Code Navigation

Bijux Proteomics Intelligence turns governed scientific evidence into reviewable analytical judgment. Navigate the package as a reasoning chain: candidate state, claim support, evidence posture, explicit policy evaluation, recommendation, review, and learning.

Reasoning map

Question	Start here	Owned result
What is being evaluated?	candidates/schema.py and candidates/records.py	candidate identity, evidence references, metrics, and state
Is the candidate valid and eligible?	candidates/validation.py, quality.py, filters.py, and selection.py	validation, QC posture, exclusions, and shortlist membership
Why is one candidate ranked above another?	candidates/ranking.py, fingerprints.py, and lifecycle.py	factor contributions, policy lineage, robustness, stability, drift, and movement
Is a claim actually supported?	claims/support.py, contradictions.py, and refusal.py	graph-backed support, conflicting evidence, and governed claim refusal
What could overturn the claim?	falsifiers.py and belief_audit.py	claim-specific falsifiers, evidence for and against, uncertainty, and next checks
What does the evidence mean in context?	interpretation/	bounded readings of quantitative, contrast, PTM, pathway, contaminant, structure, and run evidence
What action is favored under a stated policy?	judgment/policies.py, scenarios.py, recommendations.py, and paths.py	advance, hold, redesign, or scale-up analysis with reasons and alternatives
What should a reviewer receive?	reviews/decision_briefs.py, report_contract.py, boards.py, and candidates.py	aligned claim, ranking, contradiction, rationale, and follow-up artifacts
How is public scrutiny supported?	reviews/benchmark_reviews/, external_review_kits.py, independent_reruns.py, and public_scrutiny.py	benchmark-linked review packets, caveats, rerun routes, and known exclusions
How does observed outcome influence future posture?	learning/adaptation.py and learning/refinement/	prospective adaptation, convergence, and stagnation evidence

Fast reading route

Start with governance/charter.py. It defines thirteen analytical bands and the five capabilities they serve: prioritization, contradiction handling, review reasoning, interpretation discipline, and recommendation. Then inspect public_api.py before importing from package root; the root surface is curated and is not a substitute for owner-module APIs.

For one decision, trace identifiers in this order: candidate → evidence → claim → policy → scenario → recommendation → review disposition. Compare every score with its factor rows and every claim with support, contradiction, refusal, and belief-audit entries. A final recommendation without these joins is not the complete intelligence result.

Benchmark modules under judgment/ test decision behavior—blinded challenges, counterfactuals, regret, sensitivity, policy comparison, and confidence—not scientific parsing. Core owns the upstream scientific artifacts, while knowledge owns curated evidence graphs and references.