Tutorials¶

Short, practical end-to-end recipes.

Tutorial 1 — Run the demo¶

llm-pathway-curator run \
  --sample-card examples/demo/sample_card.json \
  --evidence-table examples/demo/evidence_table.tsv \
  --out out/demo/

Check:

out/demo/audit_log.tsv
out/demo/report.md

Tutorial 2 — From fgsea to audited claims¶

Run fgsea in your pipeline.
Convert results → EvidenceTable (adapter or your own export).
Run:

llm-pathway-curator run \
  --sample-card sample_card.json \
  --evidence-table HNSC.evidence_table.tsv \
  --out out/hnsC/

Tutorial 3 — Context swap (stress test)¶

Run the same EvidenceTable under a swapped Sample Card (e.g., BRCA → LUAD). Expected behavior: PASS coverage decreases; ABSTAIN reasons shift toward context issues.

(Exact CLI depends on your stress runner / paper scripts. See paper/scripts/ for reproducible sweeps.)

Tutorial 4 — Evidence dropout (stress test)¶

Drop supporting genes with probability p (seeded). Expected behavior: PASS coverage decreases; ABSTAIN reasons shift toward stress-inconclusive / instability.

Tutorial 5 — Pick a τ operating point¶

Do a τ sweep and summarize:

PASS/ABSTAIN/FAIL rates
reason code distribution
(optional) human-labeled risk among audit-PASS

Then lock τ for your analysis or deployment run.

Notes¶

For the underlying design, see Concepts.
For deterministic reproduction (benchmarks/figures/Source Data), follow paper/README.md.