The Epstein Record — Sanitized Public Reference

Demonstrated capabilities

A claim ledger with explicit limits.

The system converts heterogeneous records into atomic, reviewable propositions while recording what each source establishes, what it cannot establish, and what would change the assessment.

Provenance

Source-chain separation

Distinguishes independent corroboration from copying, circular reporting, and multiple stories derived from one underlying record.

Classification

Controlled evidence states

Uses a finite state model so adjudicated outcomes, verified records, attributed allegations, disputes, and unresolved claims are not blended together.

Evidence ceiling

No claim beyond the record

Every record states both the proposition supported by the evidence and the stronger conclusions that the same evidence does not support.

Privacy

Independent publication gate

Evaluates necessity, proportionality, lawful handling, public-interest value, and risk to victims or private individuals.

Contradiction

Responses stay with the claim

Denials, alternative explanations, conflicting records, legal responses, and material uncertainty are represented near the proposition they affect.

Corrections

Revision history is preserved

Material changes retain the previous state, new state, reason, date, and reviewer rather than silently overwriting the record.

Evidence model

Seven states, each with a defined burden.

AdjudicatedEstablished through a controlling legal disposition or equivalent official record.

Verified recordDirectly supported by an authenticated primary record.

Supported reportingReliably reported but not independently established by this project.

AllegedAn attributed allegation with clear sourcing and no claim of adjudication.

DisputedMaterially contested by credible evidence or a named response.

UnresolvedA consequential proposition for which the available record is insufficient.

Not establishedThe reviewed record does not support the proposition.

Release pipeline

A claim reaches publication only after ten gates.

Record the source identity, issuing body, date, stable identifier, document type, and handling restrictions.
Reduce the narrative into an atomic proposition with a defined subject, action or relationship, time period, and burden.
Assign and justify one controlled evidence state.
Determine whether apparent corroboration is genuinely independent.
Represent contradictions, denials, responses, and alternative explanations.
Complete privacy and public-interest review.
Check public wording against the evidence ceiling.
State the evidence or event that would trigger revision.
Generate a public-safe export that excludes restricted material.
Run deterministic schema, domain-rule, link, and production-endpoint validation.

Fictional schema examples

How the same system handles support and insufficiency.

Every person, institution, event, and record below is fictional and exists solely to demonstrate the data structure.

TER-DEMO-0001

Verified record

Example Public Official

A contemporaneous public calendar records a scheduled meeting with Example Organization on 2024-01-15.

Evidence ceiling: the record establishes a scheduled meeting entry. It does not establish attendance, knowledge, intent, wrongdoing, or the content of any discussion.

TER-DEMO-0002

Not established

Example Private Individual

An unverified online post alleges misconduct by the subject.

Evidence ceiling: the available material establishes only that an allegation was posted. It does not establish the truth of the allegation. The fictional identity is withheld to demonstrate the privacy rule for unsupported claims concerning private people.

Non-negotiable analytical rules

Association is not culpability.

Appearance in a contact book, calendar, photograph, address record, or flight record is not proof of criminal conduct.
Association, access, location, or communication does not by itself establish knowledge, intent, participation, or responsibility.
Allegations remain attributed and visibly labeled.
Legal outcomes and authenticated official records take precedence over repetition volume.
Privacy harm is weighed independently from public curiosity.
Material corrections preserve a visible revision record.
AI-assisted extraction, summaries, entity matches, and classifications remain provisional until human review.

Publication boundary

Technical availability does not equal publishability.

The public product excludes victim-identifying information, unnecessary private-person data, sealed or illegally obtained material, graphic abuse content, exposed credentials, and unsupported allegations presented without attribution or response context.

Necessity

The record must answer a defined public-interest question rather than satisfy generalized curiosity.

Proportionality

The public wording and disclosed details must be no broader than required to support the claim.

Right of response

Material denials, explanations, legal responses, and contradictory records should appear near consequential claims when feasible.

Inspectable implementation

Audit the schema, rules, examples, and automation.

Claim-ledger schemaDraft 2020-12 machine-enforced public schema. Fictional example recordsTwo public-safe demonstrations of evidence ceilings and privacy handling. ArchitectureSystem components, data flow, public boundary, and release gates. Privacy and correctionsPublication exclusions, contested claims, emergency handling, and revision policy. Automated workflowsSchema validation, domain rules, link checks, dependency review, and live smoke tests. Security policyResponsible disclosure and sensitive-data handling boundaries.