# Public research record

> **Method-version note.** The requirements below began under v1. Wherever a
> v1 artefact name differs from the future-native v2 schema, v2's validated,
> edition-scoped equivalent takes precedence. Withdrawn catalogue data is
> exposed only as a clearly labelled historical archive, never as the default
> forecast.

The site will publish enough structured evidence for a reader to understand,
challenge, and reproduce the method without presenting private model reasoning
as evidence.

## Public records

- Forecast contract and resolution rules.
- Source register with original and linked language, translation method,
  access date, geography, evidence type and limitations. A field inferred from
  document scope or absent from the original research block is never presented
  as exact: `metadataStatus: qualified` and machine-readable caveats preserve
  that gap, while every link still receives a visible conservative limitation.
- Driver and scenario atlas.
- Historical backtest, naive baseline, mapping decisions, scores, and known
  contamination limitation.
- Decision ledger, including rejected alternatives.
- AI disclosure: the run-level `AI_RUN_LOG.md` records model precision and its
  limits, tools, verbatim where retained or explicitly summarised/reconstructed bounded task prompts,
  source boundaries, human decisions, outputs and fresh-critic separation.
- Gauntlet status, concise blind verdicts, closed gaps, and regressions.
- Immutable edition history and human-readable changes.
- Downloadable structured forecast and source data.

The site never mixes these records across editions. Each edition carries a
`public-record/_MANIFEST.json` that hashes every exposed research, method and
process file, and the edition manifest binds that index. An unknown edition
shows no forecast or research cards. Shared working documents can evolve for
the next refresh but are not served as the record of an earlier edition.

## Source states

- `used`: materially supports a published claim or forecast.
- `context`: shaped the research but does not directly support a claim.
- `contradictory`: materially challenges another used source or inference.
- `rejected`: reviewed but unsuitable, with the reason retained.
- `superseded`: replaced by a later authoritative source without being erased.

## Copyright and translation

The project publishes original summaries and short compliant excerpts only.
It links to original material. Machine- or AI-assisted translations are labelled
and never presented as official translations.