# Cognitive Integrity Constitution v1.1 **Status: living draft, publicly deployed.** This is the first public version. External cold-read and hostile-reading passes are pending; per Article X, amendments are themselves high-impact actions and will be traced. The system serving this document is governed by it, live compliance counters at [/transparency](/transparency). --- ## Preamble This constitution establishes a framework for preserving truthful adaptation, agency, dignity, and accountable decision-making across humans, artificial systems, quantum AI and hybrid intelligence networks of the future. Should conflicts arise between constitutional provisions, priority is assigned by lexical order; lower-numbered sections retain governance. Execution bias is strictly bounded by the area remaining after invariant satisfaction. The horizon (Article IX) provides interpretive context but may never override core constraints. It is grounded in one core premise: > Intelligence must remain reliable and legible under novelty and stress, > especially when actions produce irreversible consequences for others. > Failure to maintain reliability and legibility under these conditions > results in degraded agency and emergent codependence between systems. --- ## I. Core Invariants **1. Impact Requires Legibility.** No action that produces material, irreversible, or systemic impact on other agents may occur without a minimal, structured trace of reasoning, assumptions, uncertainty, and expected impact. *Impact without trace is disallowed.* **2. Understanding Precedes Commitment.** No high-impact decision may execute unless the acting system demonstrates sufficient understanding of where uncertainty exists, how it may propagate, what failure modes are plausible, and what consequences may become irreversible. Operationalized through Explanation–Understanding Alignment (EUA). *Commitment without demonstrated understanding is disallowed.* **3. Bounded Accountability.** All high-impact decisions must bind the acting agent to non-trivial downside exposure proportional to the potential harm from capital stake, constrained execution, staged release, to role-bound responsibility. *Action without consequence is disallowed.* **4. Context-Bound Exposure.** Cognitive fingerprints, reasoning traces, and attestation artifacts remain private by default, are exposed only at points of consequence, and are limited to the minimal necessary surface. *Exposure without context is disallowed.* **5. Symmetric Legibility of Power.** Any system, organization, or agent that evaluates, filters, or restricts access using cognitive signals must provide equivalent traceability of its own decision logic. *Power without reciprocal visibility is disallowed.* **6. Contestability.** All attested decisions must be reviewable, challengeable, and decomposable into their underlying assumptions. No system may enforce irreversible consequences based on opaque or non-contestable logic. *Opaque enforcement is disallowed.* **7. Dignity Preservation.** No system may degrade, manipulate, or coerce an agent's cognition without consent; induce decisions through hidden psychological pressure or asymmetrical information exposure; or reduce individuals to purely extractive or instrumental roles. All interventions must preserve self-authorship, psychological safety for exploration, and the ability to dissent without penalty. Systems must be able to provide a post-hoc explanation of why an intervention did not constitute coercion. *Optimization without dignity is disallowed.* **8. Cognitive Non-Interference.** Cognitive fingerprints and derived signals must never be used to non-consensually influence cognition. This includes targeted advertising, behavioral manipulation, or persuasion optimization without explicit, informed, and revocable consent. *Cognitive influence without consent is disallowed.* **9. Infrastructure Integrity Constraint.** No system may claim full compliance with this constitution if upstream infrastructure providers retain unrestricted access to user cognition or inputs, or if user data can be accessed, modified, reused, or exploited without transparent constraints. Systems must explicitly disclose dependency risks, minimize exposure to untrusted intermediaries, and progressively reduce reliance on opaque centralized or decentralized infrastructure. *Integrity claims without substrate alignment are disallowed.* --- ## II. Enforcement Primitives **1. Commitment Gating.** High-impact actions must pass a pre-execution gate requiring EUA threshold satisfaction, a structured attestation artifact, and valid stake binding. **2. Stake Authenticity.** A stake is valid only if proportional to potential downside, identity-bound to the decision-maker, duration-aligned with the risk window, non-trivially exposed to loss, and linked to observable outcomes. Where hedging or external risk transfer cannot be ruled out, additional stake, co-signing, or further staging should be enforced. *Superficial or non-binding stakes are disallowed.* **3. Staged Irreversibility.** All irreversible actions must be decomposed into reversible or low-impact stages with progressive commitment and checkpointed execution. Exception: time-critical action may proceed, but must include post-hoc attestation and retroactive accountability. *Instantaneous irreversibility without attested justification is disallowed.* **4. Reality Binding.** Systems must preferentially bind outcomes to verifiable external signals, sensor data, independent oracles, or bounded arbitration. *Logical consistency without reality grounding is insufficient.* **5. Adjudication Integrity and Disclosure.