Dependency Governance

Lab depends on Foundation, Core, Knowledge, and Pydantic. The graph supports durable consequence records without importing Runtime orchestration or Intelligence recommendation policy into laboratory ownership.

flowchart LR
    F["Foundation"] --> L["Lab"]
    C["Core"] --> L
    K["Knowledge"] --> L
    P["Pydantic"] --> L
    I["Intelligence decision"] -. "explicit handoff contract" .-> L
    L -. "must not import" .-> R["Runtime"]

Current Dependency Contract

Dependency	Permitted role	Authority that remains external
bijux-proteomics-foundation	identifiers, provenance, shared lifecycle contracts	canonical primitive meaning
bijux-proteomics-core	scientific result and assay-relevant data contracts	scientific computation and inference
bijux-proteomics-knowledge	evidence, biological grounding, promoted-record handoff	evidence custody and claim truth
Pydantic	validation and serialization of plans, handoffs, and outcomes	lifecycle validity and scientific acceptance

Intelligence is not an installed dependency. Recommendation inputs must cross a reviewable handoff boundary so Lab can assess feasibility without inheriting or reimplementing ranking policy. Runtime is also absent: Lab describes executable work and records consequence, while Runtime owns execution mechanics.

Admission Test

A dependency is acceptable only when:

1. It supports assay design, readiness, scheduling, handoff, observation, or outcome reconciliation.
2. Lifecycle transitions remain explicit in Lab code.
3. Shared and scientific meanings remain with their upstream owner.
4. Failure, absence, and degraded operation are recordable without fabrication.
5. Operator-visible artifacts remain serializable and reviewable.
6. It does not start work, mutate Knowledge custody, or apply recommendation policy as an import side effect.

Hardware SDKs, schedulers, inventory services, and laboratory information systems belong behind adapters. Their availability and identifiers may affect readiness, but their object models must not define the durable Lab contract.

Rejection Signals

Reject a dependency that hides mutable global capacity, converts a technical failure into a biological result, requires network access to read a retained record, bypasses transition validation, or makes generated convenience objects the owner of protocol and outcome meaning.