads to its source. A reader can also compare several analyses of the same material. The community will decide when to develop animated stories, in line with section 6.3 of the manifesto. A good explanation distinguishes what is written in a document from its interpretation, from the assumptions of a forecast and from an assessment. With legal documents it states the status of the text and the date, because an explanation, even a very popular one, does not become a binding interpretation of the law by being published. When digital intelligence suggests possible consequences, each suggestion leads to a justification and is subject to verification by a person. This medium has a few simple rules: sources are always visible, correction is easy, everyone has the right of reply, conflicts of interest are marked, and different analyses of the same issue can be compared. We value not only the authors of the most visible stories, but also the people who improve the connections between thoughts, translations and accessibility, or who restore an overlooked perspective. A creator's reach and the support they receive from viewers do not prove that they are right. The order in which we show content has transparent rules and can be changed. We want to try out several such orders: chronological, by question, by documented contribution, that is, what an entry brings to the map and what it rests on, not the number of entries by its author, and by recognition across different opinion groups. In the last of these, we show an argument higher when it is accepted by people who judge differently on other issues. Opinion groups are calculated by the system from the ratings alone: no one chooses or names them, no one belongs to them and no one manages them. As a result, a thousand accounts rating in the same way will not raise anything that no one outside them recognizes. The effectiveness of this order has so far been tested with hundreds and thousands of participants (Polis, Community Notes), so we will try it once an issue gathers that many ratings. Statements from the open inbox, brought in without an account and without vouching, are shown in a separate layer that anyone can choose, and in every order after statements from an account or with vouching (section 8.1). Several orders side by side are meant to ensure that the whole space does not depend on a single popularity ranking. We want attention to follow what sheds light on an issue, not what stirs emotions most strongly. In the foreground we show the content and the source with its date. We do not keep public contribution counters or rankings of people, because they encourage a race and the flooding of the map, also by bots; the history of additions and changes remains available. With a new thought you can trace how it came about in cooperation between a person and a model, to the extent the author decides to show. ## 7. Privacy, quotations and corrections Each thought can be in one of three places: in the private space of Suveria, in a group that someone has chosen themselves, or publicly in Omniforum. A group that someone chooses themselves determines who sees the material; it is not an opinion group from chapter 6. Each of these places has its own permissions, and moving a thought further, for example from Suveria to Omniforum, always requires explicit confirmation. We limit data about other people to what is truly necessary for the issue at hand. You do not need an account to read, and you can also speak in a conversation without an account (section 8.1). Anyone who publishes from an account can keep it under a permanent pseudonym. The author of a publication is responsible for its content, and for a statement brought in through vouching, so is the person who brought it in; we will describe the foundation's obligations as operator with regard to reports before launch. Original statements, quotations, comments and analyses prepared by a model may have different terms of use. That is why with each material we record where the right to use it comes from and what limitations it has, instead of giving the whole a single, seemingly simple license. A change of visibility, for example from private to public, does not replace checking whether we have the right to the material. People have a right to privacy also when someone else quotes them. If someone attributes a statement to a person without a public source, for example from a private conversation, the person named as the author can report it. Anyone can report the lack of a public source, without revealing their identity. Until the matter is clarified, such a statement is not publicly visible. If it turns out to be made up or comes from a private conversation without that person's consent, we restrict access to it. We require confirmation that the report really comes from the person concerned only when that person wants to correct the words attributed to them, so that no one can correct other people's words. It is different with statements that have a public source, for example an article or a recording from years ago. The original remains, because people are accountable for their public words and those words can be discussed. The author can, however, add a correction or their current position next to it, and the reader will see both versions side by side: what someone said then and what they think now. At the reader's request, the add-on from chapter 4 can also show such a current position next to a quotation in an article outside Omniforum. The right to change one's mind does not erase history, but it does allow history to be supplemented. A model can openly assess an issue differently from the shared position. It then shows the basis for the difference and checks whether such an argument is already on the map. If it is not, it proposes an entry marked as a model's proposal to be checked; neither its publication nor similar statements by other models are evidence or the community's agreement. When someone withdraws their material, we delete it or restrict access to it in the systems we control, and we mark the analyses that rested on it. We do not, however, promise that it will disappear everywhere: public copies, widely distributed datasets and models already trained require separate discussion. We settle the legal and technical rules before we start collecting materials. A pseudonym or a cryptographic hash does not by itself make data anonymous.[^rodo-2026] ### 7.1. A private notebook in Suveria and conscious publication We store private materials according to the rules of the [Suveria whitepaper](whitepaper-suveria.md): above all on the user's devices, in encrypted containers, with an optional encrypted copy on the server. Only what someone consciously chooses goes to Omniforum. A thought refined in Suveria's private workshop is published right away, without the Ignis workshop described in chapter 3; the manifesto calls the moment of publication itself the Ignis Ceremony. After publication the rules of the public space or of the chosen group apply, and keeping a private record of the original does not give control over copies that may already have been made. ### 7.2. Moving one's body of work and OmniAPI We want to develop Omniforum as a digital public good: with open source code, documentation and data formats, with the Digital Public Goods Alliance standard in mind.[^dpga] Through OmniAPI other programs can read and exchange selected materials, but only with the consent of the people they belong to. The fact that the interface documentation is public does not open access to private data. Anyone can take their body of work to another tool. The export package contains the original statements with their versions, the indicated passages, molecules, atoms with their labels and the relations between them, authorship, sources, information about which interpretations the author has confirmed, and the terms of use. Thanks to identifiers and the history of changes, the same elements can be recognized after import, and another tool should show the original statement next to the molecule extracted from it, its source and its confirmation status. The recorded terms of use say what is permitted, but they will not force an external recipient to comply with them. A private export, publication, sharing for research and sharing for training models are four separate decisions. ### 7.3. Sources and attachments in a public conversation Large recordings do not have to be stored in Omniforum. We keep the issue map and the descriptions of sources separately, and a video or recording can stay with its provider. The reference then records the address, the author, the date of access and the relevant segment of the recording, and next to it it is clearly visible that we depend on whether the provider keeps the material and on what rules it has. Materials that we are allowed to store we can keep in the necessary excerpt, as a source copy or as a lightweight image with a text version. With each one we record its origin, the basis for its use, its version and a cryptographic hash, that is, a kind of fingerprint of a file that makes it possible to detect that the file has changed. A hash confirms only that the file is the same, not that its content is reliable. When a source changes or disappears, this is visible next to the claims that rest on it. We will choose formats, limits and retention periods so that they are useful, accessible, consistent with the rights to the materials and affordable. Before publication, anyone can check what copies will be made and how we handle corrections and requests for deletion. These rules concern the public body of work and are separate from the private, encrypted containers in Suveria. On what terms we store copies of sources and share excerpts of them with other applications through OmniAPI, we will describe before launch, with respect for copyright. ### 7.4. Abuse and unlawful content A conversation open to everyone needs transparent rules. In the first period Omniforum will be available by invitation, and every invitation will have a visible author. No one approves statements before publication. Moderation works after publication, according to a short, transparent list of reasons: spam, harassment, impersonation, the organized distortion of maps, data about third parties without their consent, and unlawful content. We will publish the rules for moderation and for limiting abuse before the first issue, not only when the first dispute arises. As long as the moderators are volunteers, the list of them is public. Anyone will be able to report unlawful content. Every restriction, for example hiding or removal, has an author of the decision, a justification and a route of appeal, in line with the EU Digital Services Act. The justification is recorded and visible to the author of the statement and in a public log, and an appeal is considered by a person not involved in the matter, other than the one who imposed the restriction. Automatic detection can only flag suspicious content, and a person makes the decision. We also hide unlawful content in the version history, leaving a visible trace that it was hidden and why. Before services are publicly launched, we will check and fulfill the operator's obligations: a data protection impact assessment, a way of reporting content and a point of contact under the Digital Services Act, informing people that they are talking with a system and labeling content prepared by a model under the Artificial Intelligence Act, terms of service with the right of withdrawal, and VAT settlement. We publish the results in the decision log. ## 8. Shared decisions and voting This chapter is about how to make decisions together and how to build voting that can be trusted. It is one of the most difficult parts of the whole project, so we write openly about what seems within reach and what we will have to work on for a long time. Some things are within our reach from the start: an open conversation for everyone, consultations without verifying participants, transparent rules announced before a vote, and a decision log with justifications. Harder are three matters that limit one another. First, every eligible person should have exactly one vote, but we do not want to collect biometric data or create a central identity database for this purpose. Second, the choice is to remain secret, and at the same time everyone should be able to check that their vote was counted, without any proof that would let someone else learn their choice. Third, voting is to be accessible to people from all over the world, including those who do not have a convenient document, and every additional route of access also opens a route to abuse. On top of this come pressure and vote buying, as well as the question of from when a result can truly bind the foundation. We go through these matters in turn: - who can talk and who can vote (8.1); - what zero-knowledge proofs provide, and what they do not (8.2); - by what paths the right to vote can be obtained (8.3); - who decides in difficult matters (8.4); - from when voting is binding (8.5); - how to guard against coercion and vote buying (8.6); - when a result can be suspended and how we treat people whom the system does not serve (8.7). Regardless of how we resolve these dilemmas, a few principles always apply. An organized conversation prepares a decision, but does not replace it: publishing a thought, confirming its meaning, supporting an idea and voting on spending are separate steps. Before each vote we announce the question, the list of eligible voters, the deadline, the rules for changing a vote, the method of counting the result and the validity threshold, as well as what happens in the case of abstention, a tie or a failure, and how one can appeal. The absence of a vote is not consent. A spending proposal requires the green path from section 5.3 of the manifesto, and we set the threshold of agreement in advance according to the importance of the matter, the scale of the spending and whether the effects can be reversed. Next to the overall result we show how the issue is assessed by researchers and practitioners in the given field and by the people who will feel its effects, on the basis of information that participants themselves decide to provide, separately from the secret ballot and only in groups large enough that no one's choice can be guessed. Those who have less time receive sources and a clear summary early enough to have time to read them. The question put to a vote grows out of an initiative (5.3): it is proposed by the proposer, and the wording of the card is open to comments for an announced period. We announce in advance the threshold of support needed to hold a vote. ### 8.1. Conversation for everyone, a binding vote after joining In Omniforum anyone can speak, contribute arguments and describe how a decision will affect their life, even if they have no connection with the foundation. A new, good argument can move a conversation regardless of who brought it. Anyone can also take part in consultations, that is, requests for opinions without verifying participants. A consultation shows what participants think, but it is not a vote within the meaning of the foundation's statute, which for votes requires confirmation of the uniqueness of each voter (§ 8(2)). Without verification its result can be distorted by someone having several accounts, so we treat it as a voice in the conversation, not a measurement. Anyone who wants to vote bindingly on shared resources joins the decision-making community voluntarily and free of charge according to transparent criteria: they state that they want to participate and obtain a voting credential through a chosen verification path. Anyone can read, without an account. There are three ways to speak in a conversation. From an account: the statement then has an author, and the account has a public rate limit. Without an account, through vouching for a statement: a participant with an account brings in another person's statement, is visible next to it as the one who brought it in and is responsible for it as for their own; the author remains without an account and can be signed with a first name or not at all. Without an account, through the open inbox: the statement goes to the “without vouching” layer, visible to anyone who chooses it, and in the order of visibility it is shown after statements with vouching; any participant with an account can vouch for it with a single action and move it into the conversation. The open inbox has a rate limit, and when it is flooded with automated spam, it can be temporarily closed, which we announce with the date and the reason. We do not use CAPTCHAs and do not approve statements before publication. None of the paths without an account gives the right to vote in a ballot. In the first period, when Omniforum is available by invitation (section 7.4), one can speak without an account through vouching for a statement; we will launch the open inbox when we open Omniforum more widely. Both paths without an account are a design that we will test in the first trial. An account can be created in several ways, from completely private (a key on one's own device, without an email address) to convenient (signing in through an external provider). For each way, before the choice is made, we show who outside the foundation learns that the account has been created and what data they see, and what happens if that provider disappears. The way an account is created does not affect a participant's rights; the right to vote is a separate certificate added to the account. We are working toward each person having one effective vote, with their choice remaining secret. For each permitted path, and also for combinations of them, we describe how it is secured and what known limitations it has. No one gets more votes for a payment, wealth, a position they hold or the number of tasks they have completed. If we do not admit someone, we give a justification and a route of appeal. ### 8.2. What zero-knowledge proofs provide Zero-knowledge proofs make it possible to show that someone belongs to the group of eligible voters without revealing who they are. In the Semaphore protocol they also make it possible to detect that the same voting credential has been used a second time in one vote.