Planning and promotion policy¶
Lab configuration is carried by typed plans, policies, profiles, and declared resources. There is no hidden package-wide configuration file. This keeps the difference between scientific priority, operational feasibility, and evidence promotion visible in the artifacts that cross the laboratory boundary.
Planning policy¶
PlanningPolicy weights why a pending assay is worth doing:
- uncertainty reduction;
- contradiction resolution;
- falsification value;
- impact on a blocking decision gate;
- orthogonal confirmation;
- execution burden for blocking and non-blocking assays.
The policy has a stable policy_id. Preserve it with information-gain scores
and the resulting priority queue. A higher candidate score alone does not
guarantee higher lab priority when material, capacity, assay constraints, or a
more discriminating experiment changes the expected value of work.
Design and readiness inputs¶
Experiment design operations take explicit controls rather than process-wide defaults: contrast definitions, replication expectations, randomization seed, batch and fraction capacity, multiplex channel roles, QC insertion interval, plate dimensions, carryover thresholds, and protocol attachments.
Operational readiness is a separate input surface. Declare available sample kinds, materials, reagents, instruments, staff, controls, budget, capacity, lineage, and review clearance. Supplying one class of resource must not clear a different blocker.
Outcome and promotion policy¶
AssayAcceptanceRule defines the metric, comparison operator, threshold, and
unit for one assay. OutcomePromotionPolicy controls base confidence for
passed and failed outcomes plus the uncertainty penalty. Batch promotion adds
quality, reliability, QC, and lineage requirements across a complete batch.
Promotion policy does not rewrite an observation. It determines whether the observation is eligible to become an evidence record and reports blocked assays separately.
Artifact policy¶
LabArtifactProfile and LabArtifactContractRegistry define accepted schema
versions, producers, and artifact kinds for handoff. Canonical envelopes carry
the artifact payload, document schema, and fingerprint; verification detects
payload tampering but does not prove biological correctness.
Configuration invariants¶
- Keep advisory, executable, request, instruction, outcome, and promoted evidence models distinct.
- Persist policy identifiers, design parameters, declared readiness facts, and artifact fingerprints with each handoff.
- Treat randomization seed and channel assignments as experimental provenance.
- A failed assay may produce valuable negative evidence; technical or reproducibility failure usually cannot support the same biological claim.
- Never clear a blocker by mutating a returned plan. Record the new readiness evidence and rebuild the governed handoff.