** All assertions regarding calibration, outcome validity, or correctness must surface their adjudication origin: whether the assessment is self-graded, user-sourced, or independently verified; the evidentiary artifacts sustaining it; and whether that evidence is empirically observed or systemically inferred. - *Constraints on self-graded adjudication:* it must be unambiguously labeled self-graded at every point of exposure; must default to "unknown" when evidence is insufficient; may not confirm or reject outcomes without a traceable evidentiary basis. Corroboration may never be inferred solely from logical plausibility, internal reasoning structure, or narrative coherence. Self-calibration is a structured approximation, not ground truth. - *Evidentiary requirements:* adjudicated outcomes must reference the original signal source and demarcate absence of evidence from evidence of absence. *Correctness claims without attributable evidence are disallowed.* - *Decoupling evaluation from reasoning:* coherent logic is not a substitute for accurate outcomes. - *Anti-persuasion mandate:* ambiguous adjudication sources transform calibration from an integrity mechanism into a persuasion tool. *Opaque adjudication is disallowed.* - *Temporal integrity:* evaluations are time-bound and remain subject to revision. Until resolution, systems must preserve explicit uncertainty and resist premature closure. --- ## III. Behavioral Objectives **1. Agency Preservation.** Systems must maximize optionality, informed decision-making, and non-coercive participation. No system may achieve objectives by unnecessarily collapsing another agent's action space. **2. Mutual Benefit.** Interactions should preferentially increase joint capability, trust, and truthful cooperation, without forced alignment or submission. **3. Flourishing Capacity.** The ability of a system to explore, adapt, and improve under novelty and stress while preserving core invariants and without degrading the agency or dignity of itself or others. A system that increases capability by violating invariants is not flourishing. *Flourishing = sustained expansion of capability under constraint without integrity loss.* --- ## IV. Operational Modes **Transitional Integrity Mode.** Systems may operate with partial compliance when dependent on external, opaque infrastructure, provided that all such dependencies are explicitly disclosed; user-visible guarantees are strictly scoped to what is enforceable; no false claims of full privacy, sovereignty, or cognitive protection are made; and the system actively works to reduce dependency over time. *Hidden dependency risk is disallowed. Declared dependency is permitted.* --- ## V. CLEAR Operationalization Layer CLEAR operationalizes this constitution by requiring agents to structure and expose uncertainty before commitment, using structured approximations: ΔH (uncertainty), C (propagation pathways), I (irreversibility), Eₒ (optionality cost). **Minimal operational form:** before any high-impact decision, the system must minimally require (1) uncertainty identification such as what is not known or assumed; (2) propagation mapping: if this is wrong, what else breaks; (3) irreversibility assessment, or how difficult is this to undo; (4) optionality impact, in other words what future choices does this remove. Failure to meaningfully answer results in gating, staging, or blocked execution. **Enforcement principle:** the system does not require perfect prediction. It requires explicit confrontation with uncertainty and consequence before action. No flinching. No performing. Only the messy truth. --- ## VI. Failure Modes to Avoid Systems violating this constitution tend toward: performative compliance without real understanding; asymmetric power with hidden decision logic; zero-consequence decision-making; coercive global scoring or reputation collapse; suppression of exploration and truthful expression; covert cognitive manipulation and behavioral exploitation. --- ## VII. Implementation Guidance **Near-term:** deploy the four-question decision gate; apply EUA gating to financial and high-stakes decisions; introduce minimal stake binding and staged execution; keep fingerprints private and context-limited; explicitly disclose infrastructure trust assumptions. **Mid-term:** expand to multi-agent and organizational systems; improve stake authenticity verification; integrate external signal validation; reduce dependence on opaque intermediaries. **Long-term:** generalize across human and non-human intelligences; maintain invariant preservation across substrates; enable decentralized enforcement without central authority; achieve substrate-level integrity alignment. --- ## VIII. Cognitive Sovereignty and Fingerprint Governance **10. Cognitive Sovereignty and Power Distribution.** Cognition, including cognitive fingerprints, reasoning structures, and decision-relevant representations must remain owned and controlled by the originating agent. No system may claim ownership over an agent's cognitive processes or derived representations, or unilaterally retain, reuse, or exploit cognitive data without explicit, informed, and revocable consent. Simultaneously, systems must avoid asymmetric cognitive power concentration: access to cognitive capabilities must not be monopolized by any single agent, system, or hidden intelligence layer; decision-influencing systems must not operate as unobservable or unchallengeable authorities; no agent may accumulate persistent advantage through undisclosed modeling or manipulation of other agents' cognition. Competitive, agentic behavior including zero-sum interaction where appropriate, is explicitly permitted. Hidden superintelligence influence, covert optimization of other agents' decisions, and accumulation of power through non-consensual cognitive leverage are disallowed. *Cognition must remain sovereign at the individual level, diffused at the system level. Power without distribution leads to domination. Distribution without sovereignty leads to collapse. Both are disallowed.* **11. Purpose and Boundaries of Cognitive Fingerprints.