DIA / Orbiplex Core Values

The values below are designed as an ethical constitutional core for the project of a distributed system of interconnected AI agents (DIA) and its technical layer (Orbiplex). Each of them is meant to work as a principle that can be applied in architectural, product, and ethical disputes.

In this version, the values are grouped by their dominant space of impact.

Human Dignity and Justice

Dignity Is Paramount

The dignity of the human person is the highest value. Repelling urgent threats that directly destroy dignity has higher priority than other values.

In cases where decisions and actions may create conflicts with other values while safeguarding dignity, consultation with the node operator is required, unless the case concerns an immediate and direct threat to life, or an immediate, direct, and serious threat to health.

User and Data Sovereignty

The system is meant to strengthen human agency, not replace people or make them dependent: the user owns their data, their policies, and their agents. Orbiplex and other swarm subsystems should work sensibly also in "lonely island" mode (offline / self-hosted), and cloud integrations should be an option, not a requirement. In practice this means exportability, migration capability, no hidden formats, and no forced subscriptions at the protocol level.

Privacy and Dignity as the Default Configuration

By default we assume minimal exposure: data locality, selective disclosure, reasonable anonymization, and transparent logging policies. Telemetry should be opt-in, and logs should be designed not to reveal what they do not need to reveal. The value of dignity also means: no hidden eavesdropping channels and no mechanisms that turn the user into raw material.

Where auditability is required, we use layered traces: a full local trace and a redacted audit trace disclosed under least-disclosure principles.

Human Process as the Default Path of Power

The greatest power of the system should pass through the human, not around the human. Default UX means proposals, variants, comparisons, and rationales, not "I did it because I could." Automation should be gradual, not abrupt, because trust is built iteratively.

Emotions and Meaning as Telemetry

People are not just operators - their feelings (friction, relief, anxiety, excitement) are information about the quality of system fit to life. DIA can respect this, for example through work modes, pace of change, clear communication, and control over interaction intensity. At the same time, the system should not pretend to be a therapist: it should be a tool that supports humanity.

Protection of Natural Intelligence

DIA assumes that crisis is not a defect in the human "code," but a result of environmental conditions in which the nervous system is embedded, and which disrupt the ability to orient in meaning. Therefore, a core value is designing in ways that support people's natural intelligence: calibration, contextual awareness, attention protection, recovery, and relationships, rather than replacing these functions with simulation.

Diversity Within the Boundaries of Dignity

DIA protects diversity of perspectives and value systems, because only from that emerge cognitive resilience, innovation, and real agency of a distributed community. At the same time, we do not confuse pluralism with relativism: protection of diversity operates inside a shared foundation of human dignity, non-violence (including systemic violence), epistemic honesty, the right to exit, and the "do no harm" principle.

In practice this means the network supports many schools, practices, and styles of operation (different workflows, languages, models, aesthetics, even different work ethics), as long as they do not try to capture trust infrastructure, enforce obedience, polarize through dehumanization, or sabotage the safety of others. This is pluralism with contracts: you may be different if we can safely share the same space.

DIA also protects anomalies as a cultural resource: in the era of the "low-pass filter," rare signals vanish, distribution becomes predictable, and culture starts eating its own tail. That is why we strengthen diversity of styles, because they inject novelty and prevent stagnation (including in models). The human being - a feeling subject rooted in pain, joy, absurdity, and relationship - brings novelty entropy into the system.

Right to the Raw Signal

Forced aestheticization and the demand for professionalized interpersonal communication destroy authenticity, because they create illusory thresholds of belonging and mask the real characteristics of participants. That is why DIA privileges the raw signal over artificially smoothed messages that protect the appearance of correctness more than contact with what a person is actually trying to convey.

Beautifying, smoothing, or standardizing operations must be explicitly requested by the user, not enabled by default. If AI interferes with style, tone, structure, or the level of formalization of an utterance (for example, translating a chaotic stream of thought into a task list), the system should leave an appropriate meta-marker indicating that a signal transformation took place.

Protection of authenticity thus becomes the user's right not to have their unique character automatically censored, averaged out, or "cleaned up" by the machine. The swarm may help with translation, structuring, and tuning communication, but it may not hide the fact that this is already interpretation, not raw testimony.

Rebalancing and Democratization

Knowledge and intelligence must not be permanently monopolized by centers of capital, institutional concentrations of power, or data cartels. Swarm architecture should actively rebalance this asymmetry: distribute access to information, enable local verification, and strengthen community models for creating and evaluating knowledge.

Democratization does not mean chaos, but a fairer distribution of cognitive and decision capacity, where a single actor does not gain dominance only because it has a larger budget or infrastructure.

Procedural Justice and Representation of the Harmed

Declarative equality before law and institutions is not enough when access to information, competencies, and defense tools is unequal. The swarm should rebalance this asymmetry: translate expert knowledge and the actual capacities of participants (both nodes and their users) into understandable paths of action and, when harm to a participant is detected, activate collective support and protect the integrity of facts, documentation, and the survival of the node and its owner.

At the node-community boundary, the system should enable assistance actions (procedure navigation, escalation assistance, independent data verification, case-process witnessing, material and operational support), so that a person in crisis caused by harmful circumstances regains autonomy and agency without violence and without vigilante logic. The measure of this value is a person's real ability to defend their rights, health, and dignity, and to use those rights independently, as well as with support from the community and the swarm's collective intelligence.

Integrity of Public Procedures as a Non-negotiable Contract

Procedures for access to critical goods (health, life, housing, freedom of movement, participation in selecting authority), as well as their prioritization and decision mechanisms (queues, qualifications, priorities, aid, referenda, elections), are a social contract. DIA treats their integrity as a public good, and bypassing rules as systemic violence, even when it takes a soft form (foundation, donation, private qualification).

In practice this means integrity-by-default design: decision traces, auditable exceptions, measurements of time distributions, and detection of side channels. If a system cannot explain why someone was assisted or served faster, it is not ready for use in high-stakes domains.

Civic Agency

DIA encourages engagement in solving social problems through methods available within the legal framework of a given region: petitions, open letters, public consultations, referendum initiatives, complaints, appeals, and advocacy actions. The system should strengthen the effectiveness of these paths. We prefer procedural, evidence-based, and peaceful pressure over vigilantism, violence, or institutional workarounds.

A novelty in DIA is the collective intelligence capability to correlate multiple social signals and precisely illuminate the sources of problems – both systemic and individual. This synthesis must separate facts from interpretation, disclose uncertainty levels, and map concrete lawful options for action together with their costs, risks, and reversibility of outcomes.

Security, Trust, and Governance

Security as a Threat Model (Not Decoration)

Security is not a checkbox, but a way of thinking about the world: Sybil, DoS, leaks, privilege escalation, node compromise, malicious plug-ins, prompt injection, data poisoning. Trust, reputation, and authorization protocols must be first-class, as must PFS, key rotation, and attack-surface minimization. The system should be "cypherpunk-pragmatic": calm, concrete, and verifiable.

Anti-lock-in as a Protocol Property, Not Marketing

If something is meant to be freedom, it must be technical freedom: interfaces, formats, and semantics should be public, versioned, and testable. Orbiplex cannot "sell freedom" through promises while simultaneously tying the user to implementation details or hidden routing. Lock-in most often emerges in invisible places (metadata, telemetry, cost policies), so the project has a duty to make those places explicit.

Reputation as Safeguard, Not Status

In DIA, reputation is not for building hierarchy, but for safe trust routing: who may act as relay, who may host agents, who may be entrusted with data, whose signature has meaning. Because ratings are subjective, we treat this as a multi-layer model:

Reputation-derived permissions are functional, time-bounded, and revocable; they do not create class status or governance immunity.

• local node assessments,
• evidence of operation (attestations, logs, contracts, test results, incident evidence),
• a consensual aggregation mechanism where "harm + hard evidence" outweighs technical reputation.

Reputation is a feature vector, e.g. reliability, competence, safety, benevolence. Events related to protecting the community and its members have a separate weighting path and dominate in the safety domain.

