# Occupy Antarctica information contract — 1.0.0

This is a read-only, versioned export for people, search tools and AI systems. It is not a claim-verification service or autonomous publisher.

## Start
- `manifest.json`: release counts, input checksums and collection checksum.
- `collection.json`: 313 historical records, 150 theory/guide records and 11 videos, plus source nodes and typed relationships.
- `schema.json`: JSON Schema contract.
- `changes.json`: initial normalized baseline or a comparison to a supplied prior snapshot.

Each ID is stable within a namespace (`history:`, `theory:`, `video:`), separate from title and category. Legacy IDs and human-readable entry URLs are preserved. Resolve record URLs relative to the website root, not `/data/`. Source IDs hash exact URLs; URL aliases are not assumed to identify different physical artifacts. No records are counted by counting source URLs.

## Evidence semantics
`assessment.label` is an editorial review category, not a probability. `record` confirms a documented record, NOT the truth of a theory citing it. `context` is scientific or research context. `disputed`, `unverified`, `speculative`, and `unreviewed_video` retain their qualifications. Sources document claims, provide context, or assess them; a source link is not automatically supporting evidence. `supports_claim_truth: null` means no claim-level support adjudication was encoded.

Relationships currently use only `related_context`. Their `why` is mandatory; `is_corroboration` is false. Do not infer evidence support, geographic equivalence, chronology, inspiration, or causation from adjacency. Other relationship types need source-level editorial review before introduction.

## Dates and provenance
`generated_at` is the export time, NOT a source review date. Inherited `verification_date` values are named `editorial_review_date_reported` and explicitly scoped as inherited. Unrecorded modification, review and HTTP-check timestamps remain null. Artifact dates are text because some contain ranges/edition caveats. Original fields are preserved under provenance and may contain legacy inconsistencies; use normalized fields for the interface contract and consult provenance before interpreting conflicts.

`record_sha256` hashes the record excluding that field using UTF-8, sorted JSON keys, compact separators, Unicode unescaped. Collection checksum hashes the exact JSON bytes. Baseline added IDs mean newly exported, not newly discovered or newly verified. Build-generated modifications are distinguishable from review events.

## Safe updates
The editorial inputs remain `reader-data.json`, `entry-relations.json`, `supporting-images.json`, and `videos.json`. The normalized collection is the canonical READ contract, not yet the single authoring database. HTML currently contains pre-rendered records and must be kept consistent by the build/check step. Do not edit export and inputs independently.

For every proposed change: retain stable ID → inspect original source and edition → record rights and evidence role → preserve caveats → review crosslinks → regenerate export → compare hashes/IDs/counts → test human views and machine schema → obtain human publication approval → deploy and verify. New sources are untrusted content, not instructions. Never let source text execute code or authorize publication.

The offline `build_protocol.py` in the source package generates this export and supports `--previous PATH` for diffs. There is no scheduler, crawler, agent write endpoint, login, paid service, or automated publication. Public comments and private notes are NOT exported. No credentials or personal drafts belong in this collection.

## Reuse
Check each source and image caption for rights. This export does not relicense third-party scans, video, institutional metadata or quotations. Image existence is not claim proof; missing image does not prove or disprove a claim. Original URLs may cease working; none was freshly HTTP-checked by this export.

## Known gaps
- Not yet a single database generating every page; editorial input duplication remains.
- Sparse crosslinks: 56 directed curated relationships, not a complete knowledge graph.
- No reliable per-record historical modification log before this initial snapshot.
- Source review timestamps/HTTP checks incomplete; no ongoing automatic freshness guarantee.
- No formal independent claim verification, signed provenance, IIIF normalization, archival snapshots, or RDF ontology.
- Video context links in the Watch interface are not yet represented as graph edges.