Invariants

Lab invariants prevent advisory work from becoming unauthorized instruction and prevent returned measurements from becoming stronger evidence than their QC, reliability, and promotion record permits.

Laboratory lifecycle invariants

Invariant	What remains true	Observable violation
advisory and executable plans differ	executable status requires a concrete batch, instructions, gates, and authority	assay suggestion is sent as operator instruction
readiness is conditional	material, instrument, capacity, staffing, controls, provenance, risk, budget, and authority are evaluated for the named context	inventory alone produces ready
priority never overrides safety	queue rank and schedule cannot bypass controls, dependencies, compatibility, custody, or authorization	urgent work enters a blocked batch
handoff freezes approved intent	plan identity, instructions, controls, risk, custody, target mapping, and loss report remain reviewable	export changes the approved assay silently
physical execution remains external	package records authorization and receives observations; it does not claim instrument operation	successful serialization is reported as completed experiment
observation is immutable in meaning	returned values, missingness, deviations, failures, lineage, and plan link remain distinct from interpretation	reconciliation edits the measurement record
completion and acceptance differ	operational return, QC disposition, reliability, and biological interpretation are separate	completed assay is automatically accepted
failure and inconclusive are durable	refusal, technical failure, reproducibility failure, biological non-support, and inconclusive remain recoverable	unsuccessful result disappears from the batch summary
promotion is explicit and append-only	named policy creates a downstream disposition without rewriting plan or observation	promoted claim replaces adverse QC history
feedback preserves history	Knowledge and Intelligence receive new evidence and outcome context	later outcome rewrites the earlier recommendation

flowchart LR
    P["advisory plan"] --> R["readiness and authority"]
    R --> H["frozen handoff"]
    H --> X["external physical execution"]
    X --> O["immutable observation"]
    O --> Q["QC and reliability"]
    Q --> M["promotion or hold"]

Identity across the loop

Plan, batch, assay, sample, handoff, observation, and promotion identifiers must allow a reviewer to join records without treating them as the same record. One assay can yield partial observations, deviations, reruns, and several review events while the approved intent remains unchanged.

Failure response

Block the transition whose precondition failed and retain a reason and safe next action. Do not coerce conditional readiness to ready, infer absent observations, or promote around failed QC. A refused or inconclusive transition is a valid laboratory outcome.