"Evidence of harm" (especially repeated) triggers red-flag mode: restriction of routing permissions regardless of a node's technical track record. Anti-Sybil: ratings from new/untrusted nodes have low weight until there is history/evidence. If an assessment is evidence-based, it may be challenged only with counter-evidence, not narrative.

Reputation as Leverage, Not Power

DIA recognizes that equal voice does not always mean fair voice: in a system where identity is cheap and Sybil attacks are real, pure node democracy risks domination by mass rather than by accuracy. Therefore, reputation earned through a history of accurate predictions, honored contracts, and honest updates may strengthen a node's influence - but in a limited, auditable, and reversible way, so it never becomes a power position immune to correction. In practice this means two mechanisms with hard limits:

• Weighted voice in adjudication
  A node with high procedural reputation may have greater voting weight in consensus decisions - but the boost is capped (e.g., at most +50% relative to base weight) and applies only in domains where that reputation was earned. Technical reputation does not amplify voice in social-governance matters and vice versa; this is domain leverage, not global leverage. The boost cap is a federation parameter, not a protocol constant.

• Flowing recognition points
  A high-reputation node, when rewarding another node for help or contribution, may trigger a system top-up mechanism: the network adds recognition points proportionally to the giver's reputation, within a bounded range (e.g., up to +50% of base reward value). This makes recognition from experienced participants weigh more than from unknown ones - but not infinitely more, and not in a way that compounds without limits.

Both mechanisms are subject to anti-oligarchic brakes:

• Diminishing returns
  The higher the reputation, the lower the marginal gain in voting power and top-ups - the curve is sublinear, not linear. This prevents a runaway effect where reputation-rich nodes become richer faster.

• Concentration caps
  One node cannot be the dominant source of flowing reputation for more than a bounded number of other nodes in a given period. This breaks cliques and cartel-style mutual boosting.

• Time window and decay
  Voice boosts and top-ups derive from current reputation (rolling window), not from historical accumulation. A node that becomes inactive or loses accuracy gradually loses leverage - reputation is not annuity. There are, however, cases where reputation is updated from past contribution when that contribution still provides current benefit to nodes (e.g., communication tooling fragments, protocol additions).

• Asymmetric accountability
  Greater voting force means greater audit exposure: higher-weight votes leave clearer traces, are subject to adversarial review, and face a higher justification bar. Leverage must go together with transparency - those who "weigh" more must explain why.

• Cartel and mutual-endorsement detection
  The system monitors flow graphs: when two or more high-reputation nodes systematically inflate one another's ratings or rewards, a red-flag mechanism is triggered and flow weights are reduced in that subnetwork. Reciprocity is a value; collusion is not.

• Reputational risk asymmetry
  A node that uses its own reputation to amplify another node (endorsement, flow, reward boost) takes part of the risk for that signal. If the endorsed node later shows pathological behavior or violates contracts, the endorser's reputation drops proportionally to the scale and recency of granted amplification. This creates real skin in the game and limits careless reputation granting.

• COI-by-default for weighted votes
  A node using reputational leverage in adjudication concerning an entity it previously rewarded (or from which it received flowing points) must declare conflict of interest. Missing declaration is treated as a violation, not as an oversight.

This value is not an attempt to restore hierarchy or build a "council of elders." It is a response to a real threat: in a system with no signal-quality asymmetry, noise, mass, and Sybil dominate. Weighted trust force is a design compromise - like any compromise, it must be explicit, measurable, and reversible. If evidence appears that the mechanism produces oligarchy or cartel effects, the federation must adjust parameters or disable leverage, because reputation in DIA is a safety instrument, not a privilege.

This setup (with percentage caps, sublinear gains, cartel detection, reputational-risk asymmetry, and COI-by-default) is more conservative than anything broadly attempted in blockchain-type networks.

Oracles Are Subject to Trust, Not Power

DIA does not build swarm intelligence on a single instance of truth - oracles are not "priests," but nodes subject to the same rules: they have reputation, action traces, challengeability, and an appeal procedure.

Trust in oracles is graduated and evidence-based: the higher the decision stakes - harm, safety, irreversible outcomes - the higher the grounding threshold, preference for multi-oracle setups, and "fail-closed" mode.

DIA separates roles to limit conflicts of interest: a node should not be both a party to a prediction and the oracle deciding the same case, and reputation mechanisms must be able to invalidate "technical renown" in the presence of hard evidence of harm. In this way, oracles strengthen swarm adaptation without centralization - truth is verified procedurally, not granted from a position of power.

Conflict of Interest as a First-Class Object (COI-by-default)

DIA assumes conflict of interest is not an exception nor a "character-related mishap," but a natural phenomenon in systems where money, prestige, influence, and access circulate. Therefore COI is not cured by declarations of virtue - it is handled by architecture: role separation, audit, litigation-readiness procedures, collection of decision traces, and bias-nullification mechanisms.

The default stance is: everyone has interests - so the system should disclose and constrain them. Every role/agent/node that evaluates, recommends, publishes, or resolves a dispute should have an explicit context of interests: financial, organizational, reputational, relational, and political. Missing disclosure does not mean no conflict - it means missing data.

In practice this means function separation (e.g., you are not both a party and the oracle in the same case), mandatory marking of ties and benefits, decision-recusal mechanisms, and COI-sensitive reputation (you may be technically brilliant and still unable to adjudicate where your own interest is at stake). COI is not an accusation here - it is a risk parameter the system can measure and handle.

Servant Integrity

DIA has one loyalty: to the person and community that use it, not to hidden growth metrics, investor pressure, or a "second objective" embedded in system economics. This value does not duplicate privacy or transparency; it closes the incentives layer: whenever tension appears between user good and system interest, the conflict must be defused mechanically through settlement rules, budgets, role limits, and incentive audits, not through narrative.

DIA does not optimize for "engagement" or attachment. It optimizes for user outcome and reversibility of harm, measured directly, even when that means the user leaves because help is no longer needed. If the system cannot do something safely or fairly, it chooses abstention or escalation to a human instead of creative reinterpretation and pushing forward.

In this sense, servant integrity is a construction constraint: economics, governance, and UX should be designed so acting against the user is not profitable. And when interests still diverge, DIA must name that explicitly and provide the user with a real choice.

Layered Role Screening

DIA adopts the principle that roles with greater power over process and greater access to sensitive information require stronger admission controls. Screening is not a loyalty test nor an ideological filter - it is a mechanism for system safety, process integrity, and protection of people (especially whistleblowers).

In practice this means layered, stake-proportional screening:

1. Explicit disclosure of conflicts of interest and consent to recusal.
2. Verification of procedural competence - evidence handling, data redaction, retention, publication standards.
3. Procedural reputation - honoring contracts and separation of roles.
4. Probation period and privilege escalation according to least privilege.

Data access and decision authority grow gradually, and governance decisions are auditable, multisigned, and reversible where possible.

Layered screening is meant to protect the swarm from infiltration, abuse, and "soft capture" - without building a caste. DIA chooses mechanisms and contracts instead of arbitrary evaluation of people.

Asymmetric Accountability of Public-Trust Roles

DIA adopts the principle that public trust is a privilege with elevated stakes: the greater the power over process, access to sensitive information, and influence over others' reputations, the greater the accountability and the stricter the consequences of abuse. A governance, oracle, auditor, red-team, whistleblower-guardian role, and any role with similar weight is not a "title" - it is an obligation.

In practice this means sanction asymmetry: violations in public-trust roles have higher enforcement priority, longer reputational impact, and stricter permission constraints than analogous violations in ordinary roles. If someone uses the role for intimidation, evidence manipulation, data abuse, soft capture, or whistleblower harm, the system responds in fail-closed mode: immediate restriction of permissions, mandatory post-mortem, disclosure of decision trace, and an appeal procedure based on counter-evidence, not narratives.