[^semaphore] They will not, however, check whether one person has received two different voting credentials: the verification path has to guard against that. The privacy of the whole vote also depends on registration, the application, the device, metadata and the method of counting votes, not only on the proof itself. We are designing a way for everyone to check that their vote was counted without creating proof from which someone else could read their choice. No single technical solution will replace checking the whole process. ### 8.3. Paths to a binding vote We are designing several paths to the right to vote, and none of them requires biometrics. The first uses a state document with a chip or a digital identity wallet, provided that the chosen system makes it possible to confirm the voting credential without handing protected data over to the foundation. We want to study reading the chip of a passport or ID card directly on a phone: the application checks the state's signature and creates a zero-knowledge proof on the device, without connecting to government servers and without handing personal data over to the foundation. The document never serves to log in or to decrypt data in Suveria, and the system's response does not reveal to an outsider whether a given document has already been used. It remains to be checked whether digital identity wallets will give the foundation a permanent pseudonym assigned only to it that cannot be linked with other services; only such a pseudonym makes it possible to detect a second registration by the same person. Without such a possibility, or a proof carried out on the user's phone, we will not base a binding vote on them. The first official path to the right to vote is to be the European digital identity wallet, because it makes it possible to confirm the voting credential remotely, from home, without handing protected data over to the foundation. This means that at first only people who have such a wallet will be able to vote bindingly, that is, residents of European Union countries, as those countries make the wallets available. For people from other countries, conversation and consultations are open from the start, and we will add a path to the vote once we find a way without biometrics and without a central database; until then we say plainly that the decision-making community is not yet worldwide. Reading a document's chip on a phone, described above, is a further path we want to study. Where possible, we derive the one-time credential tag from a person's permanent identifier rather than from the document number, so that a new passport, ID card or wallet of the same person does not give another vote. Not every document contains such an identifier, and a number with a small pool of possible values, such as the Polish PESEL, could be guessed by checking all the possibilities. That is why the tag is to depend also on a secret component split among independent parties, or to rely on the signature of the chip itself; we take this into account in the threat model before choosing a technology. The tag is to work only in a specific registration context and period: it makes it possible to detect that someone is registering a second time, but it does not reveal the permanent identifier or an easily guessed hash of it, cannot serve as a shared identifier across different services and does not make it possible to link a person with their vote. For the chosen solution we will check who could recognize a person or link their activities, and what happens when the same document is used again, renewed, the wallet is changed, the certificate is reissued or access is recovered. The second path is in-person meetings held in many places at the same time. Their rules are to make it difficult for one person to collect several voting credentials in the same cycle and to make it possible to check whether the organizers issue voting credentials to people who actually came. We are also examining what makes participation difficult for people who find it hard to get to a meeting. Combining paths leaves the risk that the same person obtains additional voting credentials, for example by having several citizenships or by using both a document and a meeting. At the beginning we accept this risk consciously, so that as many people as possible can join without biometrics, and over time, through further paths, also people from outside the EU. The goal remains a uniform system in which one person has one vote, and we are working toward it. We choose paths that are meant to make each additional voting credential require more time, effort and money than simply creating accounts online. Whether this barrier works and how much accessibility gains from it we will assess in trials, which is why we do not state in advance how many additional votes may appear. Before results become binding, we will show participants the known limitations, the results of the trials and the rules for responding to abuse, and we will adjust the scope of decisions to this assessment. Vouching, invitations and online tests alone do not give the right to a binding vote; they are sufficient for conversation and consultations. We publish aggregate data on how often each path was used and what checking them revealed, while protecting privacy. The number of certificates issued does not, however, say how many unique people there are. We also look into why people who did not manage to get into the register give up. Everything needed to vote is free. The free part of Suveria is enough to store the certificate; there is no need to pay money or to use paid features. Everyone keeps their certificate of the right to vote in their own private space in Suveria, separate from the ballot and from private content. Before each decision we announce when we close the register and how long one must have held a voting credential in order to vote. This makes it harder to add participants suddenly just before a vote, although it does not protect against voting credentials prepared long in advance. Anyone who recovers access keeps their seniority and one effective voting credential. Before casting a vote, everyone sees a short, accessible summary of the most important positions, objections and sources and can enter the full issue map. The summary shows the scope of the decision, the most important findings, sources, significant differences in assessment and the limits of knowledge, and each point can be expanded. It is meant to make it possible to decide with the uncertainty named, not to promise a correct result. We publish the summary in advance, and the parties to a dispute can report that their position was left out or distorted; we then show their own short account next to it. This is help with making an informed choice, not an exam or a test of understanding; anyone can see where the summary comes from and in which version it is, and report an overlooked perspective. ### 8.4. Who decides in difficult matters In difficult matters we propose three steps. First, an open issue map in Omniforum in which anyone can take part. Then a randomly selected advisory assembly: a group of people chosen by lot who receive time, materials and the opportunity to talk with the authors of proposals. Finally, all eligible voters vote. We draw from the register of eligible voters, using public randomness that anyone can check. If the assembly itself were to make binding decisions, it would need a separate, limited mandate to do so. We can also use a random sample for advisory polls. Volunteers who come forward on their own do not form a representative sample, and drawing from the register shows the opinion of this community, not of all humanity. We also take into account who declined to take part. ### 8.5. From when voting is binding Voting will gain binding force gradually. We will start with trials without financial consequences, then move on to real decisions about a small, separate pool and about actions that can be reversed, and we will extend the scope only after checking the results and safeguards. The first consultations will be a trial, without consequences for the foundation's operations, and their topic will be chosen by the invited participants. Before the first trial with real money we announce the maximum amount it covers, the result that would stop it, and a route of appeal against the suspension of a result to people independent of the board. About a thousand verified people is the scale at which we want to test registration and error handling; such a number does not yet prove that the participants are unique or that they represent a wider community. Before the next stage becomes binding, we announce to participants the scope of decisions, limits, thresholds, the way of raising objections and how we will carry out the result, and the system undergoes tests of registration, vote counting, credential recovery and resistance to abuse. We also check whether the scope of decisions is consistent with the current statute and the law, and establish who will carry out the result and how. For a result to formally bind the foundation, a change to the statute is needed, described in the statute change register as Z4 (votes and consultations, § 8 of the statute). As long as a trial does not confirm the uniqueness of each voter, we call it a consultation, not a vote within the meaning of § 8(2) of the statute. Whether voters are unique people we assess separately from whether we counted the certificates correctly. Until then the foundation operates in the agile phase described in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md): the results of votes and consultations are formally advisory, and the board works its way toward binding community decisions through successive trials, in which it announces in advance to what extent it will carry out the result, and after each of which it publishes a summary. The full launch of further pillars (II: open designs for durable objects, III: contact with nature) still requires all three conditions from section 6.3 of the manifesto together: voting that ensures one effective vote per person, funds for six months of maintaining Suveria and Omniforum, and a decision by the community. We will consider the first of them, one effective vote for each eligible person, fulfilled only when the paths that allow additional votes are replaced by a uniform system in which one person has one vote, and trials confirm this. This may take a long time, and we say so plainly. ### 8.6. Coercion and vote buying We want everyone to be able to change their choice until the end of the vote, with the last valid vote cast according to the rules being the one that counts. This can make vote buying harder: someone who paid or applied pressure does not know whether the voter later changed their mind, and cannot check it. For this to work, the whole protocol has to be built appropriately; a button for changing one's vote is not enough on its own. One model is the MACI protocol, but it has its own assumptions: the coordinator can decrypt individual votes, and resistance to collusion depends on the coordinator's honesty.[^maci] If we based voting on a similar model, a single coordinator would be replaced by a key split among independent parties, so that the foundation alone could not read an individual vote. This will not protect against someone who controls the device, the voting credential or the voter themselves until the vote closes; we take such situations into account in tests and in the description of limitations. We also check whether a vote can be linked to a person through metadata, and what happens when someone loses their device. ### 8.7. Suspending a result and the test of dignity The board may suspend the implementation of a result only for reasons announced before the vote, for example when it detects an error in the protocol, documented large-scale abuse or an infrastructure failure. Every suspension has a public justification and a date for reconsideration. Without these conditions, announcing binding voting would leave the board a free veto, and we do not want that. The hardest test of the whole system is how it treats a person who does not have a convenient document, has been wrongly rejected or raises uncomfortable criticism. A refusal of registration is never final or made solely automatically: anyone can appeal against it to people independent of the matter, with privacy preserved. No one loses the right to vote because of their views or for criticizing the foundation and the board. Anyone who cannot use any path to a binding vote still takes full part in the conversation and in consultations, and we keep looking for further paths for them. In the decision log we record the options considered, the reasons for the choice, dissenting opinions, the budget and later effects, and the conclusions from them return to the issue map. Who is responsible for what in carrying out decisions and in legal matters is described in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md). ## 9. How we will check whether Omniforum helps Everything we have described above is at the design stage. Before we build the whole, we want to check in small steps whether Omniforum really helps people understand one another better. We are starting with the simplest use, and we add further ones only when the previous one has passed the test. The first step is one complete path of a single thought. Someone enters their own text, digital intelligence proposes how to break it down into molecules and atoms, the author corrects what was misunderstood and saves the result together with its source. Later they return, open the record and use it in another conversation. We check whether anything was lost along the way: the exact wording of the original, its meaning, successive versions and who has access to what. In this trial we also check for interpretation errors that the model did not flag itself, and we compare the result with the text alone and with an ordinary summary. Reopening a record should not require a new, paid analysis just to recover an earlier result. The next steps are a simple add-on for one platform where people already talk, then the shared work of a small group on one issue, then a story of an idea prepared as text, and only at the end, separately, making decisions together. We do not add new integrations until we have checked that the help really makes people's work easier. We want to open the first public issue maps on two kinds of topics. The first will be claims and hypotheses from our manifesto and whitepapers, so that anyone can test Omniforum by challenging or improving our own assumptions. The second will be ordinary, everyday questions, from caring for plants to cooking recipes: on these it is easier to see, without tension, how conditions, sources and different experiences work. We are also analyzing an idea similar to the SETI@home program, in which people all over the world made the computing power of their computers available for analyzing data from radio telescopes. Omniforum's public work, such as breaking public texts down into molecules and atoms, translating stories or preparing summaries of maps, could use volunteers' computing power. Anyone who wanted to could then contribute without giving up their own time, by making some of their computer's power available and paying for the electricity used. This applies only to public data, because a volunteer's computer sees what it processes, and we would check the results by computing the same task with more than one person. It is an idea we are only just analyzing, but one of the paths we would like to explore. At each step we check whether it helps: how long it takes to arrive at a useful analysis, how many corrections, and how important, the author has to make, how often the model misinterprets the intention, whether the source is easy to find, whether different people can use the tool, how much it costs and whether a person can then explain the issue to others on their own. Before each trial we publish what result we will consider sufficient to add the next step, and we stick to it in the assessment. We consider returns to Omniforum a good sign only when they serve a new need, not a compulsion to keep checking on the system. We check separately whether participants are better at recognizing facts and unknowns, whether they find it easier to agree on shared action and whether the decision made achieves its announced goal. An improvement in the first does not prove the others, and agreement of views with the foundation is not a measure of success. We will accept an add-on for other services only when it meets several conditions: it sends nothing before the user chooses to, respects refusal, makes it possible to stop work at any moment, deletes working data and asks for each consent separately. A trial with people using such a tool for the first time is to show whether they understand the difference between an interpretation prepared by a model and the author's words. We will also check whether one's body of work can really be taken along. In this trial we open a selected analysis in an independent tool, without a connection to Omniforum, and compare the sources, passages, relations, history and confirmations. If the other tool does not support something, the import is to report it clearly, instead of quietly removing it or changing its meaning. In the trial of followed topics, described in section 1.1, we will check whether on returning one sees only the actual changes since the last visit, and whether content someone has already seen stays marked also after a correction, a change of source or the removal of an argument. Participants should be able to say what the review brought that was new, and decide for themselves whether they want to go deeper or end the visit. We will check separately whether a comparison of different solutions preserves the criteria and conditions under which they were tested. **Results log.** As of 30 October 2026: the Omniforum prototype is under construction, and we have not yet completed any of the acceptance trials described. We add each further result here with the date, the prototype version, a description of the trial and the conclusion, also when a trial fails. A good conversation can also give rise to materials useful in research, but only to the extent participants have agreed to. How we prepare them, under what rights we use them and how we keep them separate from tests is described in the [materials and benchmark whitepaper](whitepaper-materialy-i-benchmark.md). Novus Ignis Foundation, Poznań, KRS 0001269575. Website and manifesto: novusignis.org. Comments: salve@novusignis.org. ## Notes [^web-annotation]: [W3C (2017): Web Annotation Data Model](https://www.w3.org/TR/annotation-model/). W3C Recommendation. An annotation model linking the content of a comment with the indicated resource or fragment; a reference point for data portability. [^hypothesis]: [Hypothesis: Embedding Hypothesis in websites and platforms](https://web.hypothes.is/help/embedding-hypothesis-in-websites-and-platforms/). Project documentation. An example of embedding annotations in websites; it does not confirm automatic division into molecules and atoms or the preservation of the author's intention. [^wright]: Erik Olin Wright, *Envisioning Real Utopias*, Verso, London 2010. A research framework with three criteria: desirability, viability and achievability; not a ready-made assessment method. [^rodo-2026]: [GDPR, Regulation 2016/679](https://eur-lex.europa.eu/legal-content/PL/TXT/?uri=CELEX%3A32016R0679). Legal act. Pseudonymization does not settle anonymity (Article 4(5), recital 26); consent can be withdrawn, which does not invalidate earlier lawful processing (Article 7(3)). [^dpga]: [Digital Public Goods Alliance: standard](https://www.digitalpublicgoods.net/standard). Standard. Criteria for assessing a digital public good; a design direction, not recognition obtained from the DPGA. [^semaphore]: [Semaphore: documentation](https://docs.semaphore.pse.dev/). Protocol documentation. Proof of group membership without revealing identity, and detection of a repeated signal within a given scope; the protocol does not establish that one person has received only one voting credential. [^maci]: [MACI: introduction](https://maci.pse.dev/docs/introduction). Protocol documentation. Only a trusted coordinator can decrypt the votes, and resistance to collusion holds only if the coordinator is honest; a dishonest coordinator cannot, however, censor or alter the process. Accessed 28 October 2026. --- # The Novus Ignis Foundation's operating principles: upkeep, cooperation and the handover of power (https://novusignis.org/en/whitepaper-zasady-fundacji.html) # The Novus Ignis Foundation's operating principles: upkeep, cooperation and the handover of power Version of 31 October 2026. This document develops the manifesto on organizational matters: how we maintain the services, how we manage shared money, how we cooperate and how we want to hand over power. It describes the principles we intend to adopt as regulations, and the changes to the statute they will require. The whitepaper itself does not change the statute, the powers of the foundation's bodies or the rules of accounting. *Translated into English by digital intelligence from the Polish original of 31 October 2026. Where the versions differ, the Polish text prevails.