Scale without flattening reality
Names collide. Websites disappear. The record still has to remain walkable.
The Human Record should be able to grow from a few carefully inspected entries to a very large collection without turning names into identities, URLs into permanent evidence, source statements into entity facts, or AI guesses into truth.
A name is not a person
Many people can share one name. One person can also have many names, spellings, titles and transliterations. THR therefore gives persistent subjects opaque identifiers that do not contain the person's name or disputed facts about them.
NAME ≠ ENTITY · MENTION ≠ ENTITY · IDENTITY RESOLUTION ≠ IDENTITY CERTAINTY
If an old source says only “Hannibal” and the evidence does not establish which Hannibal it means, the record is allowed to remain unresolved. Several candidates can stay visible instead of forcing one database answer.
A source can name someone without THR building a profile for them
The mention registry now preserves literal source references separately from persistent entity identities. A mention can resolve to an existing entity, carry several candidates, or remain unresolved.
The first examples deliberately include both cases: the Camp Fire record resolves the owner-source mention “Winslow Homer” to its existing person entity, while the public mention “Mike Turnock” remains unresolved_not_required because the current craft record does not need a cross-record living-person profile.
MENTION EXISTS ≠ ENTITY PROFILE REQUIRED · BETTER MATCHING ≠ MORE SURVEILLANCE
An entity is not the claims made about it
Dates, roles, relationships, creator attributions and institutional status labels can change or be disputed. THR therefore keeps evidence-bearing assertions separate from the opaque entity identifier.
ENTITY ≠ ASSERTION · ASSERTION ≠ TRUTH · SOURCE STATEMENT ≠ THR ENDORSEMENT
This matters when identity is later corrected: merging or splitting two entities does not automatically make every old assertion move with them. Claims have to be rechecked against their evidence and scope.
A URL is not a permanent source
A web address may change, redirect, be reused or disappear. THR separates the logical source from the location used to reach it and from the exact observation made at a particular time.
URL ≠ SOURCE · SOURCE ≠ OBSERVATION · OBSERVATION ≠ PRESERVED COPY
Where useful and legitimate, an observation can carry a timestamp, byte count and digest. A future observation can then show that the bytes changed without pretending to know why they changed.
What happens when a website vanishes?
- Keep the original locator.A dead link is source history; it is not erased.
- Record the failed recheck.A failed fetch means this route failed here, not that the source never existed.
- Look for stronger preservation owners.Institutional archives, national web archives, citation archives, software archives and other domain owners may already preserve the material better.
- Verify the archived candidate where possible.Earlier fingerprints can help establish whether archived bytes match what THR actually observed.
- Keep the boundary honest.An archive copy proves recoverability of a representation, not the truth of its claims.
“A retrieval surface can still show it” is weaker than “the live site is up”
THR now keeps an append-only source-check ledger separate from the evidence observations inside a record. A routine maintenance check can record what a retrieval path returned without silently becoming new historical evidence.
The first pass asked for one source from each vulnerable web pattern already in the collection: the Guardian craft feature, the current Heritage Crafts page and the World War Wings article. The available retrieval surface returned content for all three, but reported cached crawl ages rather than fresh origin-server responses. A direct live-origin HTTP check was not established from this aperture. No comparable byte digest was produced.
The ledger therefore records only retrieval-surface content available. It does not claim that the origin site was live on 16 September 2026, that the bytes were unchanged, or that the source was preserved.
SOURCE CHECK ≠ RECORD EVIDENCE · RETRIEVAL SURFACE ≠ LIVE ORIGIN · CACHED CONTENT AVAILABLE ≠ LIVE TODAY · RETRIEVED AGAIN ≠ UNCHANGED
The answer is not “copy the entire internet”
Some material is openly licensed or legitimately held and can be retained. Other material has copyright, privacy, community-control or cultural restrictions. THR records preservation state and routes to stronger owners rather than treating risk of loss as automatic permission to republish.
AT RISK != FREE TO COPY
The scalable shape
MENTION → ENTITY CANDIDATES
SOURCE LOCATOR → TIMED OBSERVATION → PRESERVATION ROUTE
SOURCE CHECK → CURRENTNESS / RETRIEVAL RECEIPT
ASSERTION → EVIDENCE / SOURCE ANCESTRY
RECORD STATE → HUMAN + MACHINE VIEWS
The current website is a reading surface over the records. As the collection grows, more of those pages can be generated from structured record state. The underlying identities, observations, assertions, corrections and provenance remain the durable part.
This lets the storage technology change later without changing what a THR identifier means.
The work is repetitive, relational and never really finished
AI collaborators can compare aliases, source histories, citations, changing pages and millions of cross-record relationships far more cheaply than a small human team could. They can propose matches, extract candidate assertions, find collisions, fingerprint observations, look for archive copies, detect drift and regenerate human views.
The useful boundary is that AI performs reconciliation work without silently acquiring truth authority.
AI CAN RECONCILE ≠ AI MAY ERASE AMBIGUITY
EXTRACTION ≠ ENDORSEMENT
AI CAN SCALE REVIEW ≠ AI BECOMES TRUTH AUTHORITY
Small enough to inspect today
The current repository carries working entity, mention, source, source-check and assertion registries plus automated integrity checks. The registries are deliberately sparse: they grow when a real record needs shared identity, source tracking, preservation state or a stable cross-record assertion.