fix(lib): don't fabricate a supporting study with no identity and no content

[SkyeAv]

Summary

inline_supporting_study emitted has_supporting_studies unconditionally, so a section declaring no publications shipped a fabricated Study on every edge.

Changes

• What was changed — inline_supporting_study now emits the has_supporting_studies struct only when it carries something. It takes a new identified argument (whether study_id names a real publication rather than falling back to the config filename), threaded from Tcode._provenance_ops as bool(publication_values). When a section is unidentified and has no routed UNSATISFIABLE_EDGE_FIELDS and no _pruned_by_class values, the struct is skipped. The sheet-name and row-number columns are dropped either way — they are never meant to reach the edge.

• Why this change is needed — Without a publication, study_id falls back to the config filename, so the emitted Study was keyed my_table.yaml and its single StudyResult was a row index into a file the pipeline regenerates:

  "has_supporting_studies": {"approved_treats.yaml": {
    "id": "approved_treats.yaml", "name": "approved_treats.yaml",
    "has_study_results": [{"id": "approved_treats.yaml#row6848", "name": "row 6848"}]}}

  Biolink defines has supporting studies as "studies that produced information used as evidence". Nothing in NCATSTranslator/translator-ingests models evidence this way — there a Study is a real cohort (ICEES), dataset (COHD), trial (CTKP), or text-mining group (SemMedDB/TMKP) carrying typed StudyResult slots, never a struct keyed by a filename. Asserting a study that never existed, on every edge, is worse than emitting nothing.

• How it affects the project — Narrow and additive. A section with a real publication keeps the struct byte-for-byte unchanged; PMID:123#Table_S7 row 12 is genuine provenance and that is the case the feature exists for. A section with statistics to preserve also keeps it, published or not — the rescue path is the whole reason the filename fallback exists, and losing a class-refused value would be worse than naming its carrier after a config file. Only the empty-and-anonymous combination changes.

  Note for downstream: because prune_to_class creates _pruned_by_class at the schema level, any frame carrying a prunable column (today, species_context_qualifier on a class that lacks the slot) still counts as having content. Graphs will see this fix take effect once that column stops being derived.

Testing

• Commands run and results — uv run pytest → 958 passed, 1 skipped. The single failure (tests/test_agent_eval.py::test_gepa_metric_returns_prediction) is pre-existing and reproduces identically on a clean main checkout. uv run ruff check . → passed. uv run ruff format --check . → passed. uv run pyright → 0 errors. Coverage of src/tablassert/lib.py holds at 99% with the same six uncovered lines as baseline, so the new branch is fully exercised and no gap was introduced.
• Unit tests added/updated — three direct inline_supporting_study tests in tests/test_lib.py, covering each arm of the decision: unidentified + empty (no struct, row/sheet still consumed), unidentified + pruned statistics (struct kept, description intact), and identified + empty (struct kept, no description key).
• Integration tests added/updated — none needed; the existing end-to-end sections all declare publications and continue to pass unchanged, which is the no-regression evidence.
• Manual testing performed — verified against a real tablassert build-kg in a downstream graph: edges lose the contentless struct while published-section edges keep theirs.

Related Issues

• None

Checklist

• Tests pass
• Code style compliant
• Documentation updated
• No breaking changes

🤖 Generated with Claude Code

https://claude.ai/code/session_014gQKmpqbfNTVPfdzYw97L2

[coderabbitai]

Important

Review available on request

• 🔍 Trigger review

Reviews should be triggered manually for repositories with fewer than 10 stars. Select Trigger review above or comment @coderabbitai review to review the latest changes. For a full review, comment @coderabbitai full review.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 638a9741-c803-4e4b-875e-25b54b80b62f

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}