Integration Seams

Lab is the authority for operational readiness and assay handoff, not for discovering scientific signals or controlling instruments. The central integration contract is a closed, reviewable loop: recommendation enters with evidence; execution leaves with controls; observation returns with provenance; reconciliation informs the next decision.

flowchart LR
    F[Foundation contracts] --> P[Lab planning]
    C[Core evidence] --> P
    K[Knowledge context and caveats] --> I[Intelligence recommendation]
    C --> I
    I --> P
    P --> R[Readiness and lifecycle gates]
    R --> H[Controlled handoff]
    H --> E[External lab execution]
    E --> O[Observed outcomes and QC]
    O --> X[Reconciliation]
    X --> K
    X --> I
    T[Runtime transport and persistence] -. carries .-> H
    T -. carries .-> O

Producer and consumer obligations

Seam	Incoming contract	Lab-owned decision	Boundary
Foundation → Lab	identifiers, JSON models, provenance, outcomes, and compatibility primitives	operational models constructed from shared contracts	Lab does not redefine common identity or serialization semantics
Core → Lab	measured candidates, uncertainty, design context, and assay-relevant evidence	whether evidence needs can be mapped to concrete wet-lab actions	Lab does not rerun discovery algorithms or upgrade weak measurements
Knowledge → Lab	resolved entities, curated context, coverage, conflicts, and citations	which caveats and controls must survive into the handoff	a reference association is not proof of feasibility
Intelligence → Lab	bounded recommendation, rationale, expected information gain, and unresolved uncertainty	practicality, priority, burden, readiness, scheduling, and refusal	ranking is advisory until Lab evaluates operational constraints
Lab → external execution	executable plan, protocol version, samples, controls, caveats, field mapping, and artifact integrity	whether the packet is responsible and complete enough to deliver	instrument control and physical execution remain external
External execution → Lab	observations, run QC, failures, protocol context, and artifact identity	acceptance, rerun, reliability, and promotion readiness	missing or failed observations are not converted into success
Lab → Knowledge/Intelligence	provenance-preserving evidence candidates and planned-versus-observed feedback	operational interpretation of what happened and what should be reconsidered	Lab does not curate scientific truth or set advisory policy
Runtime ↔ Lab	persistence, queues, transport, and execution envelopes	semantic validation of Lab artifacts	Runtime carries state but does not own readiness or promotion decisions

Gate sequence

Scientific merit is necessary but insufficient. The design gate checks replication, contrasts, randomization, layout, and controls. The readiness gate checks evidence, provenance, materials, staffing, capacity, and backlog. The handoff gate checks risk, protocol attachments, authority, serialization integrity, and export loss. The outcome gate checks acceptance rules, QC, reliability, and failure class before any promotion.

Each gate can refuse progression with structured reasons. A refusal is a valid operational result: it preserves why work could not responsibly proceed and what evidence or control would change the decision.

Round-trip integrity

The plan, exported packet, observed result, and reconciliation report must remain linkable. Canonical artifact envelopes detect payload drift; LIMS mappings disclose fields that cannot be represented; outcome records retain requested-versus-observed context. Feedback can change later prioritization, but it never mutates the historical recommendation or lab record.