Text to evidence
Resources
Graded by how much they can be trusted about this pipeline. In-repo code and ADRs outrank everything else, because they are the thing being learned.
In-repo, authoritative
| Resource | Trust | What it settles |
|---|---|---|
| docs/data-pipelines.md | Highest | The authored-layer map: which paths write Postgres, which write Neo4j, which are one-way, which look live and are not |
| docs/extraction-checklist.md | Highest | Reading order for an entry, where each thing is recorded, notInSource vs unchecked |
| docs/companion-extraction-checklist.md | Highest | Extraction order and resume point, in the companion index's own order |
| prisma/schema.prisma | Highest | The stored shape. Its doc comments carry design reasons the ADRs do not repeat |
src/lib/history/batchSchema.ts | Highest | Every validation rule, and passageExcerpts — the positional anchor binding |
src/lib/history/importBatch.ts | Highest | How a batch becomes rows, inside one transaction |
scripts/history/verifyExcerpts.ts | Highest | The substring check that detects anchor drift; not wired into validate |
src/lib/history/sourceAccounts.ts | Highest | The read side: pagination, section index, payload shaping for the reader |
ADRs that hold the design arguments
| ADR | What it decides |
|---|---|
| 0008 | Review status is shown, never used to hide data |
| 0009 | Evidence attaches to a subject slug, not a profile |
| 0010 | data/ is authority; Postgres, Neo4j and seeds are projections |
| 0011 | Evidence role is a property of the citation, not the work |
| 0013 | Attendance is the relation; what happened is the status |
| 0016 | A page is the source's page, not the screen's |
External
| Resource | Trust | Notes |
|---|---|---|
| Siyar, al-Risalah ed. on Shamela | High for the text | The only host this project extracts from: ed. Husayn al-Asad under Shu'ayb al-Arna'ut, 3rd ed., 1405/1985. It carries the editor's footnotes, and its printed pages are the ones citations record. Its فهرس الموضوعات is where extraction order comes from. A delivery host, not an evidence class (ADR 0011) |
| Page ids | Diagnostic only | https://shamela.ws/book/10906/<id> runs monotonically through the edition, so it checks a declared volume or printed page. Note الجزء there labels the Siyar's own parts: its الجزء ٢ is this edition's volume 5 |
Communities
Not yet explored. The design questions here (provenance modelling, citable-unit granularity, editorial review states) are shared with digital-humanities and TEI work — worth finding a venue for once a concrete design proposal exists to put in front of people. Raise it when you want this pursued.