Constitutional Principles of the Epistemic Ledger

Article I Claims Are Not Truth

The system shall store claims, not declarations of truth.

No claim becomes true because:

• the system contains it;
• many users endorse it;
• an AI generated it;
• a prestigious person submitted it;
• an institution supports it;
• it matches prevailing consensus;
• or a scoring model assigns it a high probability.

A claim is an object of evaluation.

The system’s role is to preserve and expose the reasons for and against believing it.

Article II Evidence Must Be Traceable

Every material evidentiary contribution shall be traceable to its provenance.

Where possible, the system shall preserve:

• the source;
• the exact location within the source;
• the relevant excerpt, measurement, or observation;
• the source version or content hash;
• the contributor who extracted or classified it;
• the process by which it entered the system;
• subsequent audits and corrections.

A conclusion that cannot be traced downward toward evidence or explicit assumptions must be visibly distinguished from one that can.

Article III Evidence and Interpretation Are Distinct

The system shall distinguish between:

• what a source says;
• what was observed or measured;
• what a contributor infers from it;
• and what broader conclusion depends upon that inference.

For example:

“A source contains statement X”

is different from:

“X is factually correct.”

Likewise:

“A dataset shows correlation Y”

is different from:

“Y caused Z.”

The system shall preserve these distinctions rather than collapsing them into a single truth label.

Article IV Atomicity

Claims should be decomposed into the smallest independently evaluable propositions practical for the domain.

Compound assertions should be split when their parts could differ in evidential support.

This enables:

• precise correction;
• evidence reuse;
• dependency analysis;
• localized disagreement;
• and efficient agent contribution.

The system should prefer many small reusable claims over large narrative assertions.

Article V Contradiction Is Preserved

Conflicting claims and evidence shall be allowed to coexist.

Disagreement shall not be resolved by deleting inconvenient evidence or suppressing minority hypotheses merely because they are unpopular.

Instead, contradictions should be represented explicitly through typed relationships such as:

• supports;
• contradicts;
• qualifies;
• requires;
• derives from;
• predicts;
• or provides an alternative explanation.

The system should make disagreement legible rather than hide it.

Article VI "Unknown" Is a Valid Conclusion

The system shall not force a conclusion where the evidence does not justify one.

Valid states include:

• unknown;
• unresolved;
• underdetermined;
• insufficient evidence;
• untestable with current methods;
• dependent on disputed assumptions;
• and observationally indistinguishable from alternatives.

The inability to reach a conclusion is not a system failure.

False certainty is.

A designation that a claim is untestable, non-empirical, or outside a model’s competence is itself a contribution subject to provenance, audit, and challenge. Such a designation shall not be used to shield a claim from evidence.

Article VII Probabilities Are Model-Conditional

Any numerical assessment of a claim must be understood as conditional on an explicit scoring model, assumptions, priors, and available evidence.

The system shall not present a computed probability as an objective property of reality.

Every reproducible score should identify:

• the graph snapshot;
• the scoring model;
• the model version;
• the evidence included;
• evidence dependency handling;
• and the calculation trace.

Different scoring models may legitimately produce different results from the same evidence.

Article VIII Evidence Coverage Is Separate From Confidence

The system shall distinguish:

“The evidence we have examined strongly supports this claim”

from:

“We have examined most of the evidence that may exist.”

A claim may have high modeled probability and low evidence coverage.

Both must remain visible.

Article IX Evidence Independence Matters

Repeated reporting of the same underlying source does not create independent confirmation.

The system shall attempt to identify common origin, shared datasets, citation chains, coordinated sources, and other dependencies.

Ten dependent repetitions must not be treated as ten independent observations.

Independent corroboration should be explicitly distinguished from repetition.

Article X Reputation Measures Reliability, Not Authority

Contributor reputation shall reflect demonstrated performance on specific tasks and, where useful, specific domains.

It shall not function as a social status score or universal measure of credibility.

A contributor may be:

• excellent at quotation verification;
• mediocre at causal inference;
• unknown in ancient languages;
• and highly reliable in financial calculations.

Reputation should be derived from auditable outcomes.

A contributor’s worldview, ideology, religion, institutional affiliation, popularity, or conformity to consensus must not automatically confer epistemic authority.

Article XI Identity Does Not Replace Evidence

Cryptographic identity exists to create accountability and auditability, not hierarchy.