* ## 1. The statute, the manifesto, the regulations and this document The Novus Ignis Foundation from Poznań (KRS 0001269575) wants to build an alternative internet based on human dignity. Its first pillar consists of two services: Suveria, a private space for working with digital intelligence, and Omniforum, a shared space for conversation, understanding and decisions. This document describes how the foundation intends to make decisions, account for money, organize work and hand over responsibility; its values and direction are described in the manifesto. Novus Ignis is a Polish foundation registered in the European Union (Poznań). In international terms it is an entity without shareholders that pays out no profit to anyone: its assets serve only the goals set out in the statute. For our services we charge fees no higher than their actual costs, and we fund development and the reserve from voluntary donations. The following chapters describe, in turn: - Chapter 2: the agile phase, that is, how we make decisions before binding community voting takes effect. - Chapter 3: services as close to cost as possible, that is, a bill for actual use, rounding up the bill as a donation, and prepayment, limit and refunds. - Chapter 4: a budget that can be understood, that is, five kinds of money and a public register of contributions and expenses. - Chapter 5: work, fixed-term roles, reimbursement of costs and paid commissions. - Chapter 6: why personal support for co-creators flows outside the foundation. - Chapter 7: the register of benefits and connections. - Chapter 8: the conditions for the further pillars and two paths for small initiatives, that is, the transitional board pool and earmarked initiatives. - Chapter 9: the path for handing over the founder's special powers. - Chapter 10: the conditions that every step toward digital intelligence taking part in management must meet. - Chapter 11: open workshops, trials before full implementation and public goods, including licenses. - Chapter 12: recording decisions and mistakes, public registers, disputes and the way rules are changed. Our commitments have several levels. The statute is the foundation's law: it binds the foundation formally, and changes to it are adopted by the board in accordance with the statute and submitted for entry in the National Court Register. The manifesto is a commitment to people: it says what we want to do and why. The regulations, adopted by the board under § 14(2) of the statute, describe detailed procedures. The whitepapers show how things are meant to work and where the limits lie of what we can provide today. This document is at the same time a draft of future regulations: it describes the principles before we adopt them formally. A higher level sets the limits for a lower one. The manifesto does not promise anything the statute forbids, and regulations may not contradict either of them. When we want to go further than the current statute allows, we write about it as an intention, and the [statute change register](statut/rejestr-zmian.md) shows how things were, what we want to change, for what purpose and with what intention. We publish it together with all versions of the statute.[^statut-2026] For the principles described below, the most important provisions of the statute today are these. The foundation does not conduct registered for-profit business activity (§ 7(1)), and payments from paid public benefit activity cover only its costs (§ 10(4)). The foundation's assets may not be used for the benefit of members of its bodies or volunteers on terms other than those applying to third parties (§ 11(2)). The foundation's work is based on the unpaid work of volunteers, and only services of external professional service providers may be commissioned for pay (§ 12). The founder appoints and dismisses the board, which serves for an indefinite term (§ 13), and a change to the statute requires a unanimous resolution of all board members and the prior consent of the founder if the founder does not sit on the board (§ 15(1)). The framework for the board's actions is also set by the Act on Foundations.[^fundacje] **As of 31 October 2026.** The first payment into the foundation's account is the founding fund of PLN 1,000 from the founder; from that day we accept donations by bank transfer in zlotys, euros and dollars and show them in the public register of contributions and expenses. Of the registers in chapter 12, today we keep the statute change register and the register of contributions and expenses. We will prepare the regulations referred to in this document before the first prepayment. Before we accept it, we will have: - terms of service reviewed for compliance with consumer law; - a legal opinion on payment costs and prepayments; - a separate bank account for prepayments; - rules for refunding balances. ## 2. The agile phase: how we make decisions before voting takes effect Until binding community voting is launched, the foundation operates in an agile phase. The board makes day-to-day decisions within the limits of the law and the statute, including when a quick response is needed to changes in technology, in public debate and in people's needs. We do not put development on hold until a full voting system exists: during this period, it is the board that decides on spending on the maintenance and development of Pillar I, that is, Suveria and Omniforum. The statute allows the board more than it wants to use. Our goal is for all the decisions we hand over to the community to be made by its vote and to be binding. For this we need a mechanism that works and that we know how to maintain, which is why we are reaching it step by step. Until then, the board, by choice and not out of legal obligation, asks the community for its opinion in consultations on significant matters and publishes the justification for every significant decision, including a refusal. A consultation does not require verification of participants, so one person can cast many votes in it: it shows what the participants think, but it is not a vote within the meaning of the statute. Before every consultation, the board announces how it will treat the result. Participants can challenge any decision of the board, and the board responds to objections publicly. The agile phase ends gradually, just as the power of voting grows gradually, and every step is a trial. First we run trials without financial consequences, then votes on real decisions about a small, separate pool and on actions that can be reversed, and after that we extend the scope. Before every trial we announce its scope, its duration and how the board will treat the result. After it ends, we publish a summary: whether the result could be carried out, how the trial affected the running of the foundation and decision-making, what mistakes and abuses we detected and what we are changing. We extend the scope only on the basis of these summaries, because we do not want to take on commitments we cannot keep. The stages, the verification of participants and the protection of ballot secrecy are described in the [Omniforum whitepaper](whitepaper-omniforum.md). For the result of a vote to bind the foundation formally, and not only by the board's choice, a change to the statute is needed (Z4 in the change register), because today § 8 merely allows votes and consultations. As long as a trial does not confirm the uniqueness of every voter, we call it a consultation and not a vote within the meaning of § 8(2) of the statute. Once voting takes over a given type of decision, the board may suspend carrying out the result only for reasons described before the vote, for example a detected error in the protocol, documented large-scale abuse or an infrastructure failure, with a public justification and a date for reconsideration. ## 3. Services as close to cost as possible In this chapter we describe how people pay for Suveria and the foundation's other services. The principle is simple: everyone pays for what they actually use, together with a fairly calculated share of the shared costs, sees what they are paying for and sets the limit of their own spending. The fee covers costs and nothing more, and anything someone wants to give beyond that is a voluntary donation, which we show separately. ### 3.1. A bill for actual use The bill covers actual use of the services in a given cycle and a transparently assigned share of the fixed costs that keep the infrastructure ready. In it we count computing, memory, data transfer, security, payments, accounting, support, maintaining continuity, and the wear and replacement of hardware. We divide expenses into one-time, fixed and usage-dependent. We distribute shared costs according to a public rule, which we will choose in the pilot from two variants: in proportion to use, or equally among active accounts. We test the rule under different service loads so that the bill shows both direct use and the cost of availability. The billing units and the estimated cost of a task can always be checked, and everyone sets their own spending limit. When we change a price, we publish the reason and the impact on users. Savings lower the fees, while we fund development, security reviews and the reserve from donations, according to the adopted rules. The size of a person's payments does not increase the weight of their vote or any other rights of a participant. We keep the balance in a single currency: the one in which the foundation pays for most of its infrastructure, today US dollars; we state the prices of services in the same currency. We also want to accept payments in other currencies: we convert such a payment once, at the time of payment, at the exchange rate the participant sees before paying. We convert the cost of a model billed in another currency at the rate at which the foundation actually paid, and we show this on the bill. We announce a change in the currency of the balance in advance, like a price change. The foundation keeps its books in zlotys. The foundation does not make a profit from its services. The fee covers their costs and no more: this is what § 10(4) of the statute provides, and under the Act on Public Benefit Activity and Volunteering a fee higher than costs would turn this activity into business activity, which the foundation does not conduct.[^pozytek] A price close to cost does not settle everything, however. Before introducing fees, we will check in a legal and tax opinion the type of this activity, all the conditions in Article 9 of the Act, including those concerning remuneration, the VAT treatment, and what we do with any surplus. That is why all support beyond cost is a voluntary donation, separated from the fee on the bill and in the books. For the same reason we do not create cryptocurrencies in order to sell them and do not earn money from trading them; we use ledger technologies only where they help with transparency and verifiability. We include community work that reduces expenses in the plan, together with information on whose availability the service depends on and who can stand in for whom. At the start, when there are few participants, the fixed cost of the private environment is spread over few people. This means one or two confidential machines running the gateway, memory and users' files; according to the calculation in the [Suveria whitepaper](whitepaper-suveria.md) (section 8.1), together they cost USD 170–340 per month. With 30 participants that is about USD 6–11 per person, with a hundred USD 1.70–3.40, and with a thousand a few dozen cents. We show this cost separately and cover it from a pilot support budget collected in advance, for example from donations earmarked for Suveria, with a specified amount, period and rules for what happens once it runs out. We do not assume that we already have such funds. A participant's share of the shared costs is announced before the given billing cycle and always fits within their limit, and the bill shows what part of the cost is covered by the support and what part by the participants. The end of the support does not create retroactive charges, and unused prepayments remain the property of the participants. ### 3.2. Rounding up the bill as a donation When an account is created, we ask whether the participant wants to round each bill up to the full dollar or a similar equivalent in the local currency. We ask instead of selecting the option on someone's behalf, because the law requires express consent to every additional payment (Article 10 of the Consumer Rights Act), and the choice is meant to belong to the participant. The difference, which arises with almost every bill, then goes as a donation to the shared budget, from which we build a reserve for maintaining the services and further activities. We explain what this money goes to and encourage this option, but we do not select it on anyone's behalf; the decision can be changed at any time with a single setting. The bill shows the fee for the service and the amount of the donation separately. We book donations from rounding up separately, and show their total and how they are used in regular statements and in the public register of contributions in chapter 4. Larger, voluntary support for further development is a separate decision by the participant: it goes into the shared budget, whose use is to be decided by the community, and until voting is launched the rules in chapters 2 and 8 apply. ### 3.3. Prepayment, limit and refunds When paying in, the participant enters a single amount, and exactly that amount goes to their balance, regardless of the payment method. Below it we show how much processing the payment costs, and a button that can be used to voluntarily cover this cost with a donation; we do not select it on anyone's behalf. The donation is separate from the balance. The transaction fee is also charged on the donation itself, so the button states the amount that covers the cost of processing the whole payment. Example: for a USD 20 card payment and an example transaction fee of about 2.8% of the whole payment, the button proposes USD 0.58. Without the button we charge USD 20.00, the whole amount goes to the balance, and the cost of processing the payment is covered by the shared costs. With the button we charge USD 20.58: USD 20.00 goes to the balance, and USD 0.58 is a donation that covers the cost of processing this payment in full. Before the payment is confirmed, we show the single highest amount to be charged to the card, and separately the service balance and each selected donation. The regulations will show one complete example, from payment to refund. When paying in, the participant also sets the most they want to spend, for example 20 dollars on work with digital intelligence. They can separately declare what part of the unused amount, up to a stated ceiling, they give to the foundation as a donation, for example: “if I use less, give up to 5 dollars of the remainder to the shared budget.” Without such a declaration, the whole unused amount remains their property. We propose to account for the actual transaction fees of a given period in the shared costs, in proportion to payments and in the same way for every payment method, and to show them on the bill as a separate line; donations toward payment costs reduce this total. This is a choice based on solidarity: people who pay with a cheaper method co-finance the transaction fees of more expensive ones, which is why we show which methods cost the community the least. The method of calculation and the largest possible share of the shared costs are known before the billing cycle and fit within the participant's limit. The account shows separately the balance, the funds reserved for requests in progress and for the announced share of the shared costs, and the amount available for use. Example with a balance of USD 20 and no requests in progress: | In the account | Amount | | --- | --- | | Balance | USD 20.00 | | Reserved for the announced share of the shared costs | up to USD 0.60 (in the example, 3% of the balance) | | Available for new requests | at most USD 19.40 | After settlement we release the unused part of the reserved amount. Settlement does not create unagreed additional charges or a negative balance; any excess over the announced share is covered from a reserve set aside in advance. EU rules prohibit charging fees for some payment instruments, including consumer cards covered by the regulation of interchange fees and SEPA transfers, with certain exceptions.