Architecture Risks

Lab risk begins when a recommendation, plan, readiness assessment, completed run, or promoted outcome is treated as if it proves all the others.

Risk	Consequence	Control
Advisory-to-execution collapse	Scientifically useful work is scheduled without operational review	Keep advisory plans non-executable and require explicit executable plans
Stale readiness	Capacity, inventory, staffing, controls, or backlog changes after approval	Bind readiness to a dated snapshot and re-evaluate before handoff
Dependency failure	An assay starts before prerequisite evidence or work exists	Validate dependency integrity and cycles
Missing controls	Results cannot distinguish signal from technical or procedural artifact	Make control absence blocking for affected batches
Provenance gap	Observations cannot be traced to approved intent, inputs, or operators	Gate execution and promotion on lineage completeness
Failure-class collapse	Technical failure is interpreted as biological failure, or vice versa	Preserve normalized result state and failure class
Censoring blindness	Below-detection measurements are treated as ordinary quantitative values	Retain detection limit, direction, and censoring flag
Promotion leakage	Raw or weak observations enter knowledge as accepted evidence	Require a separate promotion review and normalized evidence contract
Queue bias	Candidate score crowds out cost, capacity, evidence gaps, and assay burden	Preserve prioritization factors and queue rationale

flowchart LR
    A[Advisory] --> E[Executable plan]
    E --> R[Readiness snapshot]
    R --> H[Handoff]
    H --> O[Observation]
    O --> T[Failure triage and interpretation]
    T --> P[Promotion review]
    P --> K[Knowledge evidence]

The gates deliberately repeat validation at changing boundaries. A scientifically justified assay can remain operationally irresponsible, and a well-executed assay can remain evidentially inconclusive.

Review before evidence promotion

Promotion question	Required evidence	Blocking outcome
did authorized work produce this observation?	plan, handoff, custody, operator, sample, batch, and artifact lineage	missing or conflicting provenance
was the measurement technically valid?	controls, QC, calibration, detection limits, censoring, deviations, and acceptance policy	technical or uninterpretable failure
is the result repeatable enough for its intended use?	replicate, batch, dispersion, missingness, and reproducibility assessment	unresolved reproducibility failure
what proposition can the observation address?	requested question, measured entity, direction, magnitude, context, and granularity	observation does not answer the claim being promoted
how should adverse or null evidence be represented?	failure class, uncertainty, context, and relationship to the requested result	pressure to omit a valid non-confirming outcome
who accepts the evidence relationship?	named reviewer, promotion policy, decision, limitations, and destination Knowledge identity	no accountable evidence authority

Promotion may accept, qualify, contradict, reject, or hold an observation. It must not convert technical failure into biological evidence, hide a valid null result, or strengthen the original recommendation merely because the assay was completed.