Dependencies and adjacencies

Intelligence declares Foundation, Core, and Knowledge as product prerequisites: shared documents preserve decision identity, Core defines scientific inputs, and Knowledge supplies evidence custody. Narrow review surfaces also consume Runtime and Lab records so a recommendation can be challenged against execution and downstream consequence without moving those meanings into decision policy.

flowchart LR
    foundation["Foundation\ndecision identity"] --> intelligence["Intelligence\npolicy and judgment"]
    core["Core\nscientific result"] --> intelligence
    knowledge["Knowledge\nevidence bundle"] --> intelligence
    runtime["Runtime\nexecution record"] --> reviews["review seam"]
    lab["Lab\nconsequence record"] --> reviews
    reviews --> intelligence
    intelligence --> advisory["recommend · downgrade · escalate · refuse"]

Prerequisite contract

Dependency	Intelligence consumes	It must not redefine
Foundation	candidate, claim, program, and review identities; canonical records	cross-package identity or serialization rules
Core	scientific outputs, QC, acceptance posture, and domain constraints	calculations, thresholds, or workflow meaning
Knowledge	sources, support, contradiction, gaps, and evidence sufficiency	evidence history or truth relationships
NumPy	numerical ranking, sensitivity, and scenario calculations	policy identity or explanation semantics
Pydantic	strict decision and review records	human authority or scientific validity

Review adjacencies

Runtime and Lab appear in outsider review, independent-rerun, release-candidate, and workflow-authority surfaces. These records apply pressure to a decision:

Adjacent record	Decision question	Authority retained by neighbor
Runtime run bundle	did the evidence-producing work execute under the claimed conditions?	provider, state, artifacts, comparison, and replay
Lab readiness record	is the proposed action feasible and controlled?	readiness, custody, scheduling, and refusal
Lab consequence record	did the observed outcome support, contradict, or leave the action unresolved?	measurement, QC, deviation, and outcome identity

Intelligence may change a ranking or posture in response. It may not edit the run or lab record to make the decision appear stable.

Dependency placement rules

Proposed behavior	Correct owner
scientific calculation or acceptance threshold	Core
source curation, contradiction, or evidence sufficiency	Knowledge
ranking, scenario, counterfactual, regret, confidence, or refusal policy	Intelligence
provider, checkpoint, retry, execution state, or artifact transport	Runtime
assay readiness, scheduling, handoff, observation, or consequence	Lab

Review the edge

For any new dependency or imported record, verify that:

1. its owner and immutable identity remain present in the decision record;
2. missing upstream evidence or consequence can trigger downgrade or refusal;
3. the ranking policy is reproducible without private state from the neighbor;
4. explanation distinguishes evidence movement from policy movement;
5. review-only integration does not become a circular core dependency.

Use decision foundations, recommendation challenges, and dependency governance for deeper review.