Dependencies and adjacencies

Knowledge is downstream of shared document contracts and scientific identity, but upstream of recommendation policy. Published metadata declares Foundation and Core as product prerequisites. The current Knowledge source tree imports Foundation contracts directly; it keeps Core as the neighboring scientific authority rather than importing scientific algorithms into evidence custody.

flowchart LR
    foundation["Foundation\nidentity and documents"] --> knowledge["Knowledge\nevidence custody"]
    core["Core\nscientific subjects and results"] --> knowledge
    sources["literature · databases · ontologies · run artifacts"] --> knowledge
    knowledge --> intelligence["Intelligence\ndecision consumer"]
    knowledge --> lab["Lab\nevidence consumer and outcome producer"]

The arrows into Knowledge describe prerequisites or evidence inputs. The arrows out describe record handoffs, not dependencies that grant downstream packages permission to modify evidence history.

Dependency contract

Dependency	Knowledge use	Boundary
Foundation	JsonModel, identifiers, document schemas, fingerprints, typed outcomes	canonical representation does not decide evidential truth
Core	declared scientific-semantic neighbor for proteins, features, pathways, and workflow results	calculations and scientific acceptance remain Core-owned
Pydantic	strict evidence, claim, resolution, and review records	validation cannot manufacture provenance or context
external sources	observations, assertions, identifiers, and reference relationships	retrieval success and source reputation are not support by themselves

Downstream adjacencies

Consumer	Knowledge supplies	Consumer may add	Consumer may not change
Intelligence	versioned evidence bundle, support, contradiction, gaps, sufficiency posture	ranking policy, scenarios, sensitivity, recommendation, refusal	source content, relationship, provenance, or historical resolution
Lab	evidence context for assay design and readiness	plan, handoff, observation, QC, consequence	the evidence bundle that justified the original request
Runtime	portable review input or artifact reference	execution identity and custody	evidence meaning or contradiction state

Lab observations return through an explicit ingestion route and become new evidence records. They do not edit the recommendation or evidence snapshot that preceded the experiment.

Dependency placement rules

Add a dependency only when it protects evidence meaning or a public handoff. Reject it when it imports a downstream decision or operational policy into the memory layer.

Proposed need	Placement
canonical evidence and claim documents	Foundation contract plus Knowledge model
protein, feature, or pathway scientific meaning	Core-owned contract referenced by Knowledge
source retrieval or ontology adapter	Knowledge reference boundary with version and provenance
candidate ranking or next-action choice	Intelligence
execution retry, provider, or artifact transport	Runtime
scheduling, readiness, or outcome acceptance	Lab

Review the edge

Before accepting a dependency change, verify that:

1. evidence records remain loadable without importing a downstream policy package;
2. source version, biological context, analytical context, and lineage remain explicit;
3. missing or conflicting evidence produces a gap or contradiction rather than a guessed value;
4. a downstream consumer receives immutable record identities;
5. the edge does not create a circular path through decision or laboratory state.

Continue with evidence foundations, architecture, and dependency governance.