Phase B: calibrated precursor-window tightening (-7.8% Astral wall)#26

Merged

ypriverol merged 26 commits intodevfrom

feat/precursor-window-tightening

May 1, 2026

Commits on Apr 29, 2026

docs(plans): Phase E GC-pressure follow-up — bigger heap helps 8t but not enough
ypriverol
committed
docs(plans): Phase E final disproof — anti-scaling and ForkJoin win were noise
ypriverol
committed
fix(bench-ci): unbreak PXD001819 CI after PR #23 mzIdentML removal
ypriverol
committed
feat(phase-b-telemetry): add opt-in counter for pairing fan-out verification
ypriverol
committed
fix(calibrator): isolate pre-pass at iso=[0,0] + outlier-filter residuals
ypriverol
committed
feat(phase-b): expose tightening formula constants as system properties
ypriverol
committed
feat(calibrator): stratify residuals by spec_eValue, keep top MIN_CONFIDENT_PSMS
ypriverol
committed
docs(plans): record Phase B Astral win after stratification fix
ypriverol
committed
docs(plans): SHIPPED.md Active section reflects Phase B win
ypriverol
committed
docs(plans): three-dataset Phase B validation table in SHIPPED.md
ypriverol
committed
docs(plans): Experiment 2 design — exact prefix mass-interval pruning
ypriverol
committed
feat(experiment-2): mass-interval pruning scaffold (off by default; Checkpoint 1+2 measured)
ypriverol
committed
docs(plans): Experiment 2 status header — kill gate hit on wall
ypriverol
committed
perf(experiment-2): replace TreeMap.subMap with binary-search on sorted double[]
ypriverol
committed
docs(plans): Experiment 2 status header reflects Checkpoint 3 result
ypriverol
committed
perf(experiment-2): gate bound test on peptideLengthIndex >= minPeptideLength (Checkpoint 4)
ypriverol
committed
docs(plans): Experiment 2 Checkpoint 4 — gate-on-minPeptideLength shipped, bench pending
ypriverol
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Phase B: calibrated precursor-window tightening (-7.8% Astral wall)#26

Phase B: calibrated precursor-window tightening (-7.8% Astral wall)#26
ypriverol merged 26 commits intodevfrom
feat/precursor-window-tightening

Commits on Apr 28, 2026

docs(plans): start astral-speed-improvements; fold shipped plans into SHIPPED.md

docs(plans): consolidate to 5x roadmap; adopt milestone-commit shipping model

docs(plans): Phase A retrospective; revert in-tree code via separate reset

docs(plans): Phase E retrospective — parallelism win not replicable under noise

Commits on Apr 29, 2026

docs(plans): Phase E GC-pressure follow-up — bigger heap helps 8t but not enough

docs(plans): Phase E final disproof — anti-scaling and ForkJoin win were noise

fix(bench-ci): unbreak PXD001819 CI after PR #23 mzIdentML removal

feat(phase-b-telemetry): add opt-in counter for pairing fan-out verification

fix(calibrator): isolate pre-pass at iso=[0,0] + outlier-filter residuals

feat(phase-b): expose tightening formula constants as system properties

feat(calibrator): stratify residuals by spec_eValue, keep top MIN_CONFIDENT_PSMS

docs(plans): record Phase B Astral win after stratification fix

docs(plans): SHIPPED.md Active section reflects Phase B win

docs(plans): three-dataset Phase B validation table in SHIPPED.md

docs(plans): Experiment 2 design — exact prefix mass-interval pruning

feat(experiment-2): mass-interval pruning scaffold (off by default; Checkpoint 1+2 measured)

docs(plans): Experiment 2 status header — kill gate hit on wall

perf(experiment-2): replace TreeMap.subMap with binary-search on sorted double[]

docs(plans): Experiment 2 status header reflects Checkpoint 3 result

perf(experiment-2): gate bound test on peptideLengthIndex >= minPeptideLength (Checkpoint 4)

docs(plans): Experiment 2 Checkpoint 4 — gate-on-minPeptideLength shipped, bench pending

Commits on Apr 30, 2026

docs(plans): Experiment 2 Checkpoint 4 confirmation — 5-trial bench, -2.27% Astral wall

chore: remove non-shippable runtime scaffolding; keep Phase B as the real improvement

Commits on May 1, 2026

fix(phase-b): isolate Spectrum state during calibration pre-pass

chore: align .claude/plans/ and benchmark/ci/ with dev (drop from PR diff)

feat(calibrator): expose maxSampled and minConfidentPsms as system properties