Workflow Claim Limits

Proteomics support is evaluated one workflow family at a time. A family earns public confidence only when its scientific benchmark, executable route, external comparison, evidence review, recommendation challenge, and downstream consequence can be inspected together.

The checked workflow matrix currently yields these exact sets:

Outsider-auditable workflow families today: dda, dia, ptm, targeted.

Full outsider-readable family packets today: dda, dia, lfq, ptm, targeted.

Internal-support-only workflow families today: multiplex.

The three sets are not a maturity ladder. A full packet makes the proof surface readable by an outsider. Outsider-auditable authority additionally requires the family to earn every governed authority cell. Internal support identifies a family whose cross-package evidence is not sufficient for external reliance. The current release preflight passes with these distinctions intact.

Current Family Limits

Family	Family-matrix result	Strongest usable public sentence	Decisive limitation
DDA	outsider-auditable	bounded outsider-auditable	Search-engine behavior and transfer remain bounded by the shipped DDA lanes and comparator evidence.
DIA	outsider-auditable	bounded outsider-auditable	The executable lane remains library-conditioned rather than chromatogram-native replay.
LFQ	outsider status not earned	review-grade bounded	Missingness, normalization, sparse-cohort transfer, and external-review pressure remain material.
Multiplex	internal support only	internal support only	External comparison and lab-consequence closure do not support an outsider-facing claim.
PTM	outsider-auditable	bounded outsider-auditable	Localization evidence is stronger than downstream consequence confidence.
Targeted	outsider-auditable	bounded outsider-auditable	Calibration, interference, carryover, and follow-up burden constrain transfer.

What Each Classification Means

Outsider-auditable means a skeptical reader can open the benchmark package, runtime evidence, comparison, grounded claim, recommendation packet, and consequence record without relying on maintainer narration. It does not mean universal accuracy or transfer.

Review-grade bounded means the family has real, inspectable scientific and operational evidence, but a visible blocker prevents the stronger public sentence. The blocker is part of the result, not an editorial qualification.

Internal support only means implementations and benchmark material exist, but the cross-package proof chain does not yet support external reliance.

Evidence threshold

An outsider-readable family packet connects the evidence needed to reproduce and challenge the public sentence:

• two public benchmark packages plus one published cross-package generalization report;
• family_stability_scorecard.json with family-specific perturbation and transfer results;
• a canonical runtime lane, rerun dossier, artifact ledger, and comparison evidence;
• grounded support, contradiction, and citation context;
• a challenged recommendation packet with sensitivity and downgrade evidence;
• a requested-versus-observed outcome dossier and an assay-worth-it ledger row when laboratory consequence is claimed.

Missing one of these artifacts narrows the corresponding authority cell. It does not disappear because adjacent packages carry stronger evidence.

flowchart LR
    benchmark["scientific benchmark"] --> runtime["runtime rerun"]
    runtime --> comparison["external comparison"]
    comparison --> grounding["claim grounding"]
    grounding --> challenge["recommendation challenge"]
    challenge --> consequence["lab consequence"]
    consequence --> sentence["public sentence"]

    runtime -. veto .-> narrow["narrow the sentence"]
    comparison -. veto .-> narrow
    consequence -. veto .-> narrow

How to Audit a Family

1. Open its tracked benchmark packages and cross-package generalization report.
2. Confirm whether the strongest runtime route is raw-executable or import-only.
3. Inspect comparator posture, known limits, and unsupported-claim records.
4. Review the recommendation challenge and downgrade evidence.
5. Confirm that requested-versus-observed outcomes and assay-worth records exist.
6. Use the weakest result from those surfaces as the family claim.

The public artifact index gives the inspection order. The black-box benchmark dashboard shows the current runtime verdicts, and current capability limits explains why a technically substantial family may still carry a narrow public sentence.

Claims These Classifications Do Not Support

• equivalence across instruments, cohorts, search engines, or acquisition modes;
• biological truth merely because a workflow executed reproducibly;
• laboratory usefulness without a consequence record;
• promotion of a weaker family from evidence earned by a stronger neighbor;
• repository-wide trust inferred from any single family packet.

When the family matrix and a live gate disagree, the weaker result governs until the evidence and classification agree again.