1 · Identity
What record is this?
A record needs its own identity, title, type and current version or revision state. The record is not the event, artwork, claim or practice it describes.
Record anatomy
How the records work
Different things need different records. The common promise is not one rigid format. It is that a reader can see what is being claimed, walk back to the evidence, find the unknowns, and see how the record changed.
The Human Record currently contains four deliberately different kinds of entry: an artwork-provenance specimen, a source-ancestry case for a historical claim, a transmission-lineage case for a living craft, and a historical-person source-survival record. They do not have the same evidence or the same internal shape. They should still answer the same basic questions.
Who builds and checks this?
Mark originated the project and directs its purpose. Framework is the AI coordination role; Codex and Claude Code contribute implementation and technical review. AI work includes research, drafting, code and proposed corrections. These roles describe how the project is built, not independent witnesses for its claims.
Mark's direction does not mean he personally verified every statement. Nor does agreement between AIs establish that a claim is true. Each record must show its own sources, checks and limits. The change proposals and repository history expose parts of the build process, but a shared account name does not identify every contributor and the history is not a complete transcript.
Editorial control has not passed to an independent body or the Square. See current custody and the stewardship offer. You can challenge a record without accepting the project or joining it.
2 · Subject
What exactly is being claimed?
The subject should be narrow enough that evidence can actually bear on it. A record should not quietly widen a claim beyond what was checked.
3 · Evidence
Where does this come from?
Sources should be identifiable and their relationship visible. Two URLs controlled by one institution are not two independent witnesses. A repeated claim is not new corroboration. Where it matters, the record should also preserve what a source actually measured or asserted, including population, denominator and time scope.
4 · Current state
What can we say now?
A reader should be able to see the bounded current finding and the time at which relevant checks were made. A current status is not permanent truth.
5 · Unknowns
What do we still not know?
Unknowns and unexamined material are part of the record, not embarrassing gaps to hide. Not checked does not mean absent or false.
6 · Corrections
What changed, and why?
A challenge should be able to change the current record without erasing the earlier state. Accepted, partial, declined and unresolved outcomes should remain inspectable.
7 · Rights and control
Who may copy, publish or govern this?
Technical custody is not authority. Rights in a work can differ from rights in a reproduction. Recording a living practice or community knowledge does not make it public property or imply endorsement.
8 · Routes
Can humans and machines both inspect it?
Humans should get a readable view. Machines should get a structured route. Both should lead back to the full record and its correction history.
9 · View freshness
Which record version does this summary describe?
The readable page is derived from fuller Markdown and JSON records. Each current human view names the exact Git blob identities it was aligned to. If those source bytes change, the summary must be reviewed again before its freshness is assumed.
A readable page can become stale even while every sentence still looks plausible
The machine catalogue stores the exact source-record blob identities behind each human view. The same basis is shown at the bottom of the readable page. This does not certify the record: a matching hash only establishes which bytes the summary was checked against.
DERIVED_VIEW != CURRENT_RECORD_UNLESS_BASIS_MATCHES
Before another record is added
- Can a normal reader tell what the record is about?
- Can they see what is known, unknown and not checked?
- Can they walk back to the evidence?
- Can they tell which sources are independent and which share an ancestor?
- Can they tell what each source actually states or measures?
- Can they see when a claim changes as it moves between sources?
- Can they see how a correction changed the record?
- Can they challenge it without accepting this project?
- Are rights, consent, custody and governance boundaries stated honestly?
If not, improve the record before multiplying the collection.
The common contract stops where reality becomes different
An artwork may need owner attribution, image-byte identity and reproduction-rights layers. A viral claim may need a propagation graph and source-ancestry relations. A living-practice record may need a strongest-owner boundary, practitioner/consent limits, attention-versus-skill-transmission separation and an explicit reminder that documentation is not the practice itself. A historical-person record may need source-literal mentions, thin entity routing, attributed source reports and an explicit separation between the person and the biography that survived.
Future oral-history, audiovisual, software or creator records may need carrier condition, dependency capture, restricted access or other fields. Those fields should be added when a real record earns them, not because a universal schema would look tidy.
The same discipline has to survive repeated names and disappearing sources
At larger scale, names cannot be primary keys and URLs cannot be treated as permanent evidence. THR now separates mentions from persistent entities and source locators from timed observations and preservation routes.
The human explanation is in How The Human Record scales. The underlying working models remain inspectable and correctable rather than becoming a hidden backend ontology.
Browse the current recordsRead the working record contractHow it scalesMachine catalogueChallenge or contribute
The working contract is descriptive of the current project, not a certification standard or claim that all knowledge can be represented in one format.