Laboratory planning and outcome contracts

Lab translates scientific evidence needs into operationally reviewable work and returns observed outcomes to the evidence system. Its central invariant is that scientific advice is not an execution instruction.

Advisory and executable plans

AdvisoryAssayPlan records open evidence gaps, assay recommendations, rationale, and concrete wet-lab actions. It always carries plan_kind="advisory" and executable=false.

ExperimentPlan groups governed assay requirements into ordered batches by blocking status and assay family. It retains evidence gaps, blocking review gates, sample requirements, and per-assay sample kinds. Dependencies constrain order and cycles are rejected.

ExecutableAssayPlan selects one batch and emits instruction IDs, objectives, sample kinds, blocking posture, and preflight checks. It is ready only when its declared review gates and sample-kind blockers are clear.

flowchart LR
    need["evidence need"] --> advisory["advisory assay plan"]
    advisory --> design["design and dependency review"]
    design --> batch["experiment batches"]
    batch --> readiness["materials, controls, staff, instrument, review gates"]
    readiness -->|clear| executable["executable assay plan"]
    readiness -->|blocked| refusal["refusal with missing prerequisites"]

Execution readiness describes the recorded plan and declared resources. It does not predict assay success, sample integrity, instrument performance, or data quality.

Authorization Envelope

Readiness is evidence for an authorization decision, not the decision itself. An authorized handoff must bind the plan, selected batch, instruction set, readiness report, accepted warnings, approving actor, rationale, and approval time. If any bound identity changes, the handoff returns to review.

flowchart LR
    plan["executable candidate plan"]
    readiness["readiness report"]
    authority{"named authorization"}
    handoff["frozen handoff envelope"]
    refused["refusal or deferred batch"]
    plan --> readiness --> authority
    authority -->|approved| handoff
    authority -->|withheld| refused

Design and readiness

Design contracts make contrasts, biological and technical replication, randomization, fractionation, multiplex channels, pooled references, spike-in QC, and carryover risk explicit. Operational readiness combines material and reagent inventory, staffing, instrument availability, controls, provenance, evidence sufficiency, and review backlog pressure.

Missing prerequisites are outputs, not defaults. An irresponsible handoff is represented by LabExecutionRefusal with stable reason codes and required corrections.

Observed outcomes

An assay observation preserves metric, value, unit, raw replicate values, summary statistic, dispersion, QC state, normalization, censoring, and interpretation confidence. An assay outcome distinguishes passed, biological failure, technical failure, reproducibility failure, and inconclusive states, with an explicit failure class and uncertainty.

Batch outcomes retain a rerun policy. Evidence promotion readiness checks the observation and QC posture before converting a lab result into normalized knowledge evidence.

Outcome class	Immediate consequence	Evidence consequence
passed	enter promotion review	may support the requested claim within context and QC limits
biological failure	do not repair as technical error	may contradict or narrow the biological claim
technical failure	follow the technical rerun policy	does not adjudicate the biological hypothesis
reproducibility failure	investigate repeatability and lineage	weakens promotion even if one run looked favorable
inconclusive	preserve ambiguity and define missing evidence	cannot be promoted as confirmation or contradiction

Reconciliation and feedback

Reconciliation compares planned and observed assay state, records deviations, and produces operational follow-up plus intelligence feedback. Feedback can change future priorities, but it must not rewrite the recommendation or plan that preceded the experiment.

Contract invariants

• Advisory plans cannot be mistaken for executable instructions.
• Every executable instruction belongs to one batch and names its prerequisites.
• Dependencies, blockers, and missing resources remain explicit.
• Observations preserve replicates, QC, normalization, censoring, and uncertainty.
• Technical, biological, reproducibility, and interpretation failures remain distinguishable.
• Evidence promotion is gated and provenance-preserving.
• New outcomes append to decision history rather than revising it silently.

The complete review chain is therefore: recommendation identity, advisory plan, design, readiness report, authorization, frozen handoff, returned observations, outcome classification, rerun or promotion verdict, and feedback record. Missing links must remain visible rather than being reconstructed from a final narrative.

These contracts put operational authority with laboratory reviewers while keeping the evidence and decision lineage machine-readable.