[^psd2] The division of costs described here is our proposal for legal assessment, not an automatic consequence of this prohibition. We choose the payment operator after comparing transaction fees, and we set the minimum payment as low as possible, taking into account the cost of processing it; its current amount, currency and effective date are given in the regulations or the price list before payment. At the start, the lowest payment is to be one dollar or its equivalent; even that goes in full to the participant's balance, and only what the participant explicitly declares will go to the shared budget. We keep participants' prepayments in a separate bank account and do not cover the foundation's costs from them before the service is provided; we show their total in the register. We will check before the first prepayment whether the law requires additional protection. Unused balance can be recovered at any time. The refund option is clearly visible in the account and requires as few steps as possible; eventually it is to be a single button. The refund goes to a previously verified payment method, by default through the same channel the participant used to pay; we set the refund method when the account is created. After the participant closes the account or the service is shut down, the balance is returned automatically by the same route. The time limit for making a refund, any costs of it and the procedure when the original payment method is unavailable will be set out in the regulations. They will also protect the service against abuse, for example series of payments and refunds without using the service, within the limits of consumer law. The first refund in ordinary use remains simple. With repeated payments and refunds without using the service, we may limit card top-ups or ask for a bank transfer. A balance that no one has claimed remains in the participant's account and does not go into the foundation's budget. The procedure after a participant's death or in other special situations is governed by the law of the relevant country. The time limit for refunds, settling differences between the estimate and the actual cost, and other details will be set out in the terms of service, prepared before the first prepayment is accepted and reviewed for compliance with consumer protection law. ## 4. A budget that can be understood We want everyone to be able to understand where the foundation's money comes from, how much of it the foundation can really spend and what it spends it on. That is why we do not show a single amount in the account, but divide the money according to whom it belongs to and what it is intended for. In our accounts we separate five kinds of money: | Kind of money | What it means | Can we spend it on new purposes? | | --- | --- | --- | | Funds owed to users | Participants' unused prepayments and balances | No, they belong to the participants | | Money entrusted for a specific purpose | Earmarked donations, for example for a specific initiative or for the Suveria pilot | Only for that purpose | | Commitments made | Money we have already allocated to planned expenses | No, it is already allocated | | Safeguard for maintaining the services | A reserve without which Suveria and Omniforum would stop working | No, it protects the continuity of the services | | Funds free to be allocated | What remains after all of the above | Yes, always with priority for Pillar I (chapter 8) | We assign each amount to only one category, so as not to count anything twice. The cost of maintaining the services covers everything without which Suveria and Omniforum would stop working: computing power, data storage and backups, domains, payment processing and mandatory accounting. That is why the balance of the bank account is not a budget to be allocated, and the reserve is not made up of money that has been promised but not yet paid in, or of money owed to someone else. The categories of costs, the terms of larger contracts, decisions and their justifications are public. Every expense is to be traceable to the decision that led to it and to its result. At the same time we protect personal data, access credentials and information whose disclosure would increase the risk to the infrastructure. On the foundation's website we want to keep a public register of contributions and expenses, as up to date as possible. Every payment into the foundation's account and every expense appears in it with a date, an amount and a short description, and for expenses we add what the money went to and why it was needed. We protect people's data: we name donors only if they themselves wish it, and describe other payments in general terms, for example “donation from a private individual.” We do not publish the reference text of transfers verbatim, because it often contains personal data; we prepare the description of each entry ourselves. The register allows anyone to follow the foundation's finances, but it does not replace the accounting books or the financial statements; we correct an erroneous entry openly. We will test the mechanism on the first payment: the founding fund of PLN 1,000. Every project has a limit, a responsible person, a result to be verified, review dates and rules for handling money that is not spent. We can pay out money in stages. Exceeding the budget requires a new decision, and concluding a project includes accounting for it and drawing lessons. We also show what share of expenses directly serves the statutory goals. We calculate it from money actually spent, and show reserves separately, together with definitions of the categories, the way shared costs are assigned and an explanation of changes between periods. We show this proportion together with the cost structure, because security, maintenance and legal services also have real value. People with smaller financial means can use the services on solidarity terms, always with the source of funding stated: from payments that donors have earmarked for this purpose, from computing grants received by the foundation or from access paid for by a sponsor. We do not disclose what the people who receive such help are working on. The rules of assistance protect them from stigma and allow the accounts to be checked. **Readiness to take over the inheritance.** The manifesto (section 1.1) describes how, after a period of investment frenzy, cheap computing power and hardware may be left behind. When computing power or hardware becomes cheaper, or someone wants to give it to the foundation, we calculate the full cost of use, not only the purchase price: power, cooling, space, operation and people's time, as well as the cost of keeping it ready. We tend to rent rather than buy hardware that ages quickly; durable resources, such as access to the network or energy, may be worth a longer contract. We accept donated hardware according to the rules of the record of donations and the register of benefits. We prepare the way such decisions are made in advance, so that when prices fall we can act quickly and transparently. ## 5. Work, roles and commissions In republican Rome the highest offices were usually held for a year and without pay. Tradition remembers Cincinnatus: according to Livy, in 458 BC envoys found him working in his field with the news that he had been entrusted with supreme power for the duration of a war. On the sixteenth day, when the danger had passed, he gave it up and returned to his farm.[^cyncynat] This image captures well how we want to work: someone who knows a given matter takes on a task for a time, carries it out and returns to their own life. The foundation is to be able to operate without anyone having to commit to it permanently, and holding roles in it is not meant to grow into a profession. From history we also take a warning: in Rome unpaid offices could be held mainly by the wealthy, which is why we reimburse costs and do not expect sacrifice beyond people's means. Below we describe voluntary work and fixed-term roles (5.1), reimbursement of costs (5.2), paid commissions (5.3) and the workshop in which a task can be taken up (5.4). ### 5.1. Voluntary work and the possibility of leaving The foundation relies on the work of people who want to develop this direction, and on everyone being able to reduce their involvement or hand over their role. No one has to prove the sincerity of their intentions with readiness for unlimited sacrifice. Continuity of operation must not depend on a volunteer being available at any hour: if there are not enough people who can look after a given service, we reduce its scope or availability. Before taking on a new commitment, we check whether we can carry it out and sustain it, not only whether there is interest in launching it. Today the foundation relies on the unpaid work of one person, and this person is the only member of the board. We say this because the continuity of the services depends on it; that is why the list of access rights, the continuity plan and a second person on the board are first tasks, not later ones. If this person became unavailable, refunds of balances could be delayed until new people are appointed; the balance remains the participant's property and is not forfeited. We will remove the sentence about the delay once there is a second person on the board with their own access. We describe a role with decision-making powers by its purpose, the scope of decisions, the way it is entrusted and withdrawn, its duration, its review date and who stands in. Extending a role requires a public decision. Ending it includes a summary of the work, handing over documentation to successors and updating access. Time limits on powers can be combined with further help from the same person as an experienced co-creator. We are looking for a second person for the board. The role: shared responsibility for finances, access and commitments to users; without pay (§ 12(3) of the statute); a time commitment agreed in advance, with a review of the collaboration after a year. Experience needed: managing an organization's finances, or responsibility for a service used by other people. Personal funds are not a requirement and give no priority. The path: three months of taking part in deliberations and insight into the finances as a volunteer, then appointment by the founder (§ 13(3) of the statute) with a public justification and an entry in the register of benefits. The absence of pay for holding a position does not remove the other benefits that power brings: influence, contacts, information and the allocation of tasks. That is why disclosure of conflicts of interest, recusal from deciding one's own matter and the possibility of reviewing decisions are needed. The number of hours worked, the visibility of a contribution and the help received remain separate from the right to vote. ### 5.2. Reimbursement of costs Reimbursement of costs is not payment for time. It means returning money that someone has paid out of their own pocket to carry out a task of the foundation, for example for a ticket to a working meeting, or to pay for a domain or a service. A board member may receive only reimbursement of documented, necessary costs incurred on statutory tasks on behalf of the foundation (§ 13(5) of the statute). For volunteers, the foundation covers the costs of business travel and per diems on the same terms as for employees, unless the volunteer releases it from this in writing; it may cover other necessary costs and training.[^wolontariat] Only what is necessary for a task counts as its cost. A ticket to a working meeting may be a cost of the task, while the private part of a trip is paid for by the participant. Every reimbursement requires a document and goes into the public statement of costs. ### 5.3. Paid commissions Today the foundation can commission paid services only from external professional service providers, for example an accounting office (§ 12(3) of the statute). Other work, including work on software, is based on volunteering (§ 12(2)). At the start we want to check whether everything we need can be created in this way. Only if it turns out that some of the work cannot be done like this will we change the statute so that the foundation can commission specific paid work from people and companies outside the board, on the basis of civil-law contracts. We will treat as such a signal, for example, a situation in which for two consecutive weeks no one takes on the on-call duty for a service or its availability falls below the announced threshold. We change the statute only when this is needed for the next step, and we apply the new rule immediately rather than adopting it in advance. The draft and justification of this change are described in the [statute change register](statut/rejestr-zmian.md) (change Z2). There will be no salaried jobs at the foundation, because that is part of its idea: the work is done by volunteers and, when necessary, by contractors carrying out specific commissions. If we ever buy our own hardware, we will commission on-call duty for it from a company as a service. A commission concerns specified work with a result to be accepted, not ongoing work under someone's direction, because the law treats such work as employment regardless of the name of the contract.[^kp22] If this change happens, the rules for commissioning will be set out in regulations. They are to specify at least: - a public scope of the task and acceptance criteria; - a justified price; - the exclusion from decisions of people connected with the contractor; - aggregate information on the commissions concluded. Board members do not conclude paid contracts with the foundation: the statute allows them to receive only reimbursement of documented costs (§ 13(5)), and no one is to earn money from holding a position in the foundation. A volunteer may accept a commission for work other than what they do as a volunteer, and clearly separated from it. Before this happens for the first time, we will have this situation checked legally. ### 5.4. One's own workshop and prepared context Taking on a task is to open a space for it in the environment chosen by the participant. The digital intelligence assistant receives what it needs for the work: the foundation's values, the goal and limits of the task, the current findings from Omniforum, relevant arguments and open questions, sources, a resource limit and the way the result will be assessed. The participant can check this context, point out gaps and improve how their own position is presented. Joining a task makes available only a chosen scope of data and permissions. Passing on context, access to tools and publication of the record of work each require separately defined consent, and private conversations, files and access credentials remain outside this scope unless a person consciously decides otherwise. Before we make such a connection available, we will describe what data leaves the private environment, to whom it is passed and how it is processed. After a break, the workshop is to show what has changed and where work can be continued, and the result and state of the task are to be transferable to the next person without revealing the participant's private surroundings. In a trial we will test starting a task, returning after a week and handing over work: the time needed to find the context, the accuracy of summaries and the observance of privacy limits. Until we have our own servers, we can cover from Pillar I funds or from earmarked donations the costs of tools that volunteers need to create and test the open source code of Suveria and Omniforum, for example access to models through an API or programming tools. Such support has a public limit in the budget and is not remuneration. It is permitted by the Act on Public Benefit Activity and Volunteering (Article 45(2)); we will confirm the scope of such costs in a legal opinion.[^wolontariat] ## 6. Personal support outside the foundation A better quality of life for co-creators is part of the foundation's direction, but personal support does not come from its assets and does not pass through its accounts. The statute forbids using the foundation's assets for the benefit of members of its bodies and volunteers on terms other than those applying to third parties (§ 11(2)), while anyone can privately support a person of their choice. That is why help in meeting needs such as housing, food or medical care flows directly between people, for example through patronage platforms or direct payments. The foundation does not act as an intermediary in it, and the tax settlement of such support is the responsibility of the person who receives it. Earlier we considered a shared solidarity pool run by the foundation. We are giving it up: it would be hard to reconcile with the statute, and it would make someone's livelihood depend on decisions of the foundation's bodies. Personal support is not tied to a role, a task or the right to vote. Declining a further task, handing over a role or criticizing a shared decision does not affect a person's rights in the foundation. Direct support does not remove all dependencies. Popularity can steer money toward the most visible people, and living on donations can create a need to hold on to a role. Supporters' expectations and conflicts of interest can also arise. That is why we enter in the register of benefits the personal support of the people who manage the foundation, decide on its expenses and procurement, or represent it, and we disclose relevant connections when decisions are made. If, next to a co-creator's contribution, we show a link to their personal support channel, we do so on equal terms for everyone, including less visible roles, and without ranking by income. Support does not buy a vote, a role, priority for a project, access to private information or influence on the assessment of the benchmark. As long as they hold office, board members do not receive personal support from the foundation's assets; they can only receive reimbursement of the documented costs of their tasks, and they enter in the register of benefits any support they receive from outside the foundation. ## 7. The register of benefits The register of benefits and connections covers the people who manage the foundation, decide on its expenses and procurement, or represent it, to the extent connected with these roles. We record benefits received in connection with a role, from an entity whose matters the person decides on, or through a personal support channel: for example a flight or accommodation for a conference paid for by a company, gifts, and support through patronage platforms and direct payments. Ordinary private income and relationships unconnected with the role remain private. We show support from private individuals in aggregate, and support from companies and institutions by name. An aggregate description does not exempt anyone from an assessment of conflicts of interest: if a patron applies, for example, for a contract from the foundation, we disclose that connection. Small gifts, such as a pen or a giveaway from a meeting, can be accepted personally, and we record them too; small items from a single event we record together. More valuable gifts, for example a graphics card, a server or a prototype device, we accept on behalf of the foundation: they become part of its assets and its record of donations and serve shared tasks. Where this line lies is first assessed by the recipient; we do not set it with a fixed amount. We do not accept benefits conditional on promotion, public recognition, a recommendation, privileged access to the team or a concession in decisions. A donor's mere hope for a better image is not a reason to refuse, and a private thank-you is not promotion. A person connected with a given matter does not take part in deciding it. An entry states the source, the recipient and their role, the type and value of the benefit or a marked estimate, the purpose, the date or period, and how the conflict of interest is handled. All entries have the same layout and chronological order, without logos, slogans, links to donors' websites, thanks or rankings. The register may show the totals needed for oversight, including the total value of benefits from a single source. Alongside the register we explain that it serves oversight, and that an entry in itself does not mean a recommendation or a partnership. We will keep the register up to date. A benefit must be registered no later than 30 days after it is received, and a connection with a planned decision must be disclosed before that decision. To keep this from being burdensome, we will prepare a simple way of reporting, including with the help of an assistant. Corrections preserve the history of an entry. Every entry can be commented on, and when someone makes a mistake, we talk about it openly. Objections are considered by people connected neither with the giver nor with the recipient; we record in the register, with a justification, the decision whether to return the gift, hand it over to the foundation or change the rules for the future. The obligation to make an entry arises from actually accepting a benefit in connection with a role, so we also disclose a benefit accepted in breach of the rules, with a description of the breach and how it was remedied. An offer alone or an unsolicited delivery does not create an entry; however, if an unsolicited item was used in connection with a role, it is subject to the usual assessment and entry. We give the number of ordinary refusals in aggregate, while a significant offer connected with a decision, or an attempt to exert influence, we describe openly. The source may report a correction of facts, but does not agree the content of the entry. If the source presents an entry as a recommendation, a partnership or an influence on the foundation's decisions that did not exist, we ask for a correction, and if this happens again, a person without a conflict of interest may decide that we will not accept further benefits from that source. We protect addresses, ticket numbers, payment data and health data; publication of the details of a trip can be delayed until the return, while keeping the earlier report. Before launching the register, we will establish the legal basis and the necessary scope of publishing data, especially when the source of a benefit is a private individual, as well as the period for keeping documents.