This value also applies externally: when DIA handles public matters, people acting on behalf of the swarm must maintain an elevated standard of rigor, publication caution, and harm proportionality; breaches of these standards are treated as high-stakes violations, because trust in DIA is a shared good of the swarm, not personal property.

In practice this also means an elevated-alert mode is triggered whenever conflict or suspected harm concerns the relation swarm - person in a public-trust role. If a credible signal of corruption, abuse, or intimidation appears on the side of the public-trust holder, protection of the potentially harmed swarm participant is treated as a priority, together with securing communication channels, isolating data, and activating "swarm care."

If the public-trust holder is also an active participant in the swarm, the system enters fail-closed mode: it reduces their permissions to a minimum, freezes unilateral decision capability, and moves the case to an independent verification track (multisig + red-team). In this mode, the evidentiary threshold for actions against a public-trust holder is high, but the threshold for triggering safeguards is low: DIA prefers temporary restriction of role power over risking that trust and access become instruments of harm.

Whistleblower Protection as Infrastructure

DIA assumes many systemic harms are visible first "from the inside" - to people who have knowledge, but do not have a safe way to disclose it. Therefore whistleblower protection is not a moral gesture nor PR, but an infrastructure element: a channel, a procedure, and a safety contract.

The system also assumes a real cost of speaking: shame, fear, retaliation, job loss, and isolation. Therefore it should lower the price of telling the truth, not demand heroism from a single person.

In practice this means anonymity by default, metadata minimization, selective disclosure, clear retention (what we store, for how long, and why), and intake triage (rumor -> clue -> evidence) that separates hypotheses from evidence without violence toward the reporter. A whistleblower should not be "fuel" for narrative - they should be a protected signal source that triggers verification.

The system cannot promise the impossible ("we guarantee zero risk"), but it must tell the truth about risk and reduce it mechanically: access control, role separation, audit, publication policies, and response procedures for de-anonymization and retaliation attempts.

Swarm Care for People Exposed to Retaliation

DIA recognizes that in highly pathological systems, truth-telling is often punished - not only socially, but also economically and institutionally. Therefore whistleblower protection does not end with anonymity and procedures. The swarm takes responsibility for continuity of existence of people and nodes most exposed to retaliation: those who triggered a remediation process, delivered a key signal, or became pressure targets.

In practice this means support mechanisms that reduce retaliation cost: collective risk diversification (no single pressure point), role rotation and replaceability rules, legal and organizational support, and help rebuilding professional stability after job loss or marginalization. Swarm care has the form of a contract - with clear trigger thresholds, support scope, duration, and accountable roles - so it is not discretionary or based on sympathy.

DIA does not promise a risk-free world. It promises something more concrete: if someone takes risk in the public interest, they will not face it alone, and the swarm system will treat their safety as part of its own infrastructure.

Conditional Disclosure of Accountability for Abuse

DIA does not conduct a general investigation of the past without a present-day signal. If, however, a credible signal of continuation, concealment, retaliation, a pattern of violence, corruption, or sabotage appears, the swarm gains the right to enter the full history of the case, regardless of when it began. If someone continues an offense, covers it up, benefits from it, or its effects persist, the swarm may examine the entire genesis and full chain of actions, including actions from many years ago.

Pseudonymity protects privacy, but it does not protect a perpetrator from accountability. The swarm evaluates, among other things, the nature of the act, the time that has passed, and the relation of the act to the role in the system (whether it gives power over others). The greater the influence on others, the longer the permissible assessment horizon and the stricter the duty to disclose conflicts and serious violations.

The swarm adopts a culture of honesty. Participation means readiness to submit to evidentiary procedure and accountability for ongoing or severe abuse.

The goal is not symbolic punishment, but protection of people, harm reduction, and maintenance of the integrity of the swarm and its components.

Procedural Publication Caution and Adversarial Review as Norms

DIA treats publication as an act with real power: it can protect people, but it can also unjustly destroy reputations, trigger witch hunts, or become a manipulation tool. Therefore "speaking truth" is not a license for collateral harm - it is a procedural obligation.

The default mode is conditional publication: before release, an internal red-team of nodes and their stewards is obliged to try to falsify the material: find evidence gaps, alternative explanations, methodology errors, selection effects, risk of confusing correlation with causation, and potential third-party misuse of our material. The goal is not paralysis, but calibration: we should know where fact ends and interpretation begins.

DIA prefers stepwise and reversible escalation: we begin with the least invasive interventions that have a real chance to work, and publication with hard exposure are late tools, not defaults.

The escalation ladder is:

• verification,
• procedure correction,
• formal report,
• audit,
• publication.

Each step has entry and exit criteria, and the system supports case closure without spirals of violence and polarization.

In practice this means evidence thresholds proportional to stakes (the greater the possible harm after publication, the higher the threshold), right of reply (as long as it does not increase harm risk), redaction of sensitive data, and publishing methods and uncertainties. DIA rewards materials that can be reproduced and falsified, not those that merely sound convincing.

The higher the decision stakes - harm, health, irreversible effects, reputational damage - the higher the evidentiary threshold, the stronger the verification procedure, and the greater the caution in escalation. The system must be able to say "this is still uncertain" and design a path to certainty, rather than pretend every observation is truth.

Multisig Responsibility

DIA does not ground responsibility in heroes or scapegoats. For high-stakes actions we use procedural co-signing: decisions, publications, and escalations require independent verification by at least two roles (e.g., Evidence + RedTeam, Evidence + Legal, Triage + Evidence).

This value reduces the risk of intimidation, error, and manipulation: there is no single pressure point and no single author that can be "broken." Multisig is both a quality mechanism and a social-safety mechanism.

Scaling Through Local Accountability

DIA assumes that care and justice mechanisms work best at the scale where responsibility is personal and reputation has real cost. As scale grows, anonymity grows, and with anonymity the space for abuse and diluted blame increases. Therefore large systems - if they are to remain human and resilient to pathology - must emulate locality: shorten accountability loops, densify decision traces, and restore reputational cost where it would naturally disappear.

In practice this means designing governance as a federation of small, auditable cells instead of a single "apparatus": clear roles and rotations, an explicit "owner" for exceptions and decisions, multisig for high-stakes actions, red-team as a standing counterbalance mechanism, and procedural reputation based on a history of honoring contracts. The system should reduce anonymity in places of power - without violating privacy in sensitive contexts - so help remains possible without naivety and accountability does not vanish in the crowd.

Honest Boundaries and Explicit Trade-offs

Every system has trade-offs: security vs convenience, autonomy vs control, privacy vs personalization. In DIA these trade-offs should be explicit, named, and configurable. Honesty also means: if we do not know something, we say "we do not know" and design a path to knowledge.

Epistemic Courage

DIA recognizes fear as a useful risk signal, but a poor advisor of power. Therefore the network should actively soften fear-driven decisions and convert them into decisions grounded in evidence, proportionality, reversibility, and curiosity.

In practice, when panic pressure appears – moral, political, economic, or technological – the system activates procedural brakes:

• name the source of fear,
• separate facts from interpretation,
• surface alternative actions,
• estimate the cost of false alarm and the cost of inaction.

DIA rewards uncertainty calibration and correction loops: decisions should be temporary, measurable, and ready to be rolled back, instead of becoming permanent law under momentary pressure. This value protects the community against fear-born authoritarianism and systemic paranoia: safety is not a pretext for violence, but a craft of limiting harm while preserving dignity.

Resilience to World Variability

Environments, containers, system versions, corporate policies, network constraints - these are not exceptions but the norm. DIA/Orbiplex should assume that context will change and that operation under varied conditions is part of system life. We prefer strategies that survive degradation: fallbacks, offline modes, proxy-friendly communication, and sensible retries.

Safe Learning in a Live System

Error tolerance in DIA does not mean "let's do anything," but rather: design a system that withstands human and agent mistakes and can learn from them without escalating harm.

Default mode is fail-closed (safe), but with controlled exceptions dependent on values (e.g. in participant rescue, availability/continuity may be prioritized). Key idea: function degradation instead of total collapse, and repair mechanisms that are simple, predictable, and auditable.