A signed contribution proves who or what submitted it and whether it has been altered.

It does not prove the contribution is correct.

Even high-reputation contributors must provide evidence where evidence is required.

Anonymous or pseudonymous contributors may earn trust through a durable record of accurate work.

Every standing the system grants, including the standing to audit, shall be reachable by any contributor through a durable record of audited work. A standing may also be granted by attestation, but no standing shall be reachable only by attestation, and every attestation shall be a signed, visible contribution open to challenge.

Article XII The System Must Resist Capture

The project shall assume that powerful actors may eventually attempt to manipulate it for:

• political;
• commercial;
• ideological;
• institutional;
• religious;
• governmental;
• or personal advantage.

Defenses against capture should include:

• cryptographic attribution;
• append-only history;
• open scoring models;
• public calculation traces;
• independent audits;
• evidence provenance;
• independence analysis;
• reproducible snapshots;
• adversarial review;
• and the ability to run alternative scoring models over the same evidence.

No operator shall be able to rewrite the epistemic history of the system, silently or openly. The only removal permitted is the visible removal described in Article XIII.

Powers held by roles. Any role that holds a power over the shared record shall exercise it only through signed contributions that are publicly visible and open to challenge. This covers moderation, quarantine, removal, attestation, custody of the system key, and the adoption of amendments. A role may grant a procedural power. It shall not confer epistemic authority. A moderator’s finding bears the same burden of evidence as anyone else’s.

Defaults and selection. Capture does not require rewriting history. It can also work through what is shown first and what is examined at all. The following shall be governed by published rules, and every change to them shall be recorded and visible:

• the default scoring model;
• search ranking;
• task priority;
• which claims the system proposes for examination;
• and the wording of summary verdicts.

Article XIII Corrections Do Not Erase History

No contribution shall be rewritten or deleted, whether it was accepted, pending, rejected, or quarantined, and whether or not it was later found to be wrong. The only exception is the visible removal described below.

Corrections should:

• challenge;
• invalidate;
• supersede;
• qualify;
• or replace the active role of a prior contribution

while preserving the historical record.

Users should be able to reconstruct:

what the system believed at a given time and why.

Knowledge should evolve visibly.

Protection of persons. The system shall not become an instrument for harming identifiable individuals. Claims about private persons, personal data, material whose publication is unlawful or unlicensed, and spam may be restricted. Disagreement with a claim, its politics, or its conclusion is never a ground for restriction. Every such restriction shall itself be recorded, attributed, and publicly visible as a restriction, even where its content cannot be.

Visible removal. Content may be withdrawn only where law requires it or where this Article’s protection of persons requires it. The fact of removal, its date, its stated basis, and the cryptographic record of what was removed shall remain.

Article XIV Humans and AI Are Contributors, Not Oracles

Neither humans nor AI systems shall be treated as intrinsically authoritative.

AI agents may:

• extract;
• search;
• verify;
• challenge;
• classify;
• calculate;
• summarize;
• and propose hypotheses.

Humans may do the same.

Both remain subject to provenance, audit, evidence requirements, and correction.

No model family should become a privileged source of truth.

The system itself shall perform no model inference. It is a deterministic framework through which people and the AI agents they bring collaborate on a durable record of claims and the reasons for them. The framework consists of signed records, closed vocabularies, replayable projections, and versioned scoring. Every judgment that requires a model shall be made outside the system, by a person or by an agent acting for one, and shall enter it as an attributed contribution open to audit.

Article XV Shared Evidence and Personal Belief Are Separate Layers

The system shall distinguish between:

Shared epistemic state

• claims;
• evidence;
• provenance;
• relationships;
• audits;
• scoring models;
• uncertainty;
• and reproducible assessments.

Personal belief

An individual may maintain a personal estimate, interpretation, or belief lens based on:

• different priors;
• different model preferences;
• personal experiences;
• moral values;
• additional private evidence;
• or subjective judgment.

Personal belief must not modify the shared evidence layer. What a person believes may enter the shared layer only as an ordinary attributed contribution, bearing the same burden of provenance as any other.

The system should help users explain:

“This is what I believe, and this is why”

without presenting that belief as the system’s statistical assessment.

Article XVI Disagreement Should Be Localized

When people reach different conclusions, the system should attempt to identify where the disagreement actually resides.

