38. Configuration compiler, physical-model registry, and operating-envelope solver

Deterministic graph compilation
compile(configurationId, contextCheckpoint, packLock):
  verify configuration, registry and pack digests
  materialize item occurrences, ports, connections, software and calibrations
  evaluate effectivity predicates using four outcomes TRUE/FALSE/UNKNOWN/ERROR
  reject duplicate exclusive occupancy and forbidden composition cycles
  validate frames, joints, geometry and configuration-specific transforms

  for each realized connection:
    unify port family, medium/carrier, direction, dimension and protocol
    intersect voltage/current/pressure/flow/speed/torque/load/range constraints
    require a non-empty operating intersection and declared protection/failure behavior

  build structural, support, energy, material, thermal, data, control,
        kinematic, safety and utility subgraphs
  identify strongly connected components; require a registered solver for each loop
  propagate source availability, conversion paths, observability, authority,
        fault overlays and environmental/resource prerequisites to a fixed point
  evaluate registered quantitative constraints and operating envelope
  emit CapabilityCompilationReceipt and minimal explanation DAG

Capability result with machine-checkable explanation
CapabilityCompilationReceipt {
  compilationId, compilerVersion, compilerBinaryDigest,
  configurationId, configurationDigest, contextCheckpoint,
  registryLockDigest, domainPackLockDigest,
  requestedCapabilities[], realizedGraphDigest,
  resultByCapability[{capabilityId, outcome, limitingConstraints[],
                      proofNodeRefs[], unresolvedFacts[], marginSetRef?}],
  solverRuns[], findingCodes[], resourceUsage,
  startedAt, completedAt, outputDigest
}

ProofNode {
  nodeId, ruleOrSolverRef, outcome, inputNodeRefs[], inputRecordRefs[],
  constraintValues?, explanationTermRef, sensitiveEdgePolicy
}

The public explanation is a redacted projection of this DAG. Redaction must not
change the underlying outcome, and a non-interference test must prove hidden
facts cannot be inferred through node counts, errors or timing classes.

Versioned equations and computation receipt
PhysicsModelDefinition {
  modelId, version, modelFamily, equationSetDigest,
  stateVariables[], algebraicVariables[], inputs[], disturbances[], parameters[],
  constraints[], invariants[], eventGuards[], resetMaps[],
  supportedConfigurationPredicates[], validDomainRef,
  initializationProcedureRef, solverProfile,
  verificationEvidenceRefs[], validationEvidenceRefs[]
}

ComputationRun {
  runId, modelRef, exactInputRecordIds[], inputArtifactDigests[],
  configurationId, operationId?, validTime, dataCutoff,
  parameterSetDigest, solverBinaryDigest, solverOptions,
  randomSeeds[], convergenceReport, resultArtifactDigest,
  qualityDisposition, invalidatedBy[]
}

Models are bounded claims
An equation can be correctly implemented and still be invalid for the current scale, Reynolds number, sea state, tire regime, propeller advance ratio, atmosphere, payload, structural mode or controller. Every output reports both numerical convergence and model applicability.

Quantitative and stateful operating envelope
EnvelopeDefinition {
  envelopeId, version, applicableConfigurationPredicates[], modes[],
  entryPredicate, continuePredicate, hardExitPredicate,
  hysteresisPolicy, entryDwell, softExitDwell,
  requiredEvidenceFreshness, requiredProbabilityOrBound,
  monitorDefinitionRef, fallbackRef
}

Predicate := allOf | anyOf | not |
  compare(quantity, operator, threshold, uncertaintyPolicy) |
  within(value, interval) | rateWithin(value, interval) |
  geofence | routeSegment | modeIs | configurationHas |
  resourceRemaining | healthState | linkQuality |
  modelValidatedFor | customRegistered

For constraint g_i(x,c) <= 0:
  inside only when the declared uncertainty rule is satisfied.
  probabilistic profile: P(g_i <= 0) >= p_required.
  bounded profile: entire interval inside entry bound => TRUE;
                   entire interval outside exit bound => FALSE;
                   otherwise UNKNOWN.

Digital twin as an evidence-qualified estimate
TwinStateEstimate {
  twinId, physicalAssetId, configurationId,
  stateVectorDefinitionRef, estimateArtifactDigest,
  covarianceOrEnsembleArtifactDigest, observedAt, dataCutoff,
  estimatorDefinitionRef, modelDefinitionRef, parameterSetDigest,
  inputSegmentRefs[], latencyNs, unobservableStates[],
  divergenceTests[], applicabilityState, modelState
}

modelState in {
  UNVERIFIED, VERIFIED, CALIBRATED, VALIDATED_IN_DOMAIN,
  MONITORED, OUT_OF_DOMAIN, DIVERGED, INVALIDATED
}

Confidence is a vector, never one percentage:
  implementation verification; convergence; parameter identifiability;
  calibration recency; validation-domain coverage; input quality;
  observability; model-form uncertainty; domain proximity;
  clock freshness; divergence status.