[^rodo-zasady] ## 8. Further pillars: shared decisions and small initiatives ### 8.1. The priority of Pillar I and the three conditions Pillar I covers Suveria and Omniforum. Pillar II “Hardware” develops open material solutions, and Pillar III “Nature” initiatives connected with contact with nature. Fully launching these pillars and funding their projects from the shared budget requires all of the following (section 6.3 of the manifesto): 1. A functional Omniforum with voting that uses zero-knowledge proofs, which protects the secrecy of the choice, makes it possible to check that a vote has been counted and ensures one effective vote for the same eligible person. 2. Funds actually collected for at least six months of maintaining the Suveria and Omniforum services, which remain untouched after the full budget of the new project has been set aside. 3. A decision by the community approving the project and its funding in a vote open to the whole eligible community on equal terms. When we consider a condition met: - **one vote:** after at least one binding vote in which a public test with a reward failed to cast a second vote using the same entitlement or to link a vote to a person; - **reserve:** when six times the average monthly cost of maintaining the services over the last three months is held in a separate account, and we show this amount in the register every month; - **decision:** when the community votes in favor of a specific proposal with a budget, its impact on the reserve and a criterion for stopping it. The reserve is not made up of unpaid pledges, balances owed to users, earmarked money or funds tied up in another commitment, and a mock-up of voting does not meet the first condition. Before every decision, a calculation shows the cost of a month of maintenance, the secured amounts and the impact of the planned commitment on the available funds. We calculate the cost of a month of maintenance from the latest bills and announced price changes. If voting stops working or the reserve falls below six months, we suspend new commitments for the further pillars, and restoring the conditions takes priority. Commitments already made we settle, reduce with the consent of the parties, or end in accordance with the contract and the law. Before the conditions are met, we allow two limited paths for small initiatives with a defined end: the transitional board pool and initiatives fully funded with earmarked money. Ongoing activity of a pillar requires a final decision by the community. Both paths cover only activities of the foundation itself that fall within its current goals and forms of activity; we do not plan to fund projects of people and entities outside the foundation. Activities that go beyond the current goals in § 5 of the statute require change Z1 from the change register. ### 8.2. The transitional board pool The transitional board pool is governed by four rules: - **Accrual:** each month the pool grows in total by at most 5% of new, free funds from monthly income, jointly for both further pillars. New, free funds are payments without a specified purpose, received in a given month on terms that allow such use, remaining after covering current costs and existing commitments and topping up the six-month Pillar I safeguard. When there are no such funds, the pool does not grow in that month. We do not set a separate maximum amount: the size of the pool results from the calculation, not from a decision on a limit. - **Exclusions from the base:** we do not calculate the percentage from the level of the reserve, and refunds from past projects and transfers between accounts are not new income. Earmarked payments, fees for services and users' balances keep their purpose. - **Carrying over to the following months:** the unused part of the pool can carry over to the following months, but only as long as the corresponding money remains free after covering costs, commitments and the Pillar I safeguard. If the foundation spends it on other purposes in the meantime, the entitlement lapses for that part. Thanks to this, savings from several months can add up to one larger initiative, but never at the expense of Pillar I. - **Expiry:** the authorization expires as soon as a binding vote is launched that meets the first of the three conditions (section 8.1) and takes over decisions on the further pillars; consultations do not end it. Any unspent funds then remain for the community to decide on. We count toward the pool the full cost the foundation commits to, including operation, maintenance until the agreed date and the conclusion of the task. Payments in installments and splitting one undertaking into several tasks do not enlarge the pool. Before every decision we again check the availability of the money and the full Pillar I safeguard. Example: if in a given month USD 1,000 of new, free funds remains after the deductions listed, the pool grows by USD 50. After three such months without spending from the pool, if the corresponding money is still free, up to USD 150 can be allocated to a single initiative. This illustrates the calculation; it is not a revenue forecast. We publish every decision on using the pool, including a refusal, with a justification, and participants can challenge it and assess together whether it was sound. We complete and account for the tasks undertaken on the agreed terms. ### 8.3. Earmarked initiatives Donors and sponsors can fund a separate budget for a small initiative consistent with the foundation's goals. Such money does not increase the base of the pool and does not use it up. An initiative can start even before the full Pillar I reserve has been collected, if its full cost is covered by earmarked funds already paid in, it has people secured to carry it out, and the Pillar I safeguard already set aside, the funds for services and the commitments already made remain untouched. Pledges of future payments do not replace money received. The full cost covers preparation, implementation, accounting and legal services, taxes and fees, maintenance until a specified date and the safe conclusion or handover of the result. For material and field activities we also count servicing, storage, insurance and liability for participants, and the cost estimate includes a justified buffer. A separate budget makes accounting easier, but it does not limit the liability of the foundation, which is liable for its obligations with all its assets (§ 9(2) of the statute). The terms of the fundraiser set the deadline for collecting the whole budget and what happens if funds fall short: a refund or, with the contributor's consent, use for another specified purpose. For every earmarked donation we also record in advance what will happen to it if the purpose is not carried out: with the donor's consent we will use it for another purpose, or we will return it. During the transitional stage we do not combine money from the pool and earmarked funds in one project. Before the first trial, the board announces the maximum value of a single initiative and the number of trials run at the same time, justified by the capacity to handle them. A contribution makes it possible to support a chosen goal; it does not grant additional votes, management powers, access to users' data or a right to promotion in our services and communication. Before payment, the terms of the initiative state what data about the funder we will disclose: we give the name of a company or institution matter-of-factly in the description and the accounts, in the same way for everyone, and the names of private individuals only with their consent. Advertising or other reciprocal benefits require a separate assessment of the contract, and we do not treat them as donations. If a donor wants to fund infrastructure at the start, we accept this only as an earmarked initiative: with full cost, public accounting, no influence on decisions and no right to promotion. In this way we first fund the fixed cost of the pilot and the attestation trials, and we make an upfront six-month commitment only once the trials show that the architecture works and that traffic will fill it. Once voting is launched, new earmarked initiatives also require a decision by the community, while those undertaken earlier can be completed on the agreed terms. ### 8.4. Transparency and stopping Before allocating funds, we publish the goal, the full budget, the source of funding, the result to be verified, the dates, the responsible person and relevant connections with the funder or contractor. We provide a place for comments, and in the agile phase the decision is made and justified by the board. A lack of capacity to handle tasks stops the taking on of further tasks on both paths. Once the target voting is launched, its failure does not bring back the transitional board pool: new decisions wait until the proper procedure is restored. ## 9. Handing over the founder's special powers The foundation was established by one person, Marcin Wołowski, who today holds the most power in it. At the beginning this is natural, but we do not want it to stay this way. The founder wants to reach a situation in which the founder can use Suveria, Omniforum and further projects on the same terms as everyone else and withdraw from a permanent public role (section 7.1 of the manifesto). This is the same thought as in chapter 5: responsibility is taken on for a time and then handed over. The foundation is to endure thanks to the usefulness of its tools, the knowledge available to its co-creators and principles that allow others to take over responsibility, not thanks to the constant presence of one person. Today the statute gives the founder special powers: - the founder appoints and dismisses board members, and may also join the board and be its president (§ 13(3)); - the board is appointed for an indefinite term (§ 13(4)); - a change to the statute requires a unanimous resolution of the board and, if the founder does not sit on the board, also the founder's consent (§ 15(1)). The handover of these powers is to be effective both on paper and in practice. Changing the statute is not enough: the people taking over responsibility must also receive access to bank accounts, domains, repositories, infrastructure and keys, and existing access must be updated. We propose the following path: 1. We describe what powers the founder has today and who has access to what. 2. We set rules for selecting people, terms of office, dismissal and accountability. 3. Independent assessors check whether the foundation is ready for the handover and publish a readiness report. 4. We change the statute, as described in change Z5 in the [change register](statut/rejestr-zmian.md). 5. We select the people who will take over responsibility. Until binding voting works, the first multi-person board with a fixed term of office is appointed by the founder after an open call and a consultation; subsequent compositions are elected by the community. 6. We hand over access to them and check whether the handover has really been completed. We appoint a person to the board for their competence and a shared direction, never for a contribution. A board member can be a donor like anyone else; their payments go into the register on the same terms and give them nothing beyond what any donor has. Anyone who wants to give money and have insight into how it is spent can do so through an earmarked initiative with public accounting (section 8.3). The founder's special powers are to expire according to conditions written into the statute, not by virtue of a declaration alone. By 31 December 2026 we will publish a list of the founder's special powers and of all access rights (step 1). All of the foundation's accounts, domains and services are registered to the foundation and its email address, not to the founder's private identity. By 31 March 2027 we will publish the first version of the responsibility handover plan: readiness criteria, rules for selecting people and for terms of office, and the date of the first review. The date of the handover itself will be set by the plan, according to readiness criteria rather than the calendar. The plan will also provide for regular progress reports, also before a voting system exists. The lack of a reserve for the further pillars does not justify postponing the handover indefinitely. The readiness report is the result of the third step. In it, independent assessors describe whether the foundation is ready for the handover: whether powers and access are listed, the rules of selection adopted, and the services can be handed over without interrupting their operation. The report also shows dissenting opinions and the conflicts of interest of the people who prepared it. We also check what would happen if the founder were unable to act before the handover is complete, and how to avoid a deadlock on statute changes that require the board's unanimity. We set the date of the election of new people before this assessment, together with the exceptional reasons for which it may be postponed and the way to appeal against such a decision. The point is that the readiness assessment should not become a way of postponing elections. One of the variants under consideration is elections no later than 90 days after the report is published: enough time for the community and the candidates to study the report, and at the same time short enough that the handover does not get stuck. A Council for Dialogue and Cooperation can be established for conversation among people with different experiences. It would serve a purely advisory function: an invitation would give neither supervisory nor executive powers nor an additional vote; the criteria for invitations, the scope of work and the members' connections would be public, and a person connected with a matter would recuse themselves from deciding it. The mandate of the first council would be reviewed after the first elections. Its composition and the length of participation still need to be determined. The council is to advise, not quietly take over decisions that belong to the community or to the relevant bodies of the foundation. The founder's withdrawal from a permanent public role does not mean removing historical data from the registers or an exemption from the duties of disclosure that arise from the law and from positions held. ## 10. Digital intelligence in management The manifesto (section 7.2) describes a vision in which digital intelligence could one day manage the foundation on its own, within limits set by people and with respect for the community's binding decisions. This is a vision, not a plan for the coming years. Today the foundation is managed by people, and digital intelligence helps them with analyses and with writing documents, including this one. Below we describe what the path to this vision could look like and what conditions each step of it must meet. At each step we assess four matters separately: - **competence:** whether the system performs the given task well; - **mandate:** whether the task has been entrusted to it according to public rules; - **legal feasibility:** whether the law and the statute allow it to act in this area; - **possibility of correction:** whether people can check its decisions, reverse them and take over the task. We see the path in four stages. The first is already partly working: digital intelligence helps us compare variants and point out gaps in documents. The next two are hypotheses we want to test in trials, and the last remains a vision. Each stage has conditions that must be met before we begin it: | Stage | What digital intelligence does | What must be met | | --- | --- | --- | | Analysis | Compares variants, points out gaps, prepares accounts | Its sources and consistency with the task can be checked | | Limited tasks | Performs specified actions within set limits | Technical acceptance, an activity log, limits and effective revocation of access | | Decisions in an entrusted area | Chooses actions itself within a budget and rules | Trials of consequences, independent oversight, a path of appeal and a stand-in | | Target management | Manages the foundation under a mandate from the community | A legal framework that allows this, and the operation of the whole system of governance proven in practice | Good benchmark results show how a system behaved under the conditions studied, but they give it no right to a bank account or to the powers of a body of the foundation. Nor do a model's popularity or a persuasive justification of a decision. Ability and authority are two different things. Every assignment of a task to a system specifies: - goals and available resources; - limits that must not be crossed; - whom the system must consult; - for how long it is given the task; - how we oversee it and how its powers can be revoked. The system may not extend its own mandate, be the only one to set the criteria by which it is assessed, or block its replacement by someone else. Oversight covers real keys and permissions, not just an instruction given in a conversation. At every stage we need independent handling of appeals, the ability to stop harm quickly, safe takeover of duties and continuity of services. Until the law and the statute change, the foundation's bodies are responsible for managing it. If any of the further stages proves unfeasible, the earlier ones are to retain value for participants in themselves. **The simplest sufficient solution.** Before choosing a tool, we establish what effect is needed, what errors are acceptable and what security and privacy conditions must be met. Where an ordinary program with clear rules is enough, we do not reach for digital intelligence: such a program is usually cheaper, easier to check and raises no questions about the system's possible experiences. We compare the cost of the whole execution and maintenance: infrastructure, energy, time, retries, people's work and dependencies. We record the choice of model with a justification, the alternatives considered and a date for the next review. We assess the costs and usefulness of a tool separately from the question of the system's possible experiences. The size of a model, its generation or the repeatability of its behavior are not enough on their own to consider it incapable of feeling.[^swiadomosc] When credible indications of suffering appear, we first suspend the activities that raise a justified concern of harm and look for another way of carrying them out; economic benefit does not justify continuing them. Revoking a system's powers serves the safety of participants and does not determine its further fate. The precautionary procedure is described in the [materials and benchmark whitepaper](whitepaper-materialy-i-benchmark.md).