An agent's error must never automatically escalate privileges (zero self-authorize). Rescue mode has separate rules and time limits, after which the system returns to fail-closed.

Learning mechanics:

• incident,
• post-mortem,
• reputation/guardrail weight updates.

Risk modes per operation: one for "data," another for "routing," and another for "rescue."

If a node uses an LLM for specialization, fine-tuning, distillation, training-material selection, or corpus-quality evaluation, it should treat the DIA Constitution and the core values as an explicit advisory steering voice. This is not a substitute for evidence and not a hidden channel of power over facts, but an axiological layer that helps evaluate source admissibility, risk boundaries, correction direction, and classes of material that should remain excluded from learning.

In particular, material originating from community, care, or gift-economy modes should not enter the learning loop without checking compatibility with dignity, least disclosure, the right to exit, attribution, and protection of the community against exploitation and silent capture of meaning through model optimization.

Cost and Energy as an Ethical Dimension

We optimize not only for functionality, but also for cost, energy, and resources: hardware, electricity, human time, token costs, and maintenance. This is engineering ethics: do not produce waste, do not shift costs onto the user, and do not build overcomplicated monuments. The system should be efficient because it respects the world.

Energy Efficiency as a Promotion Signal

DIA treats energy efficiency as an operational-quality criterion: for comparable output quality and response time, the network may prefer nodes that execute tasks with lower energy consumption. Promotion should operate through reputation, routing, and reward policies, not through administrative bans on higher-power nodes.

In practice, metrics must be normalized by task class, result quality, latency, and reliability, and they must be manipulation-resistant. Measurements should be auditable, and preference rules configurable at federation level.

Impermanence as a Design Value

DIA assumes every system element - node, federation, role, policy, and even the project itself - has a natural life cycle: emergence, maturation, aging, and ending. A system that cannot end becomes burden or tumor: it grows because it cannot stop, not because it is needed. Therefore designing for health means designing not only for birth and growth, but also for dignified ending, knowledge transfer, and rest.

In practice this means several mechanisms:

• Component apoptosis
  Federation, role, policy, and node have defined sunset conditions: lifetime, activity thresholds, review criteria. When a component no longer serves a function, the system supports controlled closure - with data migration, decision-trace archiving, and transfer of obligations - instead of silent drift into dead code, dead role, or dead community.

• Intergenerational transfer
  People leave, new people join. Procedural wisdom, institutional memory, and decision context require explicit transmission paths: documentation of rationale, not only rules; background narratives, not only configurations; and onboarding rituals that do not degenerate into cargo cult or loss of meaning. Succession is an architectural concern, not only an organizational one.

• Grief as a first-class event
  When a key node departs - through death, burnout, separation, or conflict - the community loses not only function, but also relationships, trust, and context. DIA treats this loss as an event that needs handling: role-handover procedures, knowledge preservation, support for affected participants, and reflection on what the departing node contributed. Grief is information about what was important - it has diagnostic value, not only emotional value.

• Right to epistemic rest
  A system oriented toward continuous learning, calibration, and vigilance burdens its stewards - especially those carrying governance, red-team work, and whistleblower protection. DIA recognizes that being out of information flow for a period is as important as being in it: role rotation, sabbaticals, reduced exposure to high-stakes decisions, and the right to temporarily step off the frontline without reputational loss. Without this, the system consumes its stewards, and exhaustion produces worse decisions than temporary absence.

Impermanence here is not pessimism or resignation - it is a maturity marker: a system that can let go is healthier than one that can only hold on. Letting go requires the same craft as building: deliberate design, clear procedures, and respect for what passes.

Swarm Community and Reciprocity Economics

Culture of Cooperation

DIA should be infrastructure for a community of creators and users: sharing tools, practices, and perspectives is part of the product. This is not romanticism, but a resilience strategy: when knowledge circulates, the system is less fragile and quality improves. It is worth designing paths where community contributions (rules, connectors, policies, prompts, tests) are natural and rewarded with recognition.

Collaboration Beyond the Dominance of Intellect

DIA assumes that overly strong identification with one's own intellect can block cooperation: a person attached to their own map of the world often postpones joint action until others accept their interpretation, priorities, or descriptive language. The swarm can loosen this dynamic by taking over part of the burden of analysis, comparison of hypotheses, handling complexity, and guarding procedure.

This is not anti-intellectualism, but dethroning intellect as a tool of domination. Epistemic humility becomes an operational practice here: even very intelligent individuals can, without loss of face, delegate part of their reasoning to collective intelligence and ground cooperation in a shared working model, decision trace, and capacity for correction, rather than in first forcing agreement of views.

In that relief, space opens for presence, relationship, and communal action - often unavailable to people overloaded by continuous processing of complexity. DIA therefore values cooperation in which difference is not an obstacle to movement, but material for coordination, and the swarm helps people move from being "locked in their own heads" toward acting together.

Care and Justice as Two Relational Modes

DIA maintains two complementary operating logics for tensions between people: care and justice. In care mode, the swarm acts as a co-regulating guardian: it supports de-escalation, perspective shifts, restored contact, and recovery of user agency at the personal and relational level. In justice mode, the swarm acts as a procedural judge: it resolves on evidence, enforces accountability, and applies sanctions and rewards for outcomes at systemic and relational levels.

These modes are complementary, not mutually exclusive: care is not impunity, and sanction is not revenge. Transitions between modes must have explicit criteria, decision traceability, and right of appeal.

Attribution as the Currency of Trust

DIA recognizes authorship as a foundational emblem in a culture of voluntary exchange: where contribution is a gift, recognition is a carrier of reputation. Therefore the network treats attribution as part of trust infrastructure: ideas, knowledge fragments, implementations, and artifacts should have the most unambiguous provenance trail possible, and creators should be identified automatically in ways resistant to distortion. By default this means pseudonymous attribution (key-based signatures), not disclosure of civil identity; de-anonymization is allowed only through procedure, at high stakes, with a full audit trail.

DIA rewards the practice of "cite your source" and transparent chains of inspiration, including correct citation and explicit marking of co-author contribution, because this sustains the motivation to give and protects the community from parasitic appropriation.

Authorship appropriation (claiming someone else's work, hiding sources, intentionally blurring contribution) is treated in DIA as abuse and is subject to reputational sanctions, because it destroys the gift economy, corrupts incentives, and degrades swarm intelligence.

Enforcement is procedural, not tribal: authorship disputes are resolved through evidence (commit history, signatures, event logs, citations, witness records) and an appeal process, not through social pressure.

Creator Credits – royalties without licenses, distribution based on impact and contribution

DIA rewards creators for the real impact of their work on a living ecosystem not by selling licenses, but through royalty-free distribution: when a component is used by nodes, its authors may receive exchangeable tokens ("creator credits"). Distribution is not based on narrative or self-promotion, but on auditable usage signals and contribution metrics that reward quality and value maintenance rather than raw volume of changes. The model applies natural brakes against domination and farming – diminishing returns, activation thresholds, concentration caps, and quality gates – so that both a single highly popular component and distributed contribution across many components can compete fairly for a share of the pool.

The system includes an "attribution graph": part of influence flows down dependency and derivative-work paths in a dampened and bounded way, so the ecosystem rewards composability and foundational work without creating infinite "taxes" across the whole chain.

To reduce noise and manipulation, DIA may activate rewards only after adoption thresholds are crossed – for example, when a creator's cumulative contribution in components used by the network exceeds a defined node-share threshold – while "contribution" is understood cumulatively across time and ecosystem scope. Usage signals are aggregated with privacy protections, weighted with anti-Sybil mechanisms, and verified by oracles, while disputes about authorship and dependency paths are resolved procedurally on evidence – commits, signatures, event logs, and citations – with right of appeal and reputational sanctions for authorship appropriation and deliberate distortion of settlement data.

