Phase 1: foundation, bug-fix upstream, source router (deepXIV + biomed)

[StartripAI]

Summary

Phase 1 of the OpenRevise unification per docs/plans/2026-05-01-openrevise-unification-design.md:

1A — Foundation restructure:

• Scaffolded pyproject.toml (Apache-2.0, name=openrevise, optional extras for preprint/biomed/mcp)
• Moved scripts/ → src/openrevise/{artifacts,gates,sources,revise,pipeline}/ with corrected imports and parents[3] repo-root resolution
• Added leak-guard regression test + pytest CI workflow with concurrency, timeout, fail-fast=false, pip cache, and permissions: contents: read
• Confirmed Apache-2.0 LICENSE; README content already neutral

1B — Bug-fix upstream from private revise:

• housekeeping --dry-run no longer mutates the run-index (two upsert_run_record sites guarded; regression tests for both COLD_ARCHIVED and EXPIRED_NONKEY_PURGED branches)
• revise_docx reports a friendly stderr error instead of raw KeyError when the input DOCX is missing word/document.xml or word/footnotes.xml
• ITT/mITT label-binding gate ported with regression test for SOP §Q15 swap detection; wired into both pipelines (run_revise_pipeline.py and _v2.py); --author default flipped Codex→OpenRevise across all pipeline scripts; meta.notes field added to patch-spec template (neutral text)

1C — Source router + preprint stack:

• Skeleton SourceRouter with source_type dispatch and BackendNotRegistered exception
• PreprintRouter with priority-ordered engines and graceful fallback (DeepXiv-SDK primary → native arXiv via feedparser fallback → soft-degrade to empty)
• Real pubmed.py, europepmc.py, pmc.py clients using NCBI E-utilities and Europe PMC REST (replaces metadata-only registry tags)
• default_router() factory wires preprint + biomed (with literature source_type alias for legacy registry compat); config/source_registry.yaml extended with arxiv/medrxiv/biorxiv source_ids

Test plan

• pytest -q green locally (17 tests including 5 new TDD regression tests)
• CI green on Python 3.11 / 3.12 matrix
• pip install -e ".[dev]" succeeds in fresh venv
• Leak-guard test catches a synthetic offender (PANOVA fixture positive control verified during Task 5 cycle)
• Confirm CI workflows trigger correctly on this PR (test)

Out of scope (separate plan)

• MCP server, CLI entry points, Claude Code skill bundle (Phase 2 / design steps 4–5)
• LabelBindingProfile registry refactor — check_label_value_consistency.py retains opioid-anchored regex (with leak-guard allow-list TODO marker pointing at design step 7)
• Private revise downstream refactor (Phase 2 / step 6)
• v0.1.0 PyPI release (Phase 2)

🤖 Generated with Claude Code