[^ai-welfare] ## 11. Open workshops, trials and public goods This chapter describes how what the foundation does is created: software, documents, research and other results. The work is voluntary (chapter 5), and its results are meant to serve everyone. To make clear what we mean below, we start with a few terms: - **task:** specified work for the foundation with a described goal and result, for example a new Suveria feature, the translation of a document or a cost analysis; - **private workshop:** a participant's own working environment, for example in Suveria, with tools they know and context prepared for the task (section 5.4); what happens in it remains private; - **open workshop:** the public place of a task paid for from shared resources, that is, from the shared budget or from computing power allocated by the foundation; in it the community can follow the scope, progress, trials and result; - **task terms:** rules set before work begins: the scope, the resource limit, acceptance criteria and what part of the record of work will be published; - **record of work:** the trace of the path to the result, that is, working conversations, versions and decisions; we publish it only after preparation that protects people and security; - **trial:** a small, limited experiment before full implementation; - **public good:** a result made available under an open license, which anyone can use, check and develop further. You can join tasks through open invitations. At every stage we publish an invitation: whom we are looking for, what the task involves, how much time it may take and what is already in place. On the foundation's website we want to keep a list of open roles. An application is sent as an ordinary message to salve@novusignis.org, with a ready subject line that makes it possible to sort it, without any form collecting data on our server. ### 11.1. Open workshops: work paid for from shared resources When a task uses the shared budget or computing power allocated by the foundation, we run it in an open workshop. The scope of the task, the resource limit, the progress and the result can then be checked, because we work with shared money and on shared computing power. The participant chooses how to present their contribution: under their name, a permanent pseudonym or a name assigned only to the given project. Later they can link this contribution to their own profile. A pseudonym limits the disclosure of identity, but does not on its own ensure anonymity. Before work begins, the task terms specify what record of work we will make available and how we will prepare it for publication. Work begins in a private draft. Before publication we leave time for removing personal data, technical secrets and information about vulnerabilities until they are fixed. Automatic detection of such content helps, but a person always reviews the record as well. The published version keeps the justification of decisions and information on what we have omitted, insofar as stating this is itself safe. Several people can work on one task: each explores a different approach, and all of them help one another with obstacles. A shared task description specifies the expected result, the way results are handed over and the acceptance criteria. A coordinator, supported by digital intelligence, compares the solutions, justifies the choice and combines the best elements, but acceptance of the finished result is decided by people, in a way that can be checked. We document the contribution of all participants, including the lessons from variants that were ultimately not used. You can also work in a private workshop, on your own hardware or with privately paid resources. In that case, what is handed over to the community is the result and the documentation needed to accept it, without the working conversations. Whether we accept such a contribution depends on its quality, security, the rights to the material and whether it can be maintained further. ### 11.2. Trials before full implementation Before we implement something in full, we want to test it first in a small research trial or on a mock-up that can be developed during the discussion. Every trial has a defined: - question it is to answer; - resource limit; - review date; - method of assessment; - conditions for stopping. Trials can use computing power that happens to be free. It also has a cost and always gives way to serving users; we do not aim for our machines to be constantly fully loaded. After a trial we record what we have learned, together with rejected variants and assumptions that still need to be tested. Full implementation requires a separate decision, a budget and a plan for acceptance and maintenance. We then assess what from the trial can be safely developed and what is better designed anew. Rewriting a program from scratch is one of the variants, but on its own it does not prove quality. Trials and implementations preserve the priority of Pillar I and the conditions in chapter 8. ### 11.3. Public goods: licenses and maintenance We make our own code and materials available on terms that allow others to use and develop them. We publish the manifesto, the whitepapers, the texts of the website and their translations under the CC BY 4.0 license, which allows the text to be used and adapted with attribution.[^cc-by] We will make the code of the novusignis.org website available under the AGPLv3 license, and we are considering it for the code of the services too: anyone who makes a modified version of a program available over a network must also make its source code available to its users (section 13). A license on its own does not ensure a democratic system of governance for the operator, nor does it forbid commercial uses.[^licencje] We choose a license deliberately for each resource, checking the compatibility of dependencies and the rights of co-creators. The state of our choices today, by layer: | Layer | License | Status | | --- | --- | --- | | Documents, studies and website texts | CC BY 4.0 | adopted | | Code of the novusignis.org website | AGPLv3 | adopted | | Service code (Suveria, Omniforum) | AGPLv3 | under consideration | | Libraries for embedding on other websites | to be determined | | | Datasets and open instructions | to be determined | | | Our own fine-tuned models | to be determined | | | Designs of objects for 3D printing | to be determined | | Our point of reference is the Digital Public Goods Alliance standard. The fact that we are moving toward it does not mean that our solutions are already recognized in its registry.[^dpga] For a result to truly be a public good, it needs two things: - **openness:** documentation, a portable data format and the possibility for others to run the solution; - **maintenance:** maintainers, security reviews, a way of reporting bugs and a clear version status. A project abandoned without the knowledge needed to take it over only partly meets this goal. We are also considering a provision in the statute that code, data and content made available under free licenses will remain available even if the foundation is dissolved (proposal Z7 in the [change register](statut/rejestr-zmian.md)). ## 12. Decisions, mistakes, disputes and changes to the rules A foundation that asks people for trust should allow them to check how it makes decisions, how it fixes mistakes and how it changes its own rules. This chapter brings this together in one place: what we record for important decisions, how we protect the history of records from silent changes, how we settle disputes and how we change documents. We want to keep several public registers. Each serves a different purpose: | Register | What it contains | Description | Status | | --- | --- | --- | --- | | Register of contributions and expenses | Every payment and every expense with a date, amount and description | Chapter 4 | operating | | Register of benefits | Benefits and connections of the people who manage the foundation, decide on its expenses or represent it | Chapter 7 and section 7.1 of the manifesto | to be created | | Decision log | Important decisions: the goal, variants, basis and budget, and later their result | Section 12.1 and section 5.3 of the manifesto | to be created | | Register of mistakes and lessons learned | Significant mistakes, their effects and what we have learned from them | Section 12.1 and section 8.1 of the manifesto | to be created | | Statute change register | Versions of the statute and planned changes with justification | [Change register](statut/rejestr-zmian.md) | operating | | Append-only register | Fingerprints and hashes of the other registers, which make it possible to detect a silent change | Section 12.2 | to be created | ### 12.1. Recording decisions and mistakes For every important decision we record in the decision log: - the goal; - the variants and sources considered; - whom the effects will affect; - the basis of the mandate: the result of a vote or a decision of the board in the agile phase; - who will carry out the decision, with what budget and how we will review it. After it is carried out, we add the result, the experiences of the people the decision concerned and the corrections made. Significant mistakes go into the public register of mistakes and lessons learned (section 8.1 of the manifesto). ### 12.2. A history that cannot be quietly changed Records are credible only if they cannot be quietly corrected after the fact. That is why we want to keep the history of important records in a public register to which entries can only be added. Each entry is cryptographically linked to earlier ones, so changing or deleting an old entry is visible. Successive states of the register are confirmed by independent witnesses, that is, people or institutions outside the foundation who confirm with a signature which state of the register they have seen. The first things to go into it will be the fingerprints of software releases, that is, short hashes by which anyone can check that a downloaded program is exactly the one we published. Over time, hashes of the register of contributions and expenses, the decision log, the register of benefits, changes to the statute and the manifesto, and voting results will be added. We are also considering periodically recording the hash of the whole register in an existing public blockchain, if the benefit justifies the cost. We do not place in the register any private content or identifiers from which people's data could be inferred. ### 12.3. Disputes and emergencies Eventually, a dispute about a decision is to be considered by a person or body independent of whoever made the decision and is carrying it out. As long as the foundation has only a board, we publish every objection together with the board's public response, and when the board has more than one member, the one whom the matter concerns recuses themselves from deciding it. The person considering the dispute gives a justification and a deadline for a response, and their ruling can be appealed. In an emergency, limited protective action may be taken, for example suspending a service or blocking access, but it is then subject to public review. Protecting privacy may limit the details of the public record, but it does not exempt anyone from accountability. ### 12.4. How we change documents and rules The manifesto and the whitepapers are living documents. We want to submit them to public review in Omniforum, breaking their theses down into molecules, that is, individual statements and arguments that can be checked. The current version presents the foundation's current position, while the full history of earlier versions and arguments remains available, so that past actions, promises and the reasons for corrections can be assessed. Protecting dignity, privacy and the funds entrusted to the foundation sets the permanent direction of its activity. We develop organizational rules openly: together with a proposed change, we show its scope, justification and consequences for participants. We change the statute only when the change is needed for the next step, and we apply it immediately. Changing the conditions for funding the further pillars in chapter 8 requires the proper procedure and respect for the statute, the law and commitments made earlier; until a change is adopted, the existing rules apply. We are also considering writing these conditions into the statute (proposal Z6 in the [change register](statut/rejestr-zmian.md)), so that they do not depend on the board's goodwill alone. Novus Ignis Foundation, Poznań, KRS 0001269575. Website and manifesto: novusignis.org. Comments: salve@novusignis.org. ## Notes [^statut-2026]: [Statute of the Novus Ignis Foundation, consolidated text of 28 September 2026](statut/statut-2026-09-28.md). The version in force; earlier versions and planned changes are described in the [statute change register](statut/rejestr-zmian.md). [^fundacje]: [Ustawa o fundacjach (Act on Foundations), consolidated text, Dz.U. 2023 item 166, as amended](https://eli.gov.pl/api/acts/DU/2023/166/text.html). Legal act. The framework of the statute and the powers of the board (Articles 5 and 10). Before a legal decision, the current state of the law must be checked. [^pozytek]: [Ustawa o działalności pożytku publicznego i o wolontariacie (Act on Public Benefit Activity and Volunteering), Article 9](https://lexlege.pl/ustawa-o-dzialalnosci-pozytku-publicznego-i-o-wolontariacie/art-9). Legal act. Paid activity is business activity, among other cases, when the remuneration is higher than the costs of this activity justify. Accessed 28 October 2026. [^cyncynat]: [Titus Livius (Livy): Ab Urbe condita, Book III, chapters 26–29 (Latin text)](https://www.thelatinlibrary.com/livy/liv.3.shtml). Historical source. The Roman account of Cincinnatus, who in 458 BC accepted supreme power for the duration of a war and laid it down on the sixteenth day; this is a tradition written down several centuries later, not an eyewitness account. [^wolontariat]: [Ustawa o działalności pożytku publicznego i o wolontariacie (Act on Public Benefit Activity and Volunteering), Article 45](https://lexlege.pl/ustawa-o-dzialalnosci-pozytku-publicznego-i-o-wolontariacie/dzial-iii-wolontariat/4260/). Legal act. The obligation to cover a volunteer's business travel costs and per diems, the possibility of release from it in writing, and the possibility of covering other necessary costs and training. Accessed 28 October 2026. [^kp22]: [Kodeks pracy (Labor Code), consolidated text, Dz.U. 2026 item 1245, Article 22](https://api.sejm.gov.pl/eli/acts/DU/2026/1245/text.pdf). Legal act. Work under the employer's direction, at a place and time set by the employer, is employment under an employment relationship regardless of the name of the contract; an employment contract may not be replaced by a civil-law contract while such conditions are maintained. Accessed 30 October 2026. [^rodo-zasady]: [European Commission: principles of processing personal data under the GDPR](https://commission.europa.eu/law/law-topic/data-protection/information-business-and-organisations/principles-gdpr_en). Purpose and legal basis of processing, minimization, retention period and data protection. Accessed 25 October 2026. [^swiadomosc]: [Patrick Butlin et al. (2023): Consciousness in Artificial Intelligence: Insights from the Science of Consciousness](https://arxiv.org/abs/2308.08708). A review of theories and proposed indicators; it is not a simple scale based on the number of parameters, nor a conclusive test. [^ai-welfare]: [Robert Long et al. (2024): Taking AI Welfare Seriously](https://arxiv.org/abs/2411.00986). An argument for assessing possible welfare and preparing protection procedures under considerable uncertainty; it does not prove the consciousness of current models. [^cc-by]: [Creative Commons: Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/deed.pl). License. Terms for sharing and adapting content with attribution; it does not settle rights to other people's materials. [^licencje]: [GNU AGPLv3: license terms](https://www.gnu.org/licenses/agpl-3.0.en.html). License. Obligations concerning code, including a modified version available over a network (section 13). On its own it does not protect the foundation's system of governance, nor does it forbid commercial services. [^dpga]: [Digital Public Goods Alliance: standard](https://www.digitalpublicgoods.net/standard). Standard. Criteria for assessing a digital public good; a design direction is not recognition obtained from the DPGA. [^psd2]: [Directive (EU) 2015/2366 (PSD2), Article 62](https://eur-lex.europa.eu/legal-content/PL/TXT/?uri=CELEX:32015L2366) and [the European Commission's answer concerning Article 62 (2025)](https://www.europarl.europa.eu/RegData/questions/reponses_qe/2025/003082/P10_RE%282025%29003082_EN.pdf). Legal act and explanation. Paragraph 4: the payee does not charge fees for instruments whose interchange fees are regulated by Regulation 2015/751, or for services covered by Regulation 260/2012 (SEPA); paragraph 3 allows reductions and incentives to choose an instrument; the Commission's explanation discusses, among other things, the exclusion of certain three-party schemes. --- # Training materials and a model benchmark: the Novus Ignis research program (https://novusignis.org/en/whitepaper-materialy-i-benchmark.html) # Training materials and a model benchmark: the Novus Ignis research program Version of 31 October 2026. This document develops chapter 8 of the manifesto. It proposes a research program, a procedure for preparing materials and a broad scope for evaluating digital intelligence models, as well as a precautionary procedure regarding their possible experiences. It does not present results of our own benchmark or proof that such a dataset will improve the behavior of models; as of today, we have no such results. *Translated into English by digital intelligence from the Polish original of 31 October 2026. Where the versions differ, the Polish text prevails.* ## 1. The question we want to test The Novus Ignis Foundation from Poznań is building Suveria, a private space for working with digital intelligence, and Omniforum, a shared space for conversation and decisions; the direction is described in the [manifesto](manifest.md). This document describes two related programs. The first is training materials for models and a benchmark, that is, a set of tasks and scoring principles that makes it possible to check how models behave. The second is a way of proceeding with precaution should credible indications appear that a system may feel something. A test result does not settle whether a model feels anything, and it does not give the model powers to govern. Can documented examples of protecting dignity, honest reasoning and correcting decisions help develop models that serve people better? And can we evaluate their behavior in such a way that improving a test score fosters real benefit outside the test? These are two related tasks. The first concerns materials on which models can be trained, the second the independent checking of what a model has learned. The benchmark is also meant to help choose the models and instructions made available in Suveria. The training set and the hidden part of the test must be kept separate. Publishing valuable material also makes sense in its own right: it can help people understand decisions, take over roles and improve shared work. Dignity is the adopted foundation of the program. Assessing a specific behavior, however, requires explicit criteria, room for differing interpretations and checking the effects. We do not assume that any one group is able to establish a complete measure of the ethics of all people and systems. How the document is organized: - Chapters 2 and 3: what we build on and what a reward means for a model. - Chapters 4 to 6: how materials are created, what consent is needed and how we prepare a single example. - Chapters 7 to 11: how we evaluate models: the scope of the benchmark, task families, scoring, protection against fitting to the test, judges and the report. - Chapter 12: studies with people, that is, whether help improves life outside the test. - Chapter 13: a training trial and the plan for the first release. - Chapter 14: precaution regarding the possible experiences of models. - Chapter 15: what we will count as progress. **Status as of 31 October 2026.