Dependencies and adjacencies

Core depends on Foundation for portable identity and document behavior, and on scientific libraries for numerical, sequence, XML, and command-line work. It also contains narrow integration seams that exchange records with Runtime, Knowledge, Intelligence, and Lab. Those seams compose an evidence chain; they do not make Core the owner of execution, evidence custody, decisions, or laboratory operations.

Dependency shape

flowchart LR
    foundation["Foundation\nidentity and documents"] --> core["Core\nscientific meaning"]
    scientific["NumPy · Biopython · defusedxml"] --> core
    cli["Click"] --> core
    core --> runtime["Runtime seam\nexecution request"]
    core --> knowledge["Knowledge seam\nbiological grounding"]
    core --> intelligence["Intelligence seam\ndecision input"]
    core --> lab["Lab seam\nvalidation context"]

The left-hand edges provide implementation prerequisites. The right-hand edges are governed integration seams. A seam may import a public record from its owner for composition, but Core must remain usable for scientific work that does not need that downstream integration.

Direct prerequisites

Dependency	Core use	Constraint
Foundation	JsonModel, identifiers, document schemas, canonical representation	shared representation must not acquire scientific policy
NumPy	numerical arrays and calculations	array behavior must be normalized before it enters a durable public record
Biopython	established sequence and proteomics primitives	adapter behavior and information loss remain explicit in Core contracts
defusedxml	defensive XML parsing	safe parsing does not establish mzML or scientific validity
Click	scientific CLI assembly	command handlers call domain owners rather than implement algorithms
Loguru	diagnostics	logs do not replace typed result, rejection, or provenance records

Cross-package seams

Neighbor	Legitimate Core handoff	Ownership that cannot move into Core
Runtime	validated workflow request, expected artifacts, scientific acceptance policy	provider selection, state, retry, checkpoint, transport, and replay
Knowledge	protein/pathway resolution input and evidence references	source curation, contradiction state, sufficiency, and evidence memory
Intelligence	scientific result and bounded acceptance evidence	ranking, scenario policy, regret, confidence, and recommendation posture
Lab	scientific QC, validation intent, and measurement requirements	readiness, scheduling, custody, observation, and consequence

Several integrated report and flagship acceptance paths touch more than one neighbor. Their job is to preserve references across owners, not to flatten the records into a new Core-owned truth object.

Dependency review

Before adding or widening an edge, answer:

1. can the scientific contract remain valid without the downstream package?
2. is Core importing a public record rather than a neighbor’s internal model?
3. does the resulting artifact retain the neighbor’s identity and authority?
4. can missing optional integration return an explicit result or refusal?
5. is the same behavior reachable without a circular product dependency?

If the proposed edge lets a Core calculation choose a provider, reconcile a contradiction, rank an action, or authorize lab work, route the behavior to the owning package instead.

See scientific package map for domain ownership and dependency governance for validation required when package or third-party dependencies change.