Repository Public Research preview
1# CLAUDE.md — WP-AI-0002: Human Text Reconstruction Transformer (HTR)23**Can a Transformer reconstruct characteristics of pre-2018 human-authored scientific prose from LLM-rewritten4passages?** Research home: UQO Research (`Recherche-UQO/ia/wp-ai-0002`). Language: English for code, docs,5paper. Identifier: **AI-0002** (series "AI", after AI-0001). Status: see `docs/STATUS.md`.67The author's full mandate (mission, principles, RQ1–RQ6, phases 0–8, gates A–F, definition of done) is8reproduced verbatim in `docs/MANDATE.md` and binds every session. Author additions (2026-11-09): the corpus must9be **large** (target the "Medium" stage, ≥ 10 000 documents, ≥ 200 000 eligible passages), must include10**economics and finance** articles and **English reference books published before 2018**, and the released11model must be **genuinely usable afterwards** (model release is a first-class deliverable, not a by-product).12PyTorch may be installed on other MacLustr nodes if training needs them.1314Operational rules that follow from the mandate:15161. **Never fabricate** data, metrics, sources, licences or checkpoints. Every number in `docs/`, `artifacts/`17 and `paper/` traces to a run id in `artifacts/runs/<run>/manifest.json` (hashes, seeds, versions, costs).182. **Rights first.** Open access ≠ open licence. `htr.licenses.policy` is the single source of truth; a document19 is trained on only if its canonical licence code is in the configured training set, and released only if in20 the release set (`CC0`, `PUBLIC-DOMAIN`, `CC-BY` by default). Dates are verified from the document or an21 official metadata record, never inferred from upload timestamps.223. **Honest claims.** Outputs are synthetic reconstructions. No claim of "restoring human authorship" or of AI23 detection. Separate verbatim reconstruction, stylistic distribution matching and factual fidelity. Test24 gains against identity (B0) and generic-paraphrase (B1) baselines; report null/negative findings.254. **No leakage.** Splits are by document, near-duplicate cluster and author group (`htr.data.splits`); held-out26 discipline, period and rewriting prompt; originals of held-out items are never used at inference.275. **Spend is opt-in.** The Anthropic adapter runs in `mock`/`dry-run` by default; real calls need28 `--live` and a budget ceiling; the pilot ceiling is in `configs/rewriting/pilot.json` and29 `docs/reproducibility/COMPUTE_BUDGET.md`. Spend so far is in `docs/STATUS.md` (from the job ledgers).30 Ask before exceeding it, before mass acquisition beyond the planned stages, before any public upload.316. **Gates in order** (A repository + governance + 100-paper smoke → B validated pairs → C splits/baselines →32 D M1/M2/M3 pilot → E scale → F release + paper). Each gate writes `docs/gates/GATE_<X>.md` with GO/NO-GO.337. **Remote** `origin = gitsrv:srv/git/wpai0002_uqo.git` (spbgit on M1M32). Commit after each deliverable.34 Raw data, pairs and checkpoints are never committed; `paper/main.pdf` is.3536Quick start: `.venv/bin/python -m pytest tests -q -m "not network"` · `scripts/vertical_slice.sh` ·37`.venv/bin/python scripts/discover_pmc.py --max-docs 100` · `docs/REPRODUCIBILITY.md`.38