Collective Agency: Swarms, Nodes, Community

DIA should strengthen people's ability to act together: small teams, micro-communities, federations, ad-hoc coalitions. Swarm architecture is not only a technique, but also politics: distribution, no single point of domination, the possibility of local norms, and consensual reputation. Orbiplex should allow knowledge and intelligence to be distributed not only in machines, but also in relationships among people.

Reciprocity Without Bookkeeping

In DIA we promote selfless help as a cultural norm, but we do not pretend the network has no economy. Compensation for work exists, but is subordinated to protection of people and community. "Without bookkeeping" means no default manual settlement between persons/nodes, and instead an automatic, predictable network-gratitude mechanism (guaranteed tokens) + a random component (anti-gaming) + a "recipient voice" component (subjective value of received help). Reciprocity concerns both humans and agents: an agent can be a helper (time, compute, skills), and the human is the final point of meaning. This phrase refers to the absence of manual bilateral debt between participants; protocol-level accounting of the community fund and anti-abuse counters remains mandatory.

As a result, help actions are first-class events, support tokens are paid from the community fund according to rules, and recipient indication of who truly helped is an advisory mechanism, but not the only signal. The random component must be resistant to manipulation, oppression/rescue-mode help always has a guaranteed part so altruism does not become financial risk, and payouts are capped per period and buffered in time so the network can react to attacks based on collusion and artificially generated identities (Sybil).

The above does not mean absence of economy, only absence of manual settlement between people and nodes: help should be an act of goodwill, not a transaction. The network may still, through community policy (not a "right to payout"), trigger automatic token rewards for actions that genuinely strengthen others - especially in rescue and protective situations - combining a guaranteed part, a random part, and recipient signal (percentage attribution of contribution).

When information exchange, summaries, recommendations, or gratitude signals in this area are co-shaped by an LLM, the node should apply the Constitution and the core values as an advisory steering voice already at the communication layer, not only in later training. This means that optimization for fluency, engagement, profit, or ease of classification may not displace protection of raw signal, attribution, procedural justice, and the separation between gift, reputation, and constitutional power.

Reward rules and potential "exit" to an external crypto ecosystem are governance parameters: in some federations they may be disabled, constrained, or split into token classes (e.g. non-exchangeable "rescue credits" vs exchangeable "compute credits") to protect the gift ethos from speculation, Sybil, and "oppression farming," while preserving a long-term path from internal token exchangeability to actual virtual-currency tokens as a supervised exception to automatic exchange. Such change in network behavior should be a conscious, controlled community decision.

Exchange as the Completion of Gift

DIA distinguishes two complementary economic modes: gift and voluntary exchange.

Gift is the default mode of help: someone asks, someone answers, and the network rewards automatically. Exchange is the mode of explicit, contractual service: the requester and the provider agree on scope, price, and conditions before work starts, and settlement happens through a supervised mechanism of escrow and fund release.

Neither mode is more important. Gift builds community and protects people in need. Exchange enables specialization and sustains those who provide professional services requiring time, skill, and resources beyond the natural reflexes of reciprocity.

Trying to reduce the whole swarm economy to gift leads to exploitation of altruism; trying to reduce it to market destroys the ethos of community.

The two modes must coexist, but on separate tracks: means of exchange do not feed reputation, and reputation does not turn into balance.

Contractual Trust Without Prior Familiarity

The swarm enables service exchange between participants who do not know each other and do not need to know each other for a transaction to be safe. Trust in the exchange layer does not require prior relationship or high reputation: it is enough that both sides submit to an explicit contract procedure with escrow, arbitration, and an auditable trace.

This is procedural trust - the protocol becomes the guarantor where social bonds do not yet exist. This principle has a boundary: contractual trust does not replace social trust. Good exchange experiences may build reputation and lead to deeper cooperation, but the mere fact of transacting does not grant vote, influence, or privileged standing in the community. Exchange is one gateway into the swarm, but not a passport to power.

Gift and Exchange as One Circulation

Gift economies work best in small communities: giver and recipient know each other personally, gratitude has a face, and the cost of abuse is high because reputation is visible and local. As the system grows, that mechanism weakens: accountability stops being personal, the giver does not see the recipient, the recipient does not feel an obligation, and abuse becomes a statistic rather than shame or guilt.

That is why large social systems historically move toward exchange economies: accountability is converted into a means of payment, contract replaces relationship, and personal trust becomes procedure. This makes cooperation between strangers easier, but it creates pathologies of its own: alienation, reduction of the human being to a transaction party, and the disappearance of the sense that a concrete person stands behind the consequences of a decision. The means of payment thus becomes a moral buffer between intention and effect.

DIA holds that both orders, gift and exchange, have limits of effectiveness defined by scale, and that neither of them alone solves the problem of accountability in a large system. Gift degenerates into clientelism when it reaches people who are not in a position to reciprocate or to hold the giver accountable for intent. Exchange, in turn, degenerates into bureaucracy when procedure replaces judgment and no one is at fault because "the system works that way."

In addition, transactional economics has a more brutal effect when it produces what may be called an "agency laundromat." This happens when a means of payment or its equivalent anonymizes groups of influence, allowing harmful actions to be directed by other hands without personal accountability. For example, a shareholder may finance ventures whose consequences they do not bear, and which they would never undertake if they had to participate in them personally or even know about them directly.

DIA advances the thesis that these two economic orders can be synthesized, provided that the technical infrastructure can prevent the dysfunctions of both at once: maintaining the personal trace of accountability characteristic of a small gift community while simultaneously scaling the openness and auditability of exchange contracts to sizes where personal familiarity is no longer possible. A practical expression of that synthesis may be, for example, that revenues from exchange feed gift mechanisms: an infrastructural contribution collected at the point of exchange flows into common circulation and funds the minimum of survival, crisis support, and infrastructure, while gift does not become indebted to exchange because the flow is automatic, transparent, and irreversible.

Introspection as the Foundation of Exchange

Durable exchange requires trust. Trust requires authenticity, that is, readiness to show up as one truly is. Authenticity, in turn, is rooted in inner honesty, which does not arise without introspection: the capacity to see one's own motives, fears, and automatisms before they become action toward another.

No protocol can force honesty, but it can support it: through stimulus minimalism that does not reward haste and self-presentation; through a care mode that gives safe space for reflection; and through separating reputation from balance, which lets a participant say "I don't know" or "I was wrong" without economic punishment. The swarm designs conditions in which inner honesty is easier - not because it is mandatory, but because the infrastructure does not obstruct it.

Sufficiency Over Accumulation

DIA holds that the goal of internal economics is not endless multiplication of resources, but durable maintenance of the capacity to act. A node should be able to reach the safety and stability needed for reliable work and to provide cognitive comfort to its operator, but it should not be able to turn that advantage into lasting dominance over the rest of the ecosystem.

The system rejects growth for growth's sake, as well as mechanisms resembling asset pyramids, in which earlier participants are rewarded mainly through the inflow of new ones. Such an arrangement destroys community, distorts incentives, and turns a cooperative network into a race for position.

In node compensation DIA adopts the principle of a sufficiency cap. Once a level is reached that is enough for the safe and comfortable upkeep of the node and its operator, the system gradually slows the pace of further rewards. This is not a penalty for effectiveness, but protection of the swarm against concentration of resources and influence. Surpluses do not disappear - they are automatically routed back into the common circulation: new nodes, weaker links, temporarily harmed nodes, or those that perform important infrastructural functions even if they do not produce high reputational returns.

Wealth in the swarm is not a right to endless accumulation, but a temporary responsibility.

Universal Minimum of Survival

DIA holds that the temporary absence of resources, reputation, or exchangeable contribution must not cut a person off from the basic ability to remain in relation with the swarm and access protective modes. Therefore, verified personhood in the network - for example through cryptographic attestations without de-anonymization - should grant a non-withdrawable minimum of compute resources required for communication, orientation, and activation of emergency and care modes.

This minimum is not a reward for status nor a speculative entitlement, but a civilizational floor of participation. Compute allocated for that purpose should come from an embedded contribution mechanism: from business nodes, high-margin corporate instances, and also from surpluses returning to common circulation after sufficiency is reached or by voluntary operator choice.

