docs(pdep): name the finite-SA-row precondition at the distrust call site (I-037)

[alongd]

What

The distrust selector's escape from the master-equation sensitivity floor was documented at the pes_loop call site as if it were unconditional: "distrust, not that coefficient, decides what survives and in what order." This corrects the record. The escape lives in the ranking (compute_distrust reads no sensitivity value), but selection is conditional on Arkane emitting a finite SA row.

Comment-only; no behaviour change.

Why (I-037)

select_by_distrust (t3/pdep/distrust.py:317) gates each candidate on evidence_by_ts_label.get(candidate.ts_label). A candidate with no finite row is skipped before the distrust energy window is applied — SKIP_UNMEASURABLE if structurally unmeasurable, else SKIP_NO_EVIDENCE — because t3.pdep.capture refuses an artifact with no coefficient to record.

An ordinary structural zero is a finite row (0.0 or ~1e-18) and is kept, so on a pristine pre-QM surface every bimolecular entrance channel survives to be ranked (confirmed by test_run_round_me_sensitivity_runs_real_arkane, where real Arkane emits a finite row for every TS). Only a NaN/absent row — a degenerate ME solve, not the structural-zero case distrust was built for — reintroduces the floor's blindness for that one in-window candidate.

Verifier

A constructed no-row network run through select_by_distrust:

candidate	in window?	Arkane row?	result
TS1 bimolecular entrance	yes	none	SKIP_UNMEASURABLE
TS2 unimolecular	yes	none	SKIP_NO_EVIDENCE
TS3 control	yes	finite	SELECTED

tests/test_pdep/test_distrust.py tests/test_pdep/test_pes_sa.py tests/test_pdep/test_pes_loop.py — 80 passed, unchanged.

Verdict

Conditional escape, condition = a finite SA row exists for the candidate. The condition holds on ordinary pre-QM surfaces (finite structural zeros), so this is a documentation matter, not a live defect there. The failure mode (NaN/absent row) is reachable only under a degenerate ME solve; whether real Arkane then writes a NaN row vs. crashing the whole SA job (which the round already catches and fails loudly) is unverified here without running Arkane on such a network.

[Copilot]

Pull request overview

Updates in-code documentation at the run_pes_loop distrust-selection call site to clarify that distrust-based ranking is independent of the sensitivity floor’s value, but candidate selection still requires the presence of a finite sensitivity coefficient row (i.e., evidence must exist for capture/recording).

Changes:

• Refines the comment above select_by_distrust(...) to distinguish ranking behavior (distrust) from the precondition for eligibility (finite SA evidence row).
• Documents the skip classifications (SKIP_UNMEASURABLE / SKIP_NO_EVIDENCE) that occur before the distrust energy window is applied.
• Adds a concrete pointer to the existing real-Arkane test demonstrating finite rows for each TS in a normal run.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.00%. Comparing base (e3f339d) to head (20f8ab3).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #211      +/-   ##
==========================================
- Coverage   82.01%   82.00%   -0.02%     
==========================================
  Files          76       76              
  Lines       12268    12268              
  Branches     2626     2626              
==========================================
- Hits        10062    10060       -2     
- Misses       1584     1585       +1     
- Partials      622      623       +1     

Flag	Coverage Δ
unittests	82.00% <ø> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.