[RobbinBouwmeester]

Makes the diaPASEF gate (extract.gate_support_min, added opt-in in #206) the default on spectra that carry 1/K0.

Change

• extract.gate_support_min becomes an option. Unset is 0.845 on 1/K0 spectra and off otherwise; an explicit value wins, and 0 turns the gate off. Old configs that set a number keep working.
• The gate is defined on the window-mean score. Under any other gate_mode it is off whatever the setting, and the extract report names the threshold only when it applied.
• Spectra without 1/K0 (Orbitrap, Astral) are unchanged.

Why. On diaPASEF the window-mean gate at 0.6 passes 33% of the library, against about 5% on Astral and Orbitrap. The second gate keeps 41% of those candidates. It is a speed default that costs identifications.

Measured (docs/35 section 16):

data	peptides at 1%	wall	FDR
six 60-min HYE diaPASEF runs (seed 0)	98,524 -> 96,955 (-1.6%)	41.6 -> 26.8 min
six 5-min HYE diaPASEF runs (2 seeds)	39,164 -> 38,478 (-1.75%)	42-46 -> 22.5-23.5 min
E. coli diaPASEF entrapment run (3 seeds)	14,121 -> 13,892 (-1.6%)		FDP 0.32-0.35% vs 0.30-0.38%

Quantification accuracy is unchanged or slightly better (ProteoBench |epsilon| 0.171 vs 0.173). For a smaller loss, 0.82 gives -0.8% on E. coli (FDP 0.32-0.37%) for less of the speed.

Other changes

• The pipeline test of the 1/K0 evidence columns turns the gate off, as it already turned the window-mean threshold off, so its synthetic target passes.
• A unit test covers the default resolution (1/K0 or not, explicit 0, other gate modes).
• docs/35 also records the end-to-end prefilter timings: 5-min HYE; 22 HeLa runs 21:04 -> 15:36 at 31.5 -> 18.6 GB; one single cell 4:31 cold, 1:37 warm.
• CHANGELOG and CLAUDE.md are updated.

Checked: cargo fmt, clippy -D warnings, workspace tests, cargo doc -D warnings, config reference regenerated. The smoke fixture has no 1/K0 arm, so it is unaffected.

🤖 Generated with Claude Code