Runtime layers¶

13 runtime modules under src/, one per major framework layer. Each is independently importable and tested in isolation.

Module	Camada	Surface	Tests
`src.pipeline`	7	`PipelineOrchestrator`, gates, locks, budget, verification, continuation	smoke + 3 unit modules
`src.ai`	8	`ModelManager`, `CircuitBreaker`, memory stores, RAG, metrics	smoke
`src.privacy`	9	PII detection, classifier, RBAC, audit chain, DSAR, deletion, retention	smoke + 2 unit modules
[`src.meta_learning`]	10	`SkillEvolutionTracker`	smoke
`src.skills`	10	`SkillLoader`, `SkillRouter`, `SkillInvoker`, `SkillVersioning`	smoke
[`src.agent_expansion`]	11	`AgentRegistry`, dynamic registration	smoke
[`src.autonomy`]	12	`ProposalBoard`, `ReflexEngine`	smoke
[`src.growth`]	14 + 19	`LifecycleAssessor` (HomoSapiens), `GrowthTracker`	smoke
[`src.body`]	15	7 hormones, `Baroreceptor`, `LymphaticSystem`	smoke
[`src.anti_obsolescence`]	16	`ObsolescenceRadar`, `DriftDetector`, `RefreshScheduler`	smoke
[`src.regulatory`]	17	LGPD/GDPR/HIPAA/PCI claims, `EvidenceLedger`	smoke
`src.cognitive`	18	`MirrorLearner`, `SleepConsolidator`, `IntuitionEngine`, `MyelinationCache`	smoke
`src.scalability`	19	`CapacityPlan`, `WaveScheduler`, `BackpressureController`	smoke

How they compose¶

   ┌─────────────────────────────────────────────────────────────┐
   │ Orchestrator (cortex / domain orch / task orch)              │
   │                                