Skip to content

Flagship Acceptance Bars

bijux-proteomics now publishes one explicit acceptance sheet per flagship workflow family instead of letting trust posture hide inside a mix of review prose, challenge assets, and release pages.

The machine-readable root is:

  • packages/bijux-proteomics-core/benchmark-assets/flagship-acceptance/

That root currently ships:

  • dda_acceptance_sheet.json
  • dia_acceptance_sheet.json
  • lfq_acceptance_sheet.json
  • multiplex_acceptance_sheet.json
  • ptm_acceptance_sheet.json
  • targeted_acceptance_sheet.json
  • acceptance_dashboard.json
  • benchmark_history_ledger.json
  • acceptance_rationale_dossier.json

What The Sheets Measure

Each family sheet publishes five exact bars tied to the real flagship package:

  • dda: search coverage, protein inference stability, calibration sanity, comparator divergence tolerance, and review-packet promotion.
  • dia: library dependence, peptide evidence coverage, protein evidence stability, quantitative coherence, and review-packet promotion.
  • lfq: missingness burden, normalization drift, differential reproducibility, comparator divergence, and recommendation promotion.
  • multiplex: interference, channel dropout, reference-channel fragility, ratio compression, and downstream review promotion.
  • ptm: localization quality, ambiguity burden, motif credibility, occupancy stability, and targetability promotion.
  • targeted: calibration quality, transition interference, heavy-light coherence, carryover posture, and follow-up promotion.

Current Trust Posture

The dashboard is the short answer:

  • acceptance_dashboard.json compares current public release language with the earned language from the live sheets.
  • trusted bounded families are expected to stay at outsider_auditable_bounded only while their sheets pass.
  • multiplex remains internal_support_only, and its sheet is intentionally failing because current imbalance, dropout, and compression pressure do not justify stronger trust.

Why These Bars Exist

The rationale dossier is the long answer:

  • acceptance_rationale_dossier.json ties every bar back to comparator pressure, literature-backed threshold anchors, benchmark difficulty, or lab consequence.
  • the history ledger records the first published threshold set so later edits cannot quietly become looser without saying so.

How To Refresh

Run:

uv run --group dev python -m bijux_proteomics.benchmarks.flagship_acceptance_assets refresh

That command rewrites all six family sheets plus the dashboard, threshold-history ledger, and rationale dossier.