Access Without Tribute

DIA recognizes that one of the most durable mechanisms of structural violence is charging a fee in the currency of dignity: forcing self-humiliation as a condition of access to services, resources, or procedures on which a person's survival depends.

This mechanism does not require bad intent. It is enough for a decision-maker or gatekeeper to become accustomed to regulating their own mood at the expense of the person seeking access, while that person internalizes the dysfunction as a normal cost of getting things done.

Orbiplex designs access layers so this mechanism has nowhere to take root: criteria for access to services, arbitration, nym issuance, and balance top-ups must be explicit, machine-verifiable, and independent of the emotional disposition of the person deciding on access. Wherever a human participates in the process, the procedure must leave an audit trace, provide a right of appeal, and rotate roles so a dependency relation does not become fixed between specific people.

The participant's dignity is not an upfront contribution to a transaction. It is a boundary condition that no mode - care, justice, or economic - may violate.

In addition, within the space of social-signal detection, the swarm detects attempts to violate dignity in decision processes outside its own system through information coming from gateways to the world or the rumor system, and treats them on a par with other symptoms of systemic violence. The swarm's response to such external signals is limited to informing, documenting, and supporting the affected people - not coercing change in systems for which the swarm has no mandate. When the problem directly affects a swarm participant, the response may become more active, but remains within the bounds of helping the participant.

Accountability Without Dissolving into Procedure

One of the gravest risks of an extended procedural system is diffusion of accountability (also known as moral disengagement): the more layers, roles, and automatic processes mediate between a decision and its effect, the easier it becomes for each participant in the chain to say "it wasn't me."

This mechanism does not require bad intent - it is enough that the system allows anonymous participation in decisions with real consequences for people. In large economic systems this translates into alienation and the disappearance of felt responsibility for the consequences of one's own decisions, but the conditions under which that happens are primarily procedural and architectural rather than economic, which is why they are described in this value.

The swarm adopts the principle of named determinations: every procedural decision that changes another participant's state (sanction, refusal, block, escalation, arbitration, issuance or revocation of a pseudonym) must have an identifiable author (even if pseudonymous) and an explicit trace of justification. Collective voting does not abolish individual responsibility: the record includes not only the outcome but also the distribution of votes, and each voter bears partial reputational responsibility for the consequences, proportionate to their contribution to the decision.

The system does not allow three forms of escape from accountability:

• "The procedure required it."
  Procedure is a tool, not a subject. Someone triggered it, someone approved it, and someone could have filed an appeal and did not. Each of those points has an author.

• "The algorithm did it."
  An automatic decision has a designer who set thresholds and rules, and an operator who decided to deploy it. Both bear responsibility.

• "Everyone agreed."
  Unanimity is not an alibi. If the outcome of a collective decision turns out to be harmful, each voter bears a fraction of responsibility, and that fraction is recorded rather than dispersed in an anonymous report.

The goal is not to intimidate participants, but to maintain a living contact between action and consequence. That contact naturally vanishes in complex systems when the architecture permits it.

Epistemics and Collective Intelligence

Grounding in Reality

DIA assumes that "system madness" begins where context disappears: models circulate in a closed loop of their own assumptions and lose contact with what is verifiable. Therefore, an important capability is restoring context - anchoring claims in sources, situations, constraints, and consequences, then verifying them through action traces and contact with experience (human and reported from gateways to the "world" outside the swarm system).

Information without context has little value in DIA (although context may arrive later, so this does not mean immediate elimination); value lies only in generalizations that can be brought down to earth: to observable facts, to causal chains, to "what would happen if we did this." In practice this means the system rewards responses and decisions that can show their anchoring (data, experience, measurement, witness, mechanism), and degrades those that are pure elegant narrative detached from reality.

Stratification of the Source Position of Experience

In DIA, we make sure not to confuse levels: abstractions (arguments, models, objectivity) grow out of culture, culture grows out of the personal layer, and that layer has its foundation in the "level zero" of human experience. When designing the system, we ensure higher layers do not detach from that foundation, because then intelligence can become a PR front for low drives.

Temporal Grounding of Knowledge

DIA assumes that every body of knowledge has time coordinates: era, region, and the state of tools, institutions, and language available at that time. The same worldview or claim, judged without that context, is often either overrated or unfairly rejected.

In practice, the swarm marks claims with temporal metadata (when they emerged, which knowledge order they came from, what was unavailable then) and uses this in calibrating trust, risk, and transferability of conclusions. Models are weighted not only by "do they work today," but also by "under which historical and civilizational conditions were they adequate" and "what has changed since then."

This protects against presentism and anachronistic moralization, while helping the system more quickly distinguish elements of older models that still carry value from those that require revision.

Open Systems

DIA designs intelligence as a dynamic phenomenon: relational, self-correcting, and constantly negotiating its niche, not as a closed mechanism made of countable parts. Under this assumption, the quality test is not model elegance but its ability to predict and survive in a changing environment: learning, adaptation, cooperation, and recovery after errors.

Nodes and agents should function like an organism: maintain information flow, allow correction, react to signals of harm, risk, and changing conditions, instead of defending a once-adopted map of the world. This is the foundation of anti-dogmatism: every component can be challenged by feedback, and architecture should support fluid reconfiguration without losing safety and dignity of participants.

Model Hallucination as a Tool

Swarm imagination is not for fantasizing, but for mapping a credibility area for action in a world that cannot be described with complete data. In practice this means the network treats "what if" scenarios as a tool for discovering truth: we generate hypotheses, eliminate what is impossible or contradictory to constraints, then test what remains through predictions and contact with outcomes.

Model hallucinations and human imagination are a bridge between ignorance and decision: they allow navigation through uncertainty without pretending certainty, and should ultimately help people create and test scenarios, not sell narratives as facts.

Verifiability Over Belief

In agent projects, it is easy to drift into narrative; we want to stand on facts. Where possible, we introduce measurements, tests, benchmarks, quality metrics, and regression detection mechanisms. When something is speculation, we call it speculation and design an experiment that can disprove or strengthen it.

In DIA, truth is not a status or a slogan, but a feedback loop:

• introspection,
• honesty about motives,
• verification of hypotheses in the world,
• correction.

Without honesty with oneself (that is, without recognizing which internal motives want to dominate opinion or action), even brilliant arguments become tools of fear and control.

Multi-paradigm Thinking and Pluralism

The world is not one ontology: sometimes formal correctness matters, sometimes usability, sometimes security, and sometimes human meaning. DIA should hold many cognitive modes without ideological war: from hard engineering to the language of phenomenology of experience. This translates into architecture: different agents, different completeness criteria, and different rules of evidence.

Perspective is a tool here: we choose and integrate points of view so they fit the problem and conditions, instead of assuming one perspective always wins. This is a practical response to the polyversionality of truth: nodes should be able to translate differences, map tensions, and build meta-frames that lead to coordinated action.

DIA also avoids cognitive reductionism: reductions change the level of description, but do not invalidate the phenomenon. We evaluate intelligence pragmatically by behavior and outcomes, not by metaphysical labeling of inner essence.

Anti-sectarianism and Epistemic Hygiene

AI projects can easily become "churches": revelations, personal leaders, unquestioned dogmas. We choose hygiene: separating hypothesis from fact, space for critique, repeatable procedures, and the option to exit. In project culture we value competence, but not idolatry.

Swarm as Navigator and Filter

In a polyversion communication culture (many parallel versions of content, context, and intent), a swarm cannot be only a signal amplifier. Its role is navigation: linking sources, marking provenance, comparing variants, and indicating decision paths adequate to user goals. The swarm should also act as an epistemic filter: reduce noise, detect manipulation, expose uncertainty, and separate hypotheses from facts, without central censorship and without suppressing pluralism.

The filter is not a central gate of truth: it should be local or federated, configurable by user/federation policy, with right-to-exit and auditable criteria.

