Statistical Genomics Foundry
Productive and statistically honest.
A knowledge base of agent skills that keeps analytical work both productive and statistically honest — using established methods correctly instead of inventing plausible-sounding ones, respecting each method's preconditions, and grounding claims in empirical checks rather than a method's name or a significant-looking number.
How work earns confidence
Family A Analyze Batch-confounding audit
Family B Referee Independent empirical check
Gate Revise Return with findings
Findings return to Family A
Molds
All molds →The Mold-primary core — abstract action templates cast into self-contained skills. Family A does the analysis; Family B referees it, so doing never self-certifies.
- audit-ancestral-reconstruction-validityJudges whether an ancestral sequence or character-state reconstruction has earned its reliability claims, emitting a per-axis pass/flag/fail verdict.Family B critique
- audit-batch-design-validityReferees a batch-aware analysis for validity, returning sourced per-axis verdicts and empirical calibration requirements.Family B critique
- audit-gene-family-birth-deathAudits a gene-family birth-death analysis for method validity: assumptions met, the named method real and appropriate, the error rate actually controlled.Family B critique
- audit-gene-loss-callDecides whether a gene-loss call is an established loss or a technical artifact, working ordered gates in which silence is never taken as evidence.Family B critique
- audit-method-validityReasons about an analysis's method validity from its method + data-handling trace, emitting a verdict that gates certification — never CERTIFY on its own.Family B critique
- audit-pangenome-claimCatches pangenome claims that are silently wrong — artifact-inflated, mis-partitioned, over-claimed, or reference-free in name only — gate by gate.Family B critique
- audit-positive-selection-claimAudits a dN/dS positive-selection claim at the layer beneath the p-value, then passes it, flags it with a remedy, or rejects it as invalid as stated.Family B critique
- audit-synteny-claimScreens a synteny or rearrangement conclusion for the cardinal sin of an artifact read as biology, flagging anything that flips with tool stringency.Family B critique
- audit-wgd-inferenceReferees a whole-genome-duplication inference against the known artifact axes; a claim that cannot clear an axis is flagged, never silently passed.Family B critique
- audit-whole-genome-alignmentReferees a whole-genome alignment and the biology drawn from it, flagging rather than passing any claim it cannot clear from the evidence supplied.Family B critique
- separate-batch-from-biologySeparates technical variation from biology with explicit design checks, pinned tooling, reproducible adjustment, and a referee handoff.Family A construct
Research
The grounding corpus — faithful, recoverable source notes that back the Molds. Not reproductions.
- Books27 own-words summaries of external textbooks, each carrying its license and attribution.
- Papers76 faithful research notes on primary papers, each with citation and a link to the source.
- Tutorials8 notes on tool vignettes — procedure, exact defaults, pinned to the version summarized.
- Patterns4 corpus leaves — named invalidity signatures and established-good practice the referee Molds delegate to.
Reference
The project's own vocabulary and rationale — how the Foundry is built and why.