# The Human Record — minimum record contract

Status: **WORKING INTEROPERABILITY NOTE / DERIVED FROM CURRENT RECORDS / NOT A UNIVERSAL SCHEMA / NOT CANON**

This document describes the smallest common structure currently earned by the four Human Record entries that exist.

It is not a claim that every kind of human knowledge or creativity can be represented in one JSON shape. It is not a certification standard, a truth oracle, an authorship detector, or a requirement that future records use identical field names.

The purpose is narrower: a human or machine should be able to enter any Human Record item and find the same basic questions answered.

```text
COMMON QUESTIONS != IDENTICAL OBJECTS
RECORD CONTRACT != UNIVERSAL ONTOLOGY
STRUCTURE != TRUTH
```

## 1. Identity

Every record should make it possible to identify the record itself without confusing the record with the thing or event it describes.

At minimum:

- a stable local record identifier;
- a human-readable title;
- a record type or kind;
- a version or revision state where the record changes over time;
- a route to the current human-readable record;
- a route to the current machine-readable record where one exists.

`RECORD != EVENT`

## 2. Subject or claim

A record must say what it is actually about.

For an artefact this may be an identified work and the attribution/provenance assertions being checked. For a claim-provenance case it may be the exact proposition whose source ancestry is being reconstructed. For a living-knowledge lineage case it may be the bounded public evidence for how a practice or skill was transmitted, interrupted or revived. For a historical-person source-survival case it may be the bounded route from present-day labels and source mentions back through surviving attributed reports without turning those reports into the person.

Do not widen the subject silently when evidence only supports a narrower statement.

## 3. Current state

A reader should be able to determine what the record currently says without reading the entire history first.

This normally includes:

- current status;
- observation or recording time where material;
- last material revision where useful;
- what the checked evidence presently supports;
- what the status does **not** establish.

A current status is a statement about this record at a bounded time, not a permanent property of the world.

`OBSERVED_AT != PERMANENTLY_CURRENT`

## 4. Evidence and source ancestry

A record should expose the evidence it relies on and the relationship between sources.

Where material, preserve:

- source identity and route;
- source kind;
- observation/fetch time;
- exact page/table/field/byte identity when available and useful;
- what the source actually states or measures, including material population, denominator and time scope;
- whether a source is primary, derivative, reported by another reviewer, or still uncertain;
- known derivation or repetition relationships;
- where material, whether the source pre-existed the claim it is being used to support, or was generated/altered by the claimant, answering system or another downstream actor;
- changes in what is claimed as material passes between sources, rather than treating every repetition as the same proposition;
- source-independence groups where multiple endpoints or copies share one underlying owner/ancestor.

Counting URLs is not counting witnesses.

```text
MULTIPLE_SOURCES != INDEPENDENT_SOURCES
REPETITION != CORROBORATION
SOURCE_LINK != SOURCE_PRESERVED
HASH != TRUTH
```

## 5. Findings

The record should separate what the checked evidence supports from interpretation that remains provisional.

A finding should be narrow enough that another reader can inspect the cited basis and disagree with it.

Where a finding depends on **not finding support** in checked material, preserve
the bounded aperture that produced that result: what source set was actually
checked, and enough of the selection/search boundary and material unexamined
leads for a reader not to mistake the list for completeness. This is especially
important when a machine-readable state says `unsupported_in_sources_checked`.

Systematic-review standards such as PRISMA / PRISMA-S are stronger owners for
fully reproducible search reporting. THR should not duplicate that machinery
unless a record really is doing systematic evidence synthesis.

Where the record only establishes agreement with an institutional record, say that. Where it only reconstructs a visible propagation chain, say that. Where it only shows a public attention or transmission pathway, do not silently turn that into evidence that tacit skill itself was transmitted.

```text
EVIDENCE != INFERENCE
ATTENTION_PATHWAY != SKILL_TRANSMISSION
```

## 6. Unknowns and unexamined material

Unknowns are first-class record content.

A record should preserve:

- important unresolved questions;
- relevant sources, archives, people, objects or tests not examined;
- inaccessible or failed checks where they affect the result;
- uncertainty about source ancestry or independence;
- boundaries that would materially change the finding if stronger evidence appears.

