CHANGELOG: v0.9.0 header, fixes moved out of Unreleased, missing entries - #346

What

CHANGELOG.md only. Brings the [0.9.0] section in line with main after #337 merged. The date stays TBD: set it when you cut the tag.

1. Header. The "DRAFT release prep (do not tag until the definitive run is integrated)" banner is replaced. It said the counts would be filled in later and named the tblastn outlier policy as on for the run. The definitive-run section below it already says the 372 genomes come from reruns with the policy off.
2. [Unreleased] fixes moved to 0.9.0. The Detect: taxid 5158 has no NCBI lineage, and --phylum leaves MAT genes unmodelled (7 of 8 Ceratocystis fimbriata assemblies uncalled) #324 empty-taxonomy record (#324: record an empty taxonomy lookup (routing_error); panel warm-up counts it and can stop the job #339) and the genes_found order (Scoring: genes_found lists optional slot genes in roster order (reproducible report order) #335) are in main and will be in the tag. Neither is in the frozen run's code stamp (7d73be5); neither changes a call. [Unreleased] is now empty.
3. Entries added for merged work the section did not list: --weak-coverage-noise-policy (Unpolished-gene noise floor is a bitscore cliff: a better MTLA1 hit can lower confidence #149, with the two limits from Weak-coverage noise (#149): act only on a determined idiomorph with best core coverage >= 50% #332), --withheld-detail-policy (Report-only: --withheld-detail-policy lists the genes of each withheld locus #313), the nearest_reference column (Label tables: the nearest_reference column, both core genes listed (#200) #311), the Basidiomycota defective-flag table change (Tables: the Basidiomycota defective flag is a model check; low-confidence models listed apart (#168) #304), the libyaml C loader (Parse curated YAML with libyaml's C loader; identical results on 823 genomes (#287) #292), the reads gate causes (Reads gate causes for v0.9.0 (#195, #196): not typable, zygosity, unchecked; cause table; site wording #325), thin-evidence typing (Reads gate: type thin-evidence species (one strain or 1-3 runs) as UNVALIDATED instead of holding them #327), clone groups (Reads clone groups by solid containment: 202 species, calibrated on same-strain pairs #320), the A. fumigatus population report (Reads: A. fumigatus population MAT report (4,849 isolates), MAT frequency, ENA FTP fallback #326), the sketch store (Sketch store: one reusable read sketch per isolate for MAT, target screens and population structure #315, Sketch store: trim a broken ENA stream to whole read pairs; retry partial downloads #319, Sketch store: Starship/bafA prevalence, co-occurrence and gain/loss on the tree (starship_dynamics.py) #329, baf family screen: bafA beyond the AF293-type allele (baf_family.py), report section 8, bafA wording corrected #330, Strain-set Starship table with the all-alleles bafA class; eAF strains (PRJNA991533) #333), the landing-page charts (Landing page: Fola and A. fumigatus read-typing charts, tables and notes (#172) #301) and the FungiDB annotation check (#290: per-tier FungiDB annotation agreement (script, test, note); no verdict changes v0.8.1 to v0.9.0 #336). Every number comes from the PR description.

For the curator

• Scope is your call. The reads and sketch-store bullets (Reads gate causes for v0.9.0 (#195, #196): not typable, zygosity, unchecked; cause table; site wording #325, Reads gate: type thin-evidence species (one strain or 1-3 runs) as UNVALIDATED instead of holding them #327, Reads clone groups by solid containment: 202 species, calibrated on same-strain pairs #320, Reads: A. fumigatus population MAT report (4,849 isolates), MAT frequency, ENA FTP fallback #326, the sketch store, Landing page: Fola and A. fumigatus read-typing charts, tables and notes (#172) #301) are there because the 0.8.1 section lists reads work. Drop any you do not want in this changelog.
• Unpolished-gene noise floor is a bitscore cliff: a better MTLA1 hit can lower confidence #149 wording. The entry states the rule and its two limits. It does not say the 109-locus list is approved: the Unpolished-gene noise floor is a bitscore cliff: a better MTLA1 hit can lower confidence #149 thread has your approval of that list (2026-11-12) and Weak-coverage noise (#149): act only on a determined idiomorph with best core coverage >= 50% #332 later restricted the rule, and I left that for you to settle. The entry says the limited version has not been run on a panel.
• Internal docs and notes PRs (Add CLAUDE.md: working rules for agents and contributors #295, Proposal: the nearest_reference column for the idiomorph labels (#200) #300, Notes: MAT loci not recovered in the v0.8.1 run, and what could improve them #307, Notes: the two-stage tblastn search measured: slower and not identical (#287) #309, Notes: the one-polished-gene measurement (edge-case action 5) #314, Notes: junction-panel and cannot-be-typed design for #195 and #196 (proposal for ruling) #321, Notes: decisions on the six open questions of the junction-panel design (#321) #340) are left out.

Not verified

• I recomputed the 202, 23 and 108 counts from the committed tables; I did not re-run anything.
• Correction (2026-11-12): an earlier version of this note said I counted 107 undetermined loci against the CHANGELOG's 108. That was my error: my count printed only the top six change categories. The tables give 108, as the CHANGELOG says.
• Docs only: I did not run the test suite. Nothing in tests/, scripts/ or src/ reads CHANGELOG.md.

🤖 Generated with Claude Code