Library and service boundaries

bijux-proteomics-intelligence installs no executable and exposes no HTTP application. Its public contract is a set of typed Python decision-support modules. This is intentional: candidate ranking and recommendation do not own runtime workspaces, authentication, process control, or automatic progression.

The package root exposes fourteen owner modules lazily:

from bijux_proteomics_intelligence import (
    belief_audit,
    candidates,
    claims,
    contradictions,
    falsifiers,
    governance,
    interpretation,
    judgment,
    learning,
    next_steps,
    posture,
    query,
    refusal,
    reviews,
)

Import a concrete class or operation from its owner module. This keeps a decision path visible in code—for example, ranking from candidates, evidence readiness from posture, scenario evaluation from judgment, and adaptation from learning.

Expose intelligence through another surface

Consumer need	Owning integration
Run or reproduce computation	bijux-proteomics-runtime command or API
Render a review packet	Application or reporting layer using a typed intelligence report
Persist evidence and contradictions	bijux-proteomics-knowledge contracts
Schedule a follow-up assay	bijux-proteomics-lab after explicit promotion
Add authentication, rate limits, or request validation	The service that publishes the endpoint

When wrapping intelligence in a CLI or service, serialize the complete typed result. Preserve reason codes, rejected candidates, evidence references, policy lineage, uncertainty, unresolved questions, and human-review flags. Returning only the preferred candidate or action turns qualified decision support into an unjustified command.

build_intelligence_decision_support_envelope() marks a recommendation as advisory by default. Only promote_intelligence_output_to_policy() creates an enforced envelope, and that operation requires a policy identifier, actor, and rationale. A transport must not infer promotion from a successful response or a high score.

For runnable Python examples, see Entrypoints and worked examples.