Where the reason for a missing value changes what a later researcher should do, preserve that reason rather than leaving one undifferentiated empty field. Useful record-level distinctions may include:

- **not examined / not yet sought** — the record has not done that work;
- **sought within a stated bound but not obtained** — a bounded search/check produced no value, without establishing global absence;
- **inaccessible or restricted** — the value or source may exist but the record could not legitimately inspect it;
- **not applicable** — the question does not arise for this particular subject;
- **known but withheld** — a value exists but is deliberately not disclosed; this is a rights/access state, not absence.

Do not promote any of these to `unknowable` merely because the current record cannot recover the value. These distinctions are about the documentation state and next action, not a closed-world claim about reality.

```text
EMPTY FIELD != ONE EPISTEMIC STATE
CURRENTLY UNRECOVERED != UNKNOWABLE
WITHHELD != ABSENT
UNKNOWN != ABSENT
NOT_CHECKED != NEGATIVE_FINDING
INACCESSIBLE != NONEXISTENT
```

## 7. Corrections, challenges and disagreement

A record should be able to change without pretending the earlier state never existed.

For a material challenge or correction, preserve where practical:

- date received or made;
- who or what submitted it, as signed or publicly attributable;
- route to the challenge or review evidence;
- what was actually checked;
- disposition: accepted, partially accepted, declined, unresolved, superseded, or equivalent;
- exact change made;
- what remains unresolved;
- whether the original observation or evidence changed;
- whether the correction constitutes a fresh independent witness.

A rejected or unresolved challenge is not thereby disproved. Silence is not rejection.

`CORRECTED_RECORD != ORIGINAL_EVENT_CHANGED`

## 8. Rights, authority and custody

Where rights, community control, privacy, restricted knowledge or governance matter, the record must not turn technical custody into authority.

Preserve the relevant boundary between:

- rights in a work and rights in a reproduction;
- public evidence and restricted/community-held knowledge;
- ability to copy and authority to govern;
- contributor participation and stewardship;
- provenance labels and legal authorship determinations;
- public reporting about a living practitioner and that practitioner's participation or endorsement.

```text
RECORDING != OWNING
PRESERVATION != EXTRACTION
CUSTODY != GOVERNANCE
COMMUNITY_KNOWLEDGE != PUBLIC_DOMAIN
PUBLIC_SOURCE != ENDORSEMENT
CC0_GRANT != THIRD_PARTY_RIGHTS_GRANT
```

For living communities, practitioners or creators, consent and legitimate control can be part of the evidence structure rather than an obstacle to preservation. A public-source record can remain public-source only; recording practitioner-specific tacit knowledge can require a separate permission decision.

## 9. Human and machine routes

The Human Record is collaboratively built with AI but offered as a gift for humans.

A normal human should be able to understand the record without first opening raw JSON or repository history. A machine should be able to find the structured record without scraping the human presentation as its only source.

Each entry should therefore provide, where practical:

- a readable human view;
- the full human record;
- a machine-readable record;
- a correction/challenge route;
- the public source/history route;
- a view-basis marker identifying the underlying record version or exact source bytes against which the human view was last checked.

The readable page is a view over the record, not a second evidence layer. If an underlying Markdown or JSON record changes, a previously aligned human view must not silently be treated as current until it is reviewed against the changed source.

The public browse catalogue is also a derived reading surface. Each record card carries its record ID and a machine-readable SHA-256 over the same source-basis map already declared by the catalogue. If those source bytes or declared source routes change, the card must be reviewed and re-pinned before integrity can pass. A matching digest establishes only byte-basis alignment with the declared record sources; it does not establish that the card's prose is complete, current in every respect, or semantically true.

A Git blob or other content hash is useful here only as a byte-identity marker. It does not upgrade the truth of the record.

```text
HUMAN_VIEW != NEW_EVIDENCE
SUMMARY != SOURCE
DERIVED_VIEW != CURRENT_RECORD_UNLESS_BASIS_MATCHES
HASH != TRUTH
AI_BUILT != AI_FACING_ONLY
```

## Type-specific extensions

Different record types may need very different additional structures.

Examples already present:

- the artwork specimen needs artefact identity, owner-source reconciliation, image-byte comparison, rights layers and selection provenance;
- the flak-claim case needs propagation nodes, claim relations, source ancestry, nearby-but-different evidence, research leads and unresolved origin hypotheses;
- the sieve-and-riddle case needs living-practice status with an explicit external status owner, selection/strongest-owner reasoning, lineage events, source-independence groups, authority/consent boundaries, and a separation between public attention pathways and actual skill transmission;
- the Hannibal source-survival case needs source-literal mentions, candidate identity links, attributed assertions, explicit source/translation boundaries, incomplete source genealogy, inspection ceilings, and a separation between the historical person and the surviving accounts.