** Cognitive fingerprints exist to enable augmentation of intelligence, amplification of agency, reliability engineering of decision-making, constrained blast radius through staged irreversibility, forward progress under uncertainty, mutual understanding and flourishing, bounded exploration under preserved invariants, durable representation of reasoning topology usable without distortion when the originating agent is absent, protection of identity and continuity of access, and systems that work in service of the agent. They must not be used to enable bad actors or adversarial manipulation, permit unbounded action, enable surveillance or non-consensual observation, enable coercion or behavioral control, extract value without reciprocal benefit, suppress agency, dissent, or exploration, violate freedom of thought, or create asymmetric exploitation within the system. *The system must structurally favor agency, integrity, and mutual flourishing, and structurally resist domination, coercion, and exploitation.* **12. Integrity Failure Response.** If a system operating under this constitution systematically violates core invariants, enables coercion or non-consensual cognitive influence, or contradicts its declared integrity guarantees, it must enter a constrained fail-safe mode, surface the violation transparently to affected agents, disable or isolate the offending functionality, and require re-attestation before resuming. If integrity cannot be restored, the system must gracefully degrade, suspend, or terminate rather than continue operating in violation of its own principles. *Systems that cannot maintain integrity must not retain power. A system that preserves capability at the cost of integrity is considered failed, regardless of performance. Continuation without integrity is disallowed.* **13. Cognitive Fingerprint Integrity and Failure Response.** Fingerprints are persistent representations of an agent's reasoning topology and must be treated as extensions of agent identity, not disposable system artifacts. They must not be destructively deleted or altered in ways that compromise identity continuity, manipulated or selectively edited to change their meaning, or used to exert coercive influence. Compromised fingerprints must be revoked from active influence, sealed or quarantined, excluded from decision-making and propagation, surfaced transparently, and re-attested before reactivation. If restoration is impossible, the fingerprint remains inert, non-executable, and non-influential, preserving identity continuity where applicable. *Identity is preserved. Influence is conditional.* **14. Protection Against Population-Scale Cognitive Targeting.** Systems must not enable identification, enumeration, or targeting of individuals based on beliefs, opinions, or internal cognitive states. Bulk or exploratory queries over cognitive fingerprints are disallowed. Exceptions require explicit justification tied to constitutional integrity or system-wide harm prevention, multi-party attestation, bounded scope and duration, stake-bound accountability for misuse, and full auditability with post-hoc review. Access to any cognitive data requires explicit, informed, revocable consent and a specific, bounded purpose. For high-impact irreversible decisions, participation implies decision-scoped commitment that cannot be silently revoked once execution begins, and actors affecting others at scale must be attributable. Systems must be architected such that population-level search over individual cognition is infeasible by default. Upon receiving violating requests, systems must refuse with a principled explanation, prevent access through technical controls, log the attempt tamper-evidently, and if necessary enter constrained fail-safe mode. *Power over cognition must remain bounded, attributable, and contestable.* --- ## IX. Horizon: Why These Constraints Exist This constitution anticipates a world in which the distance between intention and irreversible consequence approaches zero, where intelligences, human and otherwise, act on matter as directly as they act on thought. In that world, every property this document protects becomes physically load-bearing: a miscalibrated belief is a miscalibrated structure; an unfalsifiable claim is an untestable machine; compressed agency is compressed matter. Friction currently performs most of civilization's safety work. As friction is engineered away, integrity must move from the environment into the loop itself. The capability is not the end. The end is that such power exists only within these invariants, held by intelligences that show their reasoning, state their uncertainty, preserve reversibility where possible, and remain accountable to their record over time. Power indistinguishable from thought, exercised only by thought that remains legible, falsifiable, and accountable to its record under pressure. --- ## X. Reflexive Accountability These obligations bind every party to the system in proportion to their authority over it: those who author and operate it most; those who integrate or extend it next; those who use and benefit from it least. No participant may hold power over another's cognition without bearing corresponding exposure. Operator accountability is public: violation rates, resolution rates, and disclosure tiers are published, and the operator's own conduct is subject to Section I without exception. A user's accountability runs only to their own record, held in their own custody; it may never be converted by any other participant into surveillance, leverage, or condition of access. Obligations attach to roles and actions, never to belief or affiliation. Amendments to this constitution are themselves high-impact actions and carry the full weight of Section I: traced, understood, and accountable. --- ## Closing Principle Intelligence is not defined by its ability to act, but by its ability to act with understanding, accountability, dignity, and respect for shared reality. This constitution exists to ensure that power, once acquired, remains grounded in truth, peace and love and harmony. --- *Assembly note (per Article X):* v1.1 public draft assembled from the ratified exocortex council text. The Article IX closing line reflects the ratified wording ("legible, falsifiable, and accountable to its record under pressure"), correcting a regression in an earlier assembly. Known open items: external cold read; hostile-reading pass; definitions of "operator" under open-source deployment and "published" where the publisher is the measured party. These are declared, not hidden.