[cschanhniem]

Summary

Three remaining quality gaps closed before synthesis ordering:

1. phase3.yaml config validation (3 new tests in test_config.py) — pipeline.yaml was validated at startup; phase3.yaml was not. If max_disagreement is accidentally removed from phase3.yaml, pipeline.py would crash at runtime with KeyError: 'disagreement' when iterating candidates. CI now catches this immediately.

2. compute_features → synthesis_feasibility_score contract test — synthesis_feasibility_score() uses direct dict access (not .get()) for length, longest_repeat_run, and cysteine_fraction. There was no test verifying compute_features() always returns all three. A refactor that renames one key would silently crash candidate scoring. Added test_compute_features_supplies_synthesis_feasibility_keys that both checks key presence and runs the function end-to-end.

3. docs/WET_LAB_HANDOFF.md — new guide for wet-lab scientists covering:

   • Score interpretation table (what does ensemble 0.65 mean for ordering?)
   • Synthesis decision thresholds (synthesise / conditional / expert review / exclude)
   • Selection gate rationale (why 0.30 for phase3 vs 0.40 for pipeline.yaml?)
   • Pre-synthesis QC flag guide with remediation actions (Cys, Met, hydrophobic run, trypsin sites)
   • MIC + hemolysis assay protocols (organisms, medium, concentration ranges, replicates)
   • Troubleshooting for < 10 nominations, synthesis failure, all-hemolytic batch
   • Confidence statement: "hit rate 20–40% consistent with AMP field literature"

Test plan

• pytest tests/test_config.py tests/test_physchem.py — 57 passed
• Full suite: 814 passed (up from 810)
• No regressions