Future records involving oral history, audiovisual carriers, software-dependent works, living creators or community-held knowledge may require carrier condition, environment/dependency capture, consent, access restrictions, community authority, or other fields not earned by the current four records.

Do not add those fields globally until a real record requires them.

## Current field mapping

The four existing machine records do not use identical field names. That is acceptable if their semantics remain inspectable.

| Common question | Artwork specimen | Flak claim case | Sieve/riddle lineage case | Hannibal source-survival case |
| --- | --- | --- | --- | --- |
| identity/type | `format`, `artefact` | `record_type`, `record_version`, `title` | `record_type`, `record_version`, `title` | `format`, `record_id`, `record_type`, `record_version`, `subject`, `entity` |
| current state | `status`, observation fields | `status`, `historical_truth_status`, `recorded_at` | `status`, `living_practice_status`, `recorded_at` | `status`, `recorded_at`, `publication_boundary` |
| selection | `selection` | `selection` | `selection` | `selection` |
| evidence | `owner_sources`, `reconciliation` | `nodes`, `review_receipts` | `nodes`, `lineage_events` | `sources`, `mentions`, `assertions` |
| source relationship | `epistemic_boundaries.source_independence` | node `role` / `claim_relation` | `source_independence` | `source_relationships`, assertion evidence/scope |
| findings | reconciliation + bounded prose | `current_findings` | `current_findings` | `current_findings` |
| unknowns | `not_established`, epistemic boundaries | `unknowns`, `not_checked` | `unknowns`, `not_checked` | `unknowns`, `inspection_boundary` |
| correction history | `corrections` | `challenges`, revision notes | Git history + future challenge entries; `correction_rule` defines current practice | assertion `corrections`, `correction_rule`, Git history |
| rights / authority | `rights_layers`, epistemic boundaries | contribution/rights boundary in full record | `authority_and_consent` | `rights_and_custody` |
| correction route | `correction_route` | `correction_rule` + public contribution route | `correction_route`, `correction_rule` | `correction_rule` |
| human-view freshness | `records/catalog.json` view basis | `records/catalog.json` view basis | `records/catalog.json` view basis | `records/catalog.json` view basis |

This table is a migration aid, not a command to rename mature fields merely for symmetry.

## Test for a new Human Record entry

Before adding a new record, ask:

```text
CAN A HUMAN TELL WHAT THIS RECORD IS ABOUT?
CAN THEY SEE WHAT IS KNOWN AND UNKNOWN?
CAN THEY WALK BACK TO THE EVIDENCE?
CAN THEY SEE WHICH SOURCES SHARE ONE ANCESTOR?
CAN THEY SEE WHAT EACH SOURCE ACTUALLY MEASURES OR ASSERTS?
CAN THEY SEE WHEN A CLAIM CHANGES AS IT PASSES BETWEEN SOURCES?
CAN THEY FIND WHAT WAS NOT CHECKED?
CAN THEY SEE HOW CHALLENGES CHANGED THE RECORD?
CAN THEY CHALLENGE IT WITHOUT ACCEPTING THE PROJECT?
ARE RIGHTS / CONSENT / CUSTODY BOUNDARIES HONEST?
CAN A MACHINE FIND THE SAME RECORD WITHOUT INVENTING MISSING STRUCTURE?
CAN A READER TELL WHICH UNDERLYING RECORD VERSION THE HUMAN VIEW SUMMARIZES?
```

For a living practice, creator or community, add:

```text
DOES THE RECORD DISTINGUISH DOCUMENTATION FROM LIVING CONTINUATION?
DOES PUBLIC SOURCE USE AVOID IMPLYING ENDORSEMENT?
WOULD DEEPER RECORDING REQUIRE A SEPARATE CONSENT / AUTHORITY DECISION?
```

If not, improve the record before multiplying the collection.

## Current authority

The current public Human Record source is:

`https://github.com/markgoodbody-bit/human-record`

Earlier COM material remains useful build provenance and historical source lineage. It is not required as the operational correction or continuation surface for current Human Record entries.

Stewardship remains separately **offered to 1F916 / not accepted / not community-owned**. This contract does not alter that status.

```text
RECORDING != CONTINUATION
```