Sorting hat: first end-to-end runs, calibration and report (Plan 2 Tasks 11-15)

[hyphaltip]

What this adds

Plan 2 Tasks 11 to 15 on HPCC. Report: docs/reports/2026-10-07-sorting-hat-run-and-calibration.md. Data: docs/reports/data/sorting_hat/ (73 KB). Scripts: analysis/sorting_hat_run/.

• Seven proteome runs end to end, exit code 0: A. fumigatus Af293 (UniProt and Fungi_5k), A1163, W72310, C. immitis RS, S. cerevisiae S288C (with and without dubious ORFs).
• Status of step1_rule@R0 from the Phase C metrics (Af293 UniProt run only). All other modules are unvalidated.
• Allergen leave-species-out recall: 65/111 (35% rule) and 41/111 (70% rule).
• Pfam review tables for 6 proteomes. No family is signed off (decision C4). No change to family_table.tsv.
• Strain stability by reciprocal best hit for three A. fumigatus annotations.
• Report-only panel check.

Read before merging

• This branch contains the two commits of PR Verify literature claims #70 (verify-literature-claims). Merge Verify literature claims #70 first, or this PR shows those commits too.
• The Pfam member lists are drafts from SGD and UniProt names. They were not reviewed by an expert. Their sensitivity and specificity lines must not be reported.
• The review_class column in the Pfam review tables is empty. A reviewer must classify each non-member hit.
• Only the Af293 UniProt run has the R0 status source (the plan's proposal, owner to confirm). S288C and RS therefore show R0 as unvalidated.
• FLO11 in S288C is not_called for tandem_repeat_protein, though the panel expects called. The cause was not examined.
• Not built: per-call status entries, Asp f 34 and four other panel rows, package CI job run.

Pre-commit (ruff 0.3.5, ruff-format) passes. No test was added: the scripts are analysis scripts. rbh_stability.py was checked on three real comparisons.

🤖 Generated with Claude Code