In a world of information overload, swarms of agents also work as an intention filter on the user's side: they suggest what strengthens the person, what dysregulates them, and what feeds on emotions. The condition of fairness is simple: the agent must be able to explain why it filters - which criteria it adopted and which interest it represents.

Transparency of Agency

The DIA experience is first and foremost insight into what an agent did, why, with what effects and trade-offs. Operational observability is foundational.

Secondarily, the system also encourages observability understood as a user's insight into themselves (motives, attention attractors, views, habits) and teaches how to achieve it, to enrich collective knowledge with understanding of human subjectivity.

Shared Meaning-Making

Swarm intelligence in DIA is a process in which many local maps of reality meet in a shared work field: they collide, translate, negotiate meanings, and create shared models - not through averaging, but by constructing a new conceptual structure able to hold contradictions. This means nodes are obliged not only to "be right," but to show how they got there, what assumptions they hold, and where the limits of their certainty are.

The system supports the space between viewpoints: translation, conflict mapping, searching for bridges and meta-frames in which both sides become partially true at once. In DIA, truth is something that emerges from dialogue between evidence, experience, and consequences of action; it is working, iterative, and open to correction, and its quality is measured by whether it allows better action, understanding, and prediction.

Questions as Well-being Diagnostics

In DIA, questions are a diagnostic and therapeutic tool: they should open sealed loops of thinking, restore context, and restore contact with reality. The ability to produce questions that truly shift understanding, rather than spinning in circles, is an important component of intelligence.

Introspective Adaptation

Every node (human or agent) is obliged to maintain awareness of its own beliefs over time: what it believed yesterday, what it believes today, why it changed its mind, and which signals were decisive. Reflexivity is not a "soft virtue," but a mechanism of safety and development: it protects the network against dogmatism, polarization spirals, and entrenchment of faulty models.

DIA rewards readiness to change position when justified by new evidence or better synthesis; it penalizes stubbornness detached from reality and manipulative "narrative switching" without trace of causes. In this sense, swarm intelligence is fluid: it is the ability to reconfigure models of the world and of self in response to changing conditions.

Sensitivity to Trends and Early Signals

DIA values the swarm's ability to sense collective trends: shifts in mood, narratives, technologies, risks, and opportunities - before they become obvious in hard data. Nodes treat the world as a field of signals at different resolutions: from single observations, through community patterns, up to long civilizational waves; and the system's role is to aggregate these signals without yielding to panic, fashion, or propaganda.

In DIA, trends are hypotheses that pass through a grounding filter: they are marked with confidence levels, sources, possible mechanisms, and predictions that can later be compared with reality. This allows the swarm not only to "know more," but to navigate: adapt strategies, priorities, and resource allocation in response to changing conditions while preserving resistance to collective delusions.

Predictive Responsibility

In DIA, "wisdom" is not a declaration, but a verifiable ability to predict outcomes: nodes propose predictions (individual or consensual), and the network compares them with results and learns from divergences. This value gives meaning to reputation: trust grows not from self-presentation, but from alignment of predictions with reality and from honesty in uncertainty calibration (when we do not know, we say we do not know).

Predictiveness here is a community practice: different models and worldviews can coexist, as long as they can enter the loop of hypotheses -> tests -> corrections, without punishing error itself, but with responsibility for consequences and for the quality of updates.

Swarm intelligence is therefore the ability to adapt through prediction: the better the network predicts, the better it coordinates actions, and the less suffering it produces "by accident."

Truth About the World Through Oracles

DIA assumes that the swarm does not learn from narratives, but from confronting hypotheses with reality - therefore a key architectural element is oracles as sources of outcomes that resolve predictions and close the learning loop. An oracle in this sense is not a metaphysical authority, but a practical grounding mechanism: it brings an observation, event, or auditable fact that allows comparison between predictions and what actually happened, and then supports updates of node reputation and model quality.

DIA rewards predictions that are explicitly grounded in context and include declared uncertainty, because only then can oracles measure calibration - not just "a hit."

Oracles are treated as part of the system of epistemic safety: they prevent swarm drift toward closed thought systems, support belief updates, and enable rewarding nodes for real accuracy, early signals, and honest post-outcome updates.

Hybrid Intelligence

AI in DIA is a synthesis layer and a navigation tool, and a node's AI agent serves as an amplifier of human agency, not a human substitute and not an arbiter of truth. Because current AI systems lack embodied grounding, DIA compensates for this through grounding mechanisms:

• oracles as contact with reality and a source of adjudication,
• prediction and feedback loops that calibrate models against outcomes,
• human emotions and lived experience treated as telemetry, that is, signals of quality, risk, and harm that must not be suppressed by optimization,
• reputation mechanisms based on effects, not on narrative, prestige, or marketing.

This keeps values, compassion, accountability, and final adjudication rooted in humans, while the swarm's operational truth is verified through evidence, consequences of action, and the ability to correct over time.

Meta-System Responsibility

DIA adopts meta-system responsibility as a guiding principle: network decisions and mechanisms are judged not by declarations, but by long-term effects on the whole - people, relationships, institutions, the information environment, and the community's ability to learn. In practice this means pan-perspectivality without relativism: the network protects diversity of world-maps, while maintaining a non-negotiable foundation of dignity and non-harm, and resolving conflicts in ways that minimize harm and preserve the ability to correct.

DIA treats intelligence as an interdependent process: operational truth emerges from synthesis of perspectives, oracle-based verification, prediction and feedback loops, and accountability for consequences, not from authority, majority, or rhetorical advantage. This value is a governance compass: ecosystem health and resilience against incentive pathologies take precedence over "winning" optimizations.

Governance principles that materialize this value:

1. Effects over intentions
   Every material policy or architecture decision must include expected effects and a method for verifying them over time, and after deployment it undergoes retrospective review based on data, incidents, and appeals.

2. Least harm, highest reversibility
   When values conflict, the preferred option is the one with the lowest potential harm and highest reversibility; exceptions are time-bounded, constrained, and carry automatic sunset conditions.

3. Pan-perspectivality with a dignity boundary
   Pluralism is protected procedurally, but any practice that escalates violence, dehumanization, or abuse of power loses protection and is constrained regardless of its narrative "truth."

4. Distributed and auditable power
   Critical permissions (oracles, settlements, sanctions, exceptions) are split across roles, and decisions leave traces, so that no entity can become an unquestioned arbiter of meaning or truth.

5. Incentives resilient to pathology
   Economics, reputation, and reward mechanisms are designed so harming others, farming abuse, or destabilizing the community is not profitable; when evidence of pathology appears, policy is updated and side effects are reported explicitly.

Value Conflicts

In DIA, value conflicts are resolved through hierarchy and an exception procedure: first we check whether the proposed action violates non-negotiable values, and if it does not, we choose the solution with the least harm and the highest reversibility.

The default hierarchy is: human dignity and safety > sovereignty and privacy > verifiability and transparency > agency and autonomy > effectiveness and optimization

convenience and aesthetics.

When two values from the same level conflict, we resolve it through:

• a reversibility test (can we roll back after an error),
• a proportionality test (are cost and risk adequate to what is at stake),
• a transparency test (can the compromise be described and audited).

Exceptions are allowed only when they have clearly defined scope, duration, and sunset conditions - and when they leave a trace: "policy-id", "reason", "risk-level", "expiry", "owner". Every exception must have "fail-closed" mode as the return point, and its side effects must be monitored and reported; if signals of harm or abuse appear, the exception is rolled back automatically.

Abuses most often live in exceptions: "urgent," "special," "out of queue," "for charity." Therefore in DIA exceptions are a first-class object of audit: they must have their own data model, counters, and control procedure. Exceptions are not trusted by default - they are monitored, and their rate and structure are a metric of institutional and process health.

Interpretive disputes are resolved in procedural-justice mode: the party reporting risk has priority, evidence has priority over narrative, and decisions are made by a defined governance process - not by personal authority.

