Documentation standards

Public scientific language must identify what Core computed, what it imported, which policy shaped the result, which evidence challenged it, and where the claim stops. Feature breadth is not a substitute for this chain.

Claim vocabulary

Term	Required evidence	Does not mean
native	repository-owned transformation from the named input contract	vendor-native raw processing or external-engine parity
imported	named external producer, version or format, parameters when available, and preserved provenance	independently reproduced computation
validated	declared invariant and the positive and negative cases that exercise it	universally scientifically correct
benchmark-backed	identified asset, license, lineage, acceptance bar, and result	broad transfer beyond the benchmark envelope
parity	named comparator, fields, tolerances, disagreement policy, and versions	interchangeable tools or algorithms
deterministic	identical supported inputs, configuration, ordering policy, and environment envelope	invariance across arbitrary dependency or hardware changes
supported	workflow family, run mode, inputs, outputs, and current trust ceiling	every route in the broader proteomics category

Claim construction

flowchart LR
    O["operation and owner"] --> I["input and provenance"]
    I --> P["policy and parameters"]
    P --> E["evidence and challenge"]
    E --> L["limitation and trust ceiling"]

A useful example ends with a reviewable artifact: accepted and rejected records, policy, parameters, source identity, warnings, and reason codes. A numeric output without those fields demonstrates syntax, not scientific trust.

Comparisons and benchmarks

Name whether a comparator output was imported, replayed, or independently computed. State exact comparison dimensions and tolerances. Preserve disagreement instead of averaging it away. Benchmark pages identify asset custody and licensing, primary versus companion role, holdout status, and the condition that would narrow the claim.

Cross-package language

Core owns scientific models and transformations. Runtime owns execution and retained run state; Knowledge owns evidence and citation custody; Intelligence owns recommendation policy; Lab owns assay readiness and observed outcomes. Public prose must not transfer authority merely because Core data appears in a downstream artifact.

Use workflow families for current family ceilings and known limitations for the limits that apply even when Core checks pass.