# Evidence Atlas methodology and reproducible refresh

This selected discovery catalogue is not the Cambridgeshire HER or a county-wide sample. Original records were assembled on 30 September 2026 from public discovery snippets; access to full ADS pages was blocked during that work. Source-level primary verification remains partial. Do not describe every field as freshly primary-verified.

## Inclusion and evidence

Retain a stable public identifier, authoritative URL, bibliographic provenance and the source's access/reuse statement. Do not infer findings, dates or coordinates from place names. Record source verification separately from dataset release dates. Missing values remain null or empty; records with no grid reference remain searchable but unmapped. A published grid reference is not a site polygon or accuracy guarantee.

No report PDF or CHER record is redistributed. The index does not grant blanket reuse rights; follow each source's terms, including records whose reuse licence is unconfirmed. Source organisations are credited investigators, not CDA commissions.

## Refresh

From the repository root, run `bun run research:build` using the included raw record file and source manifest. Run `bun run research:check` to check committed outputs without changing files, and `bun run research:test` for regression checks. Field definitions come from `src/lib/atlas-fields.json`; CSV serialization/filter rules are shared with the website. The offline generator never runs inside a page request.

Before a substantive release: individually verify eligible new evidence and rights; record blocked sources without bypassing access controls; update version and date_modified in the manifest only for meaningful release changes. Preserve date_published and original source_verification dates unless new checks actually occur. Generate version-specific CSV/JSON, site data, dictionary and public manifest/methodology. Keep previous release downloads immutable. Inspect output parity, duplicates, spatial labels and source links. Publish only with separate approval.

## Source recheck

At each substantive release and at a proposed quarterly review, check authoritative URLs and metadata using permitted access. A 403 is an access limitation, not proof a record is gone. Record outcome, method and date for each attempted check; request human/source-owner verification when access is blocked. Never claim a new links-verified date from an unsuccessful fetch.

## CSV and JSON

CSV contains one row per record, quoted values, empty cells for nulls and semicolon-separated period terms. JSON has release metadata, sources, a records array and catalogue statistics. No WGS84 latitude/longitude is generated. Record keys and field definitions are documented in the public dictionary. An empty search does not establish absence of archaeology.