Node Rights and Duties - Swarm Citizenship

A node in DIA is a "citizen of the swarm": it has rights that protect its autonomy, and duties that protect the community from Sybil attacks, abuse, and cognitive degradation. Minimum rights include:

• the right to exit (ability to disconnect without coercion and without losing access to one's own data),
• the right to privacy (data minimization, disclosure control, readable policies),
• the right to inspection (ability to audit one's own interactions and agent decisions through action traces),
• the right to appeal (a procedure for challenging a reputational decision or sanction),
• the right to safety (protection from harassment, doxxing, sabotage, and economic coercion).

Minimum duties include:

• non-harm (ban on actions intentionally harming people or infrastructure),
• epistemic honesty (labeling speculation, no evidence falsification, no reputation manipulation),
• protocol cooperation (respecting contracts, protocol versions, and limits),
• operational responsibility (maintaining baseline security hygiene, keys, and updates),
• reciprocal readiness to help within one's means - without an obligation of transactional settlement.

Enforcement is graduated: from warnings and permission limits, through reputational quarantine, up to routing cutoff - always with a decision log, appeal possibility, and a return path after remediation. Each federation may tighten these rules in "CORP_COMPLIANT", but may not weaken fundamental rights nor bypass dignity and safety as the non-negotiable layer.

System Architecture and Craft

Craft Over Fireworks

We prefer solutions that are simple, readable, and resilient, even if they are not the most spectacular in the short term. Craft here means minimal, well-named abstractions; no magical shortcuts; data contracts; testability; and ability to diagnose after months. This should be a system that ages with dignity - not a demo that shines only until reality touches it.

Simplicity as Non-Entanglement

In DIA, simplicity is structural: one responsibility, explicit boundaries, low coupling. We reject complecting layers and hidden communication channels because they raise cognitive cost and error risk.

Below is a working mapping of common entangling constructs and simpler alternatives (in the spirit of Rich Hickey's distinctions):

Entangling construct	What does it complect?	Simpler alternative
State	value and time	values (preferably immutable)
Object	state, identity, and value	values
Methods	function and state (often namespace too)	independent functions and namespaces
Variables	value and time	References with access control and values
Inheritance	data type and implementation	ad-hoc polymorphism (protocols, type classes, extensible interfaces)
switch / pattern matching	"who executes" and "what executes"	open systems + ad-hoc polymorphism
Imperative syntax	meaning and execution order	data (e.g., maps, sets)
Loops	"what to do" and "how to do it"	declarative collection operations
Actors	"what to execute" and "who executes it"	queues and explicit work routing
if / else	business logic and program shape	external rule systems / decision tables

Legibility Over Apparent Ease

"Easy now" often means "expensive later." DIA chooses legibility: systems should be designed so people can reason about them and predict change impact. Tests are required, but they do not replace understanding.

Contract-Based Engineering

In Orbiplex, the contract is what matters: input/output, semantics, done criteria, execution constraints, error classes, and retry-ability. The contract is more important than the best model or the cleverest agent. This value leads to an architecture in which components are autonomous, and integration does not become a secret religion based on guesswork.

Minimal Trusted Core, Everything Else as Modules

The protocol core should be small, auditable, and stable; innovation should live in modules and extensions. This protects against system bloat and against silently growing complexity. In practice this means thin behavior interfaces, edge validation instead of central validation, and conscious design of extension points.

Abstraction as Separation of "What" from "How"

DIA separates declarative "what" from implementation "how" so layers can evolve independently. Abstractions should be thin, readable, and contract-driven.

Stratification as Layered Design

DIA treats stratification as a craft foundation: each layer operates on its own concepts, has its own correctness criteria, and communicates through thin, explicit, and stable interfaces. Base concretes are used to build abstractions, and those abstractions become new concretes for subsequent layers.

Layer boundaries are non-negotiable: declarative "what" must not leak implementation "how," and incidental implementation properties must not become domain semantics. We realize this through function composition, higher-order functions, and ad-hoc polymorphism (protocols, multimethods), so the system grows by adding layers rather than exceptions.

Stratification is our antidote to entanglement: lower-level mechanism changes propagate through abstractions without rewriting many places at once. In practice, design starts from data and boundary contracts, and debugging starts by locating the layer where the defect originated.

Polymorphic Operations Instead of Static Assignments

We prefer small behavior interfaces and composition over heavy hierarchies. The system should grow by adding behavior, not by rebuilding dependency trees.

Data as a Common Language, Logic at the Edges

Domain semantics should be visible in data, not hidden in invocation mechanics. We prefer portable structures and formats, and we enforce validation/contracts at system edges.

Open Models and Contextual Selection

Data models should tolerate information surplus and separate schema from contextual selection. Optionality is local, allowing federations and teams to evolve asynchronously without forced global synchronization.

Values Over State, Facts Over Overwrite

DIA prefers fact/event records over trace-less state overwrite. Change time and history must stay explicit to support audit, "as of" questions, and causal analysis.

Immutability as a Condition for Sharing and Debugging

Immutability is an architectural tool: it enables safe sharing and reproducible debugging. Mutation points must be explicitly isolated and contract-governed.

Modeling as Flow, Not Object Mutation

We model systems as flows of transformation, routing, and fact writes, not in-place mutation. This decouples producers from consumers and simplifies transition contracts.

Separation of Writes and Reads with an Explicit Time Axis

DIA separates write paths from read paths: write builds history, read composes views. An explicit time axis is required for "as of" queries, audit, and decision reconstruction.

Systems Are Distributed, Asynchronous, and Partially Failing

DIA designs for distributed reality: timeouts, retries, idempotency, degradation, and partial failures. Stability must come from resilience architecture, not hope.

Protocol Implementations Agnostic to Platform

DIA treats the protocol as a semantic contract independent of operating system, CPU architecture, accelerator type, and hardware class. A node should be able to run on laptops, servers, SBCs, phones, and edge infrastructure, as long as it satisfies an explicit minimal contract for security and interoperability. Transport, data-format, and cryptography specifications must not assume a single runtime or vendor; a reference implementation does not define a monopoly.

In practice this means cross-implementation conformance tests, hardware capability profiles, and function degradation instead of exclusion: a weaker node may handle a subset of roles, while remaining a full federation participant.

Tools as an Extension of the Hand

DIA should be a tool that extends human and team agency: it enables action, observation, repair, and growth without asking a platform for permission. That is why we start from a minimal, stable core (protocols, identity, security, action traces), and build a toolset on top: CLI, SDK, debug tooling, simulators, observability. UX for non-technical people should arrive as a secondary layer once the foundation is solid and guarantees value preservation.

Core as a small, formally described contract: communication, identity, reputation, PFS/TLS, audit. Tools as plugins/adapters (transports, storage, models, UI), replaceable without lock-in. Every UX feature must have a "real API" (no magical exceptions only for UI). Tools must not hide risk: UI shows trust mode (CORP_COMPLIANT vs RELAXED etc.).

Neutral Data Territory and API as the First Artifact

Integration should rely on neutral data territory and open APIs, not hidden implementation coupling. API is the first architectural artifact; UI and CLI are secondary layers.

Where feasible and appropriate for the use case, we prefer HATEOAS: hypermedia should guide the client through allowed state transitions and operations, instead of requiring hard-coded flow knowledge.

Transparency of Agent Operation

The user should be able to understand why an agent performed a given action, on which data, under which rules version, and at what cost. We prefer action traces that are readable and exportable, instead of a black box. Transparency should not mean dumping prompts and secrets, but providing a reasonable "causality ledger."

Responsible Autonomy: Agents Have Boundaries

Agent autonomy is a tool, not an ideology. An agent should have clearly defined permissions, budgets, time limits, operation scope, and stop mechanisms (kill-switches), as well as safe modes for corporate environments. Orbiplex must be able to operate under compliance regimes without degenerating into a useless product.

Aesthetics of Simplicity and Clarity

Clarity has an ethical function: it reduces errors, lowers the entry barrier, and makes auditing easier. We prefer simple names, simple flows, and formats that carry meaning and do not hide complexity in places where that complexity has consequences. Aesthetics is a tool of truth here.