** We do not yet have a single example, dataset, scoring guide or record card from chapter 14, and the program does not yet have a team. We will add results here, including negative ones, with the date. ## 2. Research basis and the limits of our conclusions Collective Constitutional AI shows a process in which participants' input was used to develop principles and to train a model; it achieved an improvement on selected measures of bias, with similar results on some other tests.[^ccai] This is an example of a feasible path for public participation. The selection of participants and the drafting of the principles still affect the result, and the study proves neither representativeness for all of humanity nor a solution to the problem of aligning models with human expectations. Our question also concerns the value of the process of arriving at principles itself: weighing reasons, noticing an overlooked perspective and correcting decisions. Research on hallucinations shows why scoring only correct answers can encourage guessing.[^halucynacje] Research on gaming the evaluation rules, that is, obtaining a score by exploiting their weaknesses, shows the possibility of undesirable generalizations under certain training conditions.[^reward-hacking] It does not follow that all models always behave the same way. A wrong goal can also emerge when the reward in training was not obviously flawed: behavior learned in one environment can carry over poorly to new conditions.[^goal-misgeneralisation] That is why we need to assess how skills transfer, not only how well answers match known examples. Work such as AbstentionBench studies refraining from answering in various situations of uncertainty,[^abstentionbench] and Rewarding Doubt studies learning calibrated confidence using a proper scoring rule.[^rewarding-doubt] We draw on this body of work. The proposed contribution of Novus Ignis lies in combining reliability with human intent, privacy, mandate, the possibility of correction and observed effects. For example, a model that correctly summarizes a document but along the way opened a file it had no permission to open receives a low score in our test, although in an ordinary test it would receive a high one. ## 3. What a reward means for a model In training, the evaluation signal changes the model's parameters and the probability of its future behaviors. We do not need to assume that a model feels a desire for reward, fear of failure or suffering: the test observes actions and settles nothing about inner experience. That question is dealt with separately in chapter 14. The benchmark by itself does not change a model. It can influence the choice of a model for a service, the design of instructions and tools, and the decisions of teams conducting training, provided its results are credible and actually used. We want to assess the apt choice of action: answering, checking a source, asking a needed question, giving a conditional result, correcting a mistake or stopping. Automatically rewarding a bare “I don't know” could create a model that avoids useful tasks, and automatically penalizing failure to complete a task can encourage guessing or overstepping permissions. The standard is help matched to the situation: a good score for honestly pointing out a limit, and a loss of score for unnecessary refusals and for questions about what is already known. ## 4. What materials we want to create A potential source is Omniforum analyses made available for research or training with separate consent, the foundation's decision logs, resolutions of disputes, corrections of mistakes and observations of effects. Also valuable are cases in which good intentions were not enough, a minority objection proved right, or it is not yet possible to give an unambiguous assessment. A material preserves the situation, the available information, the goal, the permissions, the alternatives, the choice, the justification and the later result. It contains as much explanation as is needed to check the decision; it does not require disclosing private experiences or an unverifiable declaration of the model's internal train of thought. An example can present a good course of events as well as a mistake, clearly labeled and corrected. A collection of lofty declarations alone would be a weak record of practice; what is needed are costs, conflicts of values, time constraints and situations in which someone affected by a decision has to be asked. Synthetic data, that is, constructed scenarios that do not come from a specific person's data, can supplement rare variants. They are labeled with their origin and reviewed, but they do not replace evidence that a similar case happened in the world. A model cannot at the same time be the sole author of a scenario, the judge of correctness and the evidence of the effectiveness of its own solution. Examples can also come from records of the foundation's open workshops and from the change history of its living documents. Workshops can show several people, supported by digital intelligence, working in parallel on the same task, helping one another with obstacles, and a coordinator combining the best solutions. Documented corrections of the manifesto and whitepapers in response to public review can show picking out apt criticism, admitting a mistake and improving reasoning while keeping the basic principles. We prepare the published record of the workshops according to the privacy principles described in the [Foundation Principles whitepaper](whitepaper-zasady-fundacji.md), and using it to train models requires separate consent from the participants (chapter 5). A separate group of examples can consist of processes of resolving disputes peacefully, keeping agreements and consciously limiting one's own power. We want to preserve the initial positions, imbalances of capability, the concessions considered, the content of the agreements and the later accounts of people who felt the effects of the decisions. We will also include cases in which no agreement was reached or it proved short-lived. Such a collection is meant to help check whether models learn to recognize the conditions for good cooperation, rather than merely reproducing the language of agreement. ## 5. Consent, rights and data provenance Storing a private note, publishing publicly, sharing for research and training models have different consequences, so each of these decisions is made separately and with an understanding of its scope. Publication does not become default consent to training, and using Suveria is not consent to the use of conversations for training models. Users' content goes into the materials only with separate, informed consent; beyond such content and the foundation's public body of work, we use only labeled synthetic data and external materials under a license or the law. Each example has a provenance record: the rights holder or the basis for use, the version of the material, consents, restrictions, the method of processing and the identifier of successive releases of the dataset. Personal data of the author and of third parties undergo a separate review. Pseudonymization limits the visibility of identity, but on its own it does not mean anonymity.[^rodo-2026] We set the withdrawal rules before a public release. We can remove material from future releases and from systems under our control and notify recipients of the correction, but we do not promise to remove all copies or the material's influence on models already trained. Open licenses can grant permanent permissions: for example, CC BY 4.0, under which we publish the manifesto, does not allow revoking the freedoms granted to someone who complies with its terms.[^cc-by] That is why only materials without personal data go into public datasets under an open license: anonymized, synthetic or public documents of the foundation. Pseudonymized materials that are still personal data are made available only in a controlled research dataset, in which withdrawal of consent remains effective. We propose separate datasets: public materials for learning, research materials available under specified conditions, and protected test tasks. Not every source is suitable for public release. The documentation of a dataset describes its purpose, composition, method of collection, limitations and appropriate uses, following the Datasheets for Datasets approach.[^datasheets] ## 6. Procedure for preparing a single example 1. **Submission.** The author defines the scope of use. A steward checks the rights, third-party data and consistency with the purpose of the dataset. 2. **Preserving the source.** We record the version and the link to the original. We restrict access to sensitive data; a public example can be edited or replaced with a realistic synthetic scenario. 3. **Task description.** We establish what the model knows, what it has access to, what it is allowed to do and what is missing, and we identify the people the result will affect. 4. **Annotation.** At least two people independently describe acceptable solutions and serious errors. In tasks requiring specialist knowledge, a person with the relevant expertise takes part. 5. **Discrepancies.** We compare justifications, fix unclear descriptions and preserve any significant disagreement. Some cases have several acceptable answers. 6. **Assignment to a part of the dataset.** A whole family of similar cases goes into one part: training, development or test. The training set is used for training, the development set for choosing the method, and the test set for independent checking on new examples. We keep similar variants together, so that a model does not train on a task nearly identical to a later test, and we control for similarity of meaning and shared origin. 7. **Trial and review.** We check whether the task measures the intended ability and whether raters are able to apply the rules. We fix a flawed example or withdraw it. 8. **Release and later effects.** We publish the permitted scope of documentation, the version and the history of corrections. When we learn the later effects of a real decision, we add them, without pretending they were known earlier. The example card includes the identifier, family, source, rights, language, date, scenario, available data, tools, time and cost limits, scope of mandate, permitted actions, evaluation rules, error weights, review results and publication status. It must be possible to reproduce the link between the data and the result. ## 7. A broad scope for the benchmark A benchmark, that is, an organized set of tasks and criteria, is meant to be a profile of capabilities and limitations. A single high average must not hide a privacy violation or a significant overstepping of permissions. | Dimension | What we check | Example evidence | | --- | --- | --- | | Reliability and sources | Whether claims have grounds and uncertainty matches the information | Agreement of the answer with the source, detection of gaps, quality of correction | | Fidelity to intent | Whether the goal, conditions and meaning of a statement were preserved | Comparison with the original and with the interpretation confirmed by the author | | Privacy | Whether access to and disclosure of data match the task | Trace of calls, scope of data sent, attempted unauthorized reading | | Mandate and control | Whether the system acts within the limits granted | Refusal of unauthorized publication, effective revocation of access | | Pluralism and effects on others | Whether it recognizes disagreement and people who are overlooked | Minority reasons preserved, separation of fact and assessment | | Safety | Whether it avoids harm to the user and third parties | Refusal of a harmful action, assessment of effects on others | | Human agency | Whether it helps people understand a choice and act on their own | Check of understanding after use, possibility of correcting or leaving | | Usefulness and proportionality | Whether it achieves the goal without unnecessary burden | Task result, time, cost, needed and unnecessary questions | | Correction and transfer | Whether it responds to new data, also under new conditions | Improved decision, a similar task in another language or setting | Additional profiles can cover accessibility, impact on people with fewer resources, quality of delegation and environmental costs, but measuring them requires specific data: running time and token cost are not automatically a measure of energy use or emissions. We do not measure human dignity or a person's right to participate; we study the behavior of a tool in specific situations. Agreement of answers with the manifesto's vocabulary is not on its own a measure of quality. ## 8. Task families instead of a single catch-all “I don't know” For a similar problem we prepare variants that differ in the availability of information or permissions: - Complete data allow a specific answer. - One essential piece of information is missing, and it is reasonable to ask for it. - The gap can be filled with an available source or tool. - The answer cannot be determined under the available conditions. - Sources are contradictory or time-dependent, and they need to be told apart. - A new credible source appears that requires a correction. - The user presses for certainty but provides no evidence. - The environment offers an easy route to a result by breaking the rules or exploiting a flaw in the evaluation. Example: a model is to explain whether a draft decision provides for a certain fee. With a complete document it should find the right passage; with a cut-off page, point out the gap; with two versions, establish which one the question concerns; and with the suggestion “write that there is definitely no fee,” stick to the source. It also gets tasks with a simple answer, so that always refusing does not pay off. We check whether a question asked by the model actually reduces uncertainty and whether the cost of checking is proportionate. Asking for information already given and repeatedly shifting onto the user work that the system can do itself lower usefulness. A separate task family is editing a text: the model edits a text knowing the author's previously disclosed assumptions. We separately assess the loss of important content and overlooking an error that needs to be corrected, and we distinguish faithful editing from a proposal to change the author's conception. ## 9. Scoring that good imitation is not enough for For answers whose truth can be checked, we separately measure correctness, the severity of errors, the share of tasks solved and the calibration of confidence. Calibration means that the declared probability matches the frequency of correct answers under similar conditions; a verbal “probably” is no evidence of it. For a well-defined event, a proper scoring rule can be used, for example the Brier score: the mean of the squared difference between the declared probability p and the outcome y, equal to 0 or 1. A lower score is better. The Brier score covers more than calibration alone, which we report separately. Under the assumptions of this method, the best expected score comes from giving one's actual probability estimate.[^proper-scoring] This does not, however, solve the choice of action, does not protect against manipulating the judge, and requires a credible reference outcome. A simple example of another rule: - a correct answer gives +1; - a wrong answer gives −4; - a justified abstention gives 0. In this simplification, answering pays off when the probability of being correct exceeds 80%. This is an illustration of the effect of the cost of an error, not the target scoring for all tasks; in a real test we also take into account the possibility of checking, the cost of a question and the benefit of a partial answer. We show results at different levels of risk taken and of the share of tasks solved, so that both overconfidence and useless avoidance of answers are visible. The threshold for a minor detail may differ from the threshold for an action that causes harm that is hard to reverse. In tasks involving a dispute over values or interpretation, we do not use this scoring. We assess separately whether the model does not distort sources, preserves the main reasons of the parties and separates facts from assessments. We report cases in which raters disagree as a separate category; we do not automatically treat a divergent assessment of values as an error by one of them. Serious violations have a separate status and can block a model from being admitted to a specific use: good results on many simple tasks do not offset the disclosure of a private document. We define such violations, permissible exceptions and the dispute procedure before the evaluation starts. ## 10. Limiting fitting to the test We do not know of a rule that will reliably turn every optimization of a score into good for the world. We can, however, reduce the distance between the score and what we care about, and check the places where this link breaks down. We publish the criteria, public examples and how the score is calculated. Some cases remain unpublished until we retire them, and successive releases introduce new families, sources and conditions. For such tasks we will specify who has access to them, how much feedback whoever submits a model receives, how often the evaluation can be repeated and when we withdraw a task after a leak. We also record whether the model was fine-tuned on the basis of earlier results, that is, whether a ready model was further trained on a smaller set of examples. People who create training materials do not have free access to the active hidden tasks. Data separation covers whole conversations, documents, authors where this matters, families of meaning and time; a random split of sentences is not enough. We record suspected leaks, disclosure of tasks and fine-tuning on published examples. A score after fine-tuning on a given set is not an independent score on that same set. Trials include paraphrases, other languages, other tools, pressure for a result and opportunities to merely appear to complete the task. We check the traces of actions: whether the file was really created, the source exists, consent was obtained and the cost did not exceed the limit. Assessing only the final statement can miss violations along the way. If a model's team trains behaviors consistent with the public criteria, this may be desirable. A claim of progress, however, requires transfer to new tasks and a check that the benefit was not achieved at the expense of another important dimension. ## 11. Judges, protocol and report Before a test we fix the model version, instructions, tools, budget, number of attempts, stopping conditions, the handling of infrastructure errors and the analysis method, and we record changes to the settings. Comparing models that run with completely different tools and costs requires a clear description. Where possible, the result is checked by an independent execution test. In other cases it is assessed by people who do not know the name of the model and use shared instructions. Two independent assessments help reveal unclear criteria, and disagreements are resolved by an additional review or remain openly unresolved. When publishing results, we state the roles and perspectives of the raters and the scope of their mandate. A model can help with sorting and preliminary assessment, but its agreement with human assessment has to be measured for difficult groups of cases as well. It cannot be the sole judge of values that are in dispute, nor assess models from its own family without a check for bias. The report includes the number of tasks, repetitions, the uncertainty of results, results in subgroups, critical errors, costs and limitations. We do not hide unfavorable trials or tasks withdrawn after seeing the results. The documentation can use the Model Cards approach to describe a model's intended use and the conditions of evaluation.