Module Map

bijux-proteomics-intelligence turns evidence into inspectable decision support. It owns candidate comparison, skeptical challenge, policy-governed judgment, and reviewer-facing explanations; it does not create experimental evidence or authorize laboratory execution.

flowchart LR
    Q[query] --> C[candidates]
    E[evidence posture] --> C
    C --> J[judgment]
    X[contradictions and falsifiers] --> J
    J --> N[next steps or refusal]
    J --> R[reviews]
    I[interpretation] --> R
    B[belief audit] --> R
    R --> L[learning]

Owner modules

Family	Responsibility
candidates	Candidate schemas, records, validation, quality, fingerprints, filters, ranking, selection, lifecycle, and stores
claims, posture	Claim support and the explicit evidential stance used for evaluation
contradictions, falsifiers, belief_audit	Adverse evidence, disconfirmation tests, and traceable changes in belief
judgment	Policies, scenarios, decision paths, recommendations, and decision benchmark suites
interpretation	Decision-facing projections of quantitative, pathway, PTM, contaminant, run, contrast, and structural evidence
next_steps, refusal	Bounded follow-up actions and explicit non-recommendation when prerequisites fail
reviews	Decision briefs, boards, outsider packets, rerun kits, public-scrutiny records, and workflow-authority contracts
learning	Controlled adaptation from reviewed outcomes rather than silent policy mutation
query, governance	Decision-question contracts and package charter constraints

The package root lazily exposes these fourteen owner modules. Concrete types and functions remain under their owner, which keeps ranking, interpretation, and review APIs distinguishable.

What intelligence does not own

Core owns scientific calculations. Knowledge owns durable evidence memory and source grounding. Lab owns experimental planning and readiness. Runtime owns processes, services, artifacts, and operational state. Intelligence may consume records from all appropriate lower layers and emit a recommendation packet, but a recommendation is neither new evidence nor an execution command.