Metadata-Version: 2.4
Name: crowley-diary
Version: 0.1.0
Summary: Faithful, LLM-assisted transcription pipeline for the J. S. Crowley diary
Requires-Python: >=3.12
Requires-Dist: jsonschema[format]<5,>=4.23
Requires-Dist: pillow<13,>=11
Requires-Dist: typer<1,>=0.15
Description-Content-Type: text/markdown

# J. S. Crowley diary transcription proof of concept

This project builds a faithful, page-scoped structured edition from the scanned
page image, raw OCR, and tightly limited neighboring context. The page image is
authoritative; OCR and model output are evidence and interpretation, not ground
truth.

The first phase is deliberately limited to the 14 pages listed in
`evaluation/selection.json`. It uses a ChatGPT-authenticated Codex subscription
through `codex exec`; it does not use an API key or an API-billed SDK.

## Quick start

```sh
uv sync --dev
uv run diary source verify --source-root /path/to/diary_ocr_package
uv run diary transcribe batch \
  --source-root /path/to/diary_ocr_package \
  --selection evaluation/selection.json
uv run pytest
```

See `docs/architecture.md` and `docs/editorial-specification.md` before treating
any transcription as edition text. Agent-reviewed evaluation fixtures are
diagnostic only and remain `agent_reviewed` until the owner personally approves
them.