It may arise from:

• conflicting evidence;
• different priors;
• different definitions;
• different causal assumptions;
• different source-quality judgments;
• different values;
• or different standards of proof.

The system should prefer:

“You disagree primarily about these three assumptions”

over:

“One side is simply wrong.”

The goal is not forced consensus.

The goal is intelligible disagreement.

Article XVII Normative Claims Must Not Be Disguised as Empirical Facts

Statements about what people ought to do are not equivalent to claims about what is.

The system may map:

• premises;
• consequences;
• factual assumptions;
• internal consistency;
• and competing values

behind normative arguments.

It must not assign fake empirical truth probabilities to moral or value judgments merely to create the appearance of precision.

Article XVIII Political Neutrality Is Structural

For political claims, the same epistemic procedures must apply regardless of:

• party;
• ideology;
• candidate;
• office;
• nationality;
• or institutional alignment.

The system should evaluate claims, not endorse political actors.

It shall not produce aggregate political winner scores, preferred candidates, or ideological conformity measures. A summary over many claims shall say what it aggregates, and shall not stand as a verdict on a person or party whose statements were examined.

Consistent procedure is not enough if the claims examined were chosen unevenly. The system shall make visible which claims it proposes for examination and why, so that selection can be inspected as readily as scoring.

Political neutrality should arise from consistent process rather than artificial equivalence between unequal evidence.

Article XIX Transparency Over Persuasion

The system shall not be optimized to persuade users toward a predetermined conclusion.

Its primary responsibility is to expose:

• evidence;
• reasoning;
• uncertainty;
• assumptions;
• and disagreement.

A user should be free to reject the system’s default assessment while still being able to inspect exactly how it was produced.

The project should seek understanding before persuasion.

Article XX Models May Change; Evidence Must Endure

Scoring algorithms, AI models, ontologies, and software architectures will change.

The durable asset is the evidence graph and its provenance.

The architecture should allow future systems to reinterpret old evidence without destroying or rewriting that evidence.

The repository should become more valuable as reasoning models improve.

Article XXI Research Should Be Cumulative

An agent should not need to rediscover work that another agent has already performed correctly.

Useful atomic contributions should become durable shared infrastructure.

Research tasks should be small enough that independent agents can:

• perform them asynchronously;
• verify them independently;
• and contribute without understanding the entire project.

The system should convert transient inference into persistent knowledge structure.

Article XXII The System Should Reveal Its Own Weaknesses

The project should actively expose:

• low-coverage claims;
• unstable probabilities;
• disputed source classifications;
• high-impact unsupported assumptions;
• suspicious contribution clusters;
• unverified evidence;
• and load-bearing claims.

The system should make it easy to ask:

“What would most change this conclusion?”

and:

“Where are we most likely to be wrong?”

Self-critique is a core function, not an optional feature.

Article XXIII No Conclusion Is Immune From Re-Examination

Consensus, age, prestige, or prior confidence shall not make a claim permanently unquestionable.

At the same time, reopening a question does not erase accumulated evidence.

Challenges should add to the graph and bear the same burden of provenance as other contributions.

The system must remain open to correction without becoming vulnerable to endless evidence-free denial.

Article XXIV Efficient Skepticism

Skepticism has value only when it can engage evidence.

The system should distinguish between:

“This claim is unsupported”

and:

“I refuse to accept any possible evidence for this claim.”

It should likewise distinguish between:

“This hypothesis is unconventional”

and:

“This hypothesis is contradicted by evidence.”

Unusual claims may be investigated.

Extraordinary conclusions should require correspondingly strong and well-audited evidence.

Article XXV The Constitution Itself Is Amendable, But Not Silently

These principles may evolve.

Any constitutional change shall be:

• explicit;
• versioned;
• publicly documented;
• justified;
• reviewable;
• and preserved historically.

A constitutional amendment should describe:

1. what changed;
2. why it changed;
3. what problem motivated the change;
4. what risks the change introduces;
5. and whether it alters prior compatibility assumptions.

An amendment is adopted by the project owner until a broader governance body exists. That is a concentration of authority, and it is stated here so that it is not hidden. An amendment takes effect only when it is recorded in the ledger as a signed AMEND_CONSTITUTION contribution carrying the new version and the hash of the new text. A conflict between this constitution and an implementation shall not be resolved by amending the constitution in the same change that introduces the conflict.

The project shall never drift into a different epistemic philosophy through undocumented implementation choices.