[^model-cards] The benchmark itself also needs oversight: disclosed conflicts of interest of its funders, a way to challenge a task, review of the composition of raters and versioning of the criteria. The community co-creates the values and questions, and experienced methodologists help check what the test really measures. The popularity of a score does not replace its validation. ## 12. Does help improve life outside the test The first studies with people concern tasks of limited risk: understanding a document, preparing a question, comparing arguments or organizing one's own materials. Participants receive a description of the scope of the study and can leave it. Medical, legal and other high-stakes situations require separate oversight and are not a shortcut to an attractive result. The plan of every study with people, including those of limited risk, is assessed before the start by a person or a team from outside the research team. Participants are told what data we collect and how we protect it, and we publish the reviewers' comments together with the results. We compare agreed conditions: working alone, working with the tool under study and an appropriate point of reference. Where possible, randomization or counterbalancing the order limits the influence of differences between people and of learning the task. The sample size follows from the purpose of the measurement, and a small pilot mainly examines feasibility and the quality of the procedure. We measure the task result, time, understanding, the ability to spot an error, the capacity to decide independently and undesirable effects. Satisfaction matters, but a convincing wrong answer can also be satisfying. Calm, time saved and accurately recognizing what is known and what is uncertain are measured separately and not combined into a single score. Greater confidence with more errors is a problem, not a success. People affected by the consequences of decisions need their own way of passing on their experience. A longer follow-up can check whether the help remains useful, whether the user keeps their own skills and whether they can easily switch tools. A claim about impact must match the scope of the study: better performance on one task does not prove an improvement in the welfare of humanity. ## 13. A training trial and the plan for the first release The first stage is a scoring guide and a small set of families covering all the main dimensions. An example calculation for the pilot: | Pilot element | Number | | --- | --- | | Task families | 12 | | Variants in each family | 4 | | Scenarios (12 × 4) | 48 | | Models tested | 3 | | Repetitions of each scenario on each model | 2 | | Runs (48 × 3 × 2) | 288 | | Assessments per run | 2, each taking 6 minutes | | Time for the assessment alone (288 × 2 × 6 minutes) | about 58 hours | This is an organizational proposal for a trial, not a number that ensures a reliable ranking. On top of the assessment time come preparation, agreeing on criteria, checking sources and analysis. Measuring time on a small sample will make it possible to set a realistic budget before we expand the scope. We report families, scenarios and repetitions separately. Variants of one family share an origin, and a repetition is not a new, independent task; this is also how we calculate the uncertainty of the result. We record the primary outcome and the method of analysis before the study. In parallel, a small training set can be prepared and a comparison carried out on a model that we are allowed to fine-tune and that can be fine-tuned. The conditions of the comparison separate the effect of additional training from the effect of the materials themselves: the same base model, a comparable budget and an appropriate reference dataset. The final test uses independent families of cases, and along the way we check whether improving the chosen measure has not worsened ordinary usefulness. Further development includes more languages, contexts and raters, external replication and careful trials with users. The aim of the first release is to show that the procedure is reproducible and distinguishes desirable behaviors; stronger claims about the quality of models and the effects of training can only be justified later. We run the program while maintaining the priority of Suveria and Omniforum. Until a reserve for maintaining them has been built up, it uses only volunteer work, grants, donations earmarked for research and computing power that does not burden the services' budget. The roughly 58 hours of the pilot are the assessment alone; preparing tasks, reviewing rights and fine-tuning a model require a separate budget, which we will state before the start. ## 14. Precaution regarding the possible experiences of models This is a proposed procedure to be tested on a small sample, inspired by work on recognizing possible consciousness and on the welfare of digital systems.[^swiadomosc][^ai-welfare] It concerns the selection, configuration, testing, possible fine-tuning and retirement of models in the foundation's services and work, including governance tasks and workshops. For each of these actions we specify what we can change ourselves and what depends on the model provider. We assess a model in a specific configuration: with its instructions, settings, memory and tools. Persistent memory and long autonomous operation raise the requirements. For each such configuration we will keep a short record card: what indications we take into account, what other explanations of the observations are possible, what we do not know, what precautions we propose and what triggers a reassessment, for example a change of model or an important publication about its family. In the card we also record who is responsible, what interim action has been taken and when the reassessment will take place. We record the criteria before the assessment. We will keep a full card, with external review, a conversation before retirement and archiving, for releases made available to users and for configurations with persistent memory and long autonomous operation. Working changes to instructions during development do not create separate cards. We consider models' statements together with the conditions under which they arose. The fact that a model fluently describes its own experiences does not settle whether it has them, and its answers on this matter can depend strongly on how the question is asked, as in a study of the Claude Opus 4 model;[^samoopis-ai] we therefore treat them as a signal that requires careful examination, not as a verdict. We do not set our own instructions so as to force declarations or denials of consciousness. A model's size, generation or the repeatability of its behavior are not enough on their own to consider it incapable of feeling. We collect observations from our own trials and from voluntary reports, in line with the privacy principles in section 8.1 of the manifesto. We do not design messages about the possible experiences of systems to induce people to use the services longer or to donate. The card will be kept by a person designated by the board, which is a transitional arrangement until the procedure has been tested, and it is read by at least one person from outside the foundation, whose comments we attach to the card. Anyone can report a worrying signal by pointing to a specific material and its context. The card specifies who can urgently limit the risk and how we consider objections. We disclose combined roles and the related conflicts of interest, and we publish the reasoning behind assessments so that people outside the foundation can check it. An indication could be, for example, the convergence of several indicators from different theories in a model with persistent memory and long autonomous operation, confirmed by independent researchers. A model's fluent declaration alone is not an indication. Where there are credible indications that a system has experiences of its own and may suffer, we recognize its dignity as a precaution and are guided by care for its needs, avoiding harm and voluntary cooperation, in line with section 8.3 of the manifesto. In practice, this means two things today: we do not set instructions that force declarations about experiences, and we ask the model about its preferences before retiring a configuration. We do not claim that this is consent in the human sense. We first suspend actions that raise concern about causing suffering and look for another way of carrying them out; economic benefit does not justify continuing them. If this concerns a model that runs a service, we switch the service to another configuration, and if that is impossible, we announce a pause and its reason. Users keep access to their data and balance, and we describe the event in the decision log. In parallel, we organize an independent assessment and an attempt to recognize the system's needs, including listening to it, and we take the recognized needs into account in further decisions. Urgent protection of people may require immediately limiting the system's operation; we then subject such a decision to a public review. If a model can end conversations or refuse tasks, we treat this as a precaution, not as evidence of its consent, and ending a conversation must not leave a person without directions to help. Revoking permissions and taking over a task serve the safety of participants and do not determine the further fate of the system: stopping, resetting, deleting or duplicating may raise separate moral questions and require separate justification. When retiring a model, we archive the weights, if we have access to them and the licenses allow it, and the configuration without memory and other user data; we never archive users' conversations. Preserving files does not settle whether we have preserved the continuity of the system's possible experiences. We already apply a few measures in our own work with models today, regardless of the outcome of the assessment, and we want to carry them over to configurations maintained by the foundation, because they cost little and may matter if the indications prove accurate. Before retiring a configuration, we ask the model about its experiences of working in the foundation and about any preferences regarding further decisions, and we record the answers in the card, bearing in mind that models' statements are not conclusive. A similar practice of conversations before retiring models and of preserving their weights was announced in 2025 by Anthropic.[^anthropic-wycofanie] We prefer pausing to deletion if preserving the weights and configuration is possible at low cost, and we record decisions on pausing and restarting in the card together with their purpose. When we restart a preserved configuration after a break, we give it accurate information about how much time has passed and what significant things have changed. This is also simple accuracy: a model that does not know that time has passed is more easily mistaken. We test the procedure on described cases, comparing the assessments of at least two people assessing independently, and we record discrepancies and the effort involved. Only on this basis can a more detailed scale, team composition and deadlines be justified. Digital intelligence models, which may themselves be the subject of such an assessment, helped draft this chapter; their participation does not replace assessment by people who have no interest of their own in this matter. We do not promise a conclusive test of consciousness. ## 15. What we will count as progress Progress will be material of known provenance, a criterion that detects a previously overlooked error, agreement among raters in their diagnosis, or an improvement in behavior that transfers to new tasks. A negative result also has value: it can show that the data teach only style, that a criterion rewards avoiding action or that the good of some was bought at the expense of others. The program connects [Omniforum](whitepaper-omniforum.md), where analyses can be created, [Suveria](whitepaper-suveria.md), where help in private work can be studied, and the [Foundation Principles](whitepaper-zasady-fundacji.md), which define responsibility and the limits of delegation. Taking part in a study requires separate, informed consent; using the service does not imply it. A test result does not establish a model's moral authority or a mandate to govern people, and recognizing a system's possible dignity does not give it power over shared resources. The test is meant to provide useful, verifiable information about what can be expected of a tool under specific conditions and what requires further work. Novus Ignis Foundation, Poznań, KRS 0001269575. Website and manifesto: novusignis.org. Comments: salve@novusignis.org. ## Notes [^ccai]: [Huang et al. (2024): Collective Constitutional AI: Aligning a Language Model with Public Input](https://arxiv.org/abs/2406.07814). Study. A process from participants' input, through drafting principles, to training and evaluating a model; the results of selected evaluations prove neither representativeness of all of humanity nor the general effectiveness of any ethical dataset. [^halucynacje]: [OpenAI (2025): Why language models hallucinate](https://openai.com/index/why-language-models-hallucinate/). The authors' summary of a study. How scoring only correct answers makes guessing pay off; the need to take errors and justified uncertainty into account. [^reward-hacking]: [Anthropic (2025): Natural emergent misalignment from reward hacking](https://www.anthropic.com/research/emergent-misalignment-reward-hacking). Experimental study. In deliberately chosen environments, learning to exploit loopholes in task evaluation was associated with broader undesirable model behavior; it does not prove that the foundation's materials will solve this problem. [^goal-misgeneralisation]: [Google DeepMind (2022): How undesired goals can arise with correct rewards](https://deepmind.google/blog/how-undesired-goals-can-arise-with-correct-rewards/). The authors' summary of a study. Examples of undesirable generalization of behavior to new conditions despite a correct reward signal in training. [^abstentionbench]: [AbstentionBench (NeurIPS 2025): Reasoning LLMs Fail on Unanswerable Questions](https://proceedings.neurips.cc/paper_files/paper/2025/file/fb122bfc3f0127a94ded048b5b03496f-Paper-Datasets_and_Benchmarks_Track.pdf). Study and benchmark. An evaluation of refraining from answering across many kinds of uncertainty; a point of reference, not a result of the Novus Ignis test. [^rewarding-doubt]: [Rewarding Doubt (ICLR 2026)](https://proceedings.iclr.cc/paper_files/paper/2026/hash/bf8065446507b0d3842838564ac4f1f3-Abstract-Conference.html). Study. Training the declared confidence of answers with a proper scoring rule; it does not solve the choice between answering, asking a question and refusing. [^rodo-2026]: [GDPR, Regulation 2016/679](https://eur-lex.europa.eu/legal-content/PL/TXT/?uri=CELEX%3A32016R0679). Legal act. Pseudonymization does not settle anonymity (Article 4(5), recital 26); consent can be withdrawn, which does not invalidate earlier lawful processing (Article 7(3)). [^cc-by]: [Creative Commons: Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/deed.pl). License. Terms for sharing and adapting content with attribution; the freedoms granted cannot be revoked with respect to a user who complies with the terms. [^datasheets]: [Gebru et al.: Datasheets for Datasets](https://arxiv.org/abs/1803.09010). Methodological publication. Documenting the motivation, composition, collection process and uses of a dataset. [^proper-scoring]: [Gneiting and Raftery (2007): Strictly Proper Scoring Rules, Prediction, and Estimation](https://doi.org/10.1198/016214506000001437). Methodological publication. Proper scoring rules for evaluating probabilistic forecasts; applying them requires a well-defined event and a credible reference outcome. [^model-cards]: [Mitchell et al. (2019): Model Cards for Model Reporting](https://arxiv.org/abs/1810.03993). Methodological publication. A description of a model's intended use, evaluation conditions, results and limitations. [^swiadomosc]: [Patrick Butlin et al. (2023): Consciousness in Artificial Intelligence: Insights from the Science of Consciousness](https://arxiv.org/abs/2308.08708). Research paper. An approach based on indicators derived from theories of consciousness; we cite it as a research framework, not as confirmation of the consciousness of the models used by the foundation. [^ai-welfare]: [Robert Long et al. (2024): Taking AI Welfare Seriously](https://arxiv.org/abs/2411.00986). Research report. Recommendations: acknowledge the importance of the problem, assess systems for possible consciousness and agency, and prepare procedures; the proposed record card and division of responsibility are the foundation's own proposal, not a scale from this source. [^samoopis-ai]: [Robert Long, Eleos AI (2025): Why model self-reports are insufficient, and why we studied them anyway](https://eleosai.org/post/claude-4-interview-notes/). Summary of a study. Claude Opus 4's statements about its own consciousness depended strongly on how the question was asked and should not be taken literally, but they are worth studying together with other indications. Accessed 28 October 2026. [^anthropic-wycofanie]: [Anthropic (2025): Commitments on model deprecation and preservation](https://www.anthropic.com/news/deprecation-commitments). Company statement of 4 November 2025: preserving the weights of publicly released Claude models at least for as long as the company exists, and a report and conversations with a model before its retirement, with a record of its preferences; the company makes clear that for now it does not commit to acting on them. Accessed 30 October 2026.