W1 address pour — extract-v1 candidates: 715 figures, 52 events, 638 places, 187 peoples, 26,045 attestations
receipt:addresses-w1 · v1 · ingest_verified
sha256 2f420db81f62cc470ea12713f3bc91e7b946443011d31b3e3bfcee7254d1bda2
A receipt records what the house consulted and what it did not establish. The open state is part of the record, not a defect in it.
- Perseus Digital Library — Herodotus, The Histories (Godley translation)A. D. Godley, Loeb Classical Library, Harvard University Press / William Heinemann, 1920–25license: public-domain (US: pre-1930 publication); Perseus digital edition CC BY-SA 4.0, attribution recorded in ops/corpus-staging/SOURCES.mdpinned: github.com/PerseusDL/canonical-greekLit @ d121d1539cc37bfc9274bb81c6b6ec773111f098 (sparse checkout data/tlg0016, fetched 2026-07-07); file tlg0016.tlg001.perseus-eng2.xml; per-file sha256 in ops/corpus-staging/sources.manifest.json
- Perseus Digital Library — Plutarch, Parallel Lives (Perrin translation)Bernadotte Perrin, Loeb Classical Library, Harvard University Press / William Heinemann, 1914–1926license: public-domain (US: pre-1930 publication); Perseus digital edition CC BY-SA 4.0, attribution recorded in ops/corpus-staging/SOURCES.mdpinned: github.com/PerseusDL/canonical-greekLit @ d121d1539cc37bfc9274bb81c6b6ec773111f098 (sparse checkout data/tlg0007, fetched 2026-07-07); 66 files tlg0007.tlg001–tlg0007.tlg066; per-file sha256 in ops/corpus-staging/sources.manifest.json
Deterministic extraction (extract-v1: proper-noun mid-sentence frequency with a capitalization-ratio gate; person/place context patterns; explicit event phrases widened by place+battle-vocabulary co-occurrence) over the W0 corpus. 715 figure, 52 event, 638 place, 187 people candidates; 26,045 attestation rows carry the exact surface forms and per-episode counts. For figures, places, and peoples the attestation sums reconcile to mention_count exactly. The 674 harvested-but-unclassified names and the 20 era phrase candidates are NOT in this database — they stay in the local ops pool (addresses.jsonl) per the unmatched-pool precedent (D-LEXICON-001) and the book-gated era plan (W3). Membership is recorded on the rows themselves (attestation.receipt_id = receipt:addresses-w1), not enumerated here.
What this does not prove- extract-v1 is frequency + pattern extraction — not NER and not curation; misfiled rows exist at the margins (the prototype found 'Roman' filed as figure).
- A candidate address may merge homonyms — three different wars answer to 'Salamis', many men to 'Alexander'; splitting them is W2 curation.
- Event attestations are place-name + battle-vocabulary co-occurrence, not verified narrations of the event; an event's mention_count counts explicit event-phrase hits only, so it will not equal its attestation sum.
- figure_kind is 'unassigned' on every candidate — person|legendary|deity is a W2 curation verdict.
- Mention counts are extraction-grade, not catalog-grade; no address carries year anchors until the era spine (W3) lands.
- For a handful of very frequent addresses (>400 mentions) the prototype artifact's episode_count was ref-capped; the attestation rows here are the uncapped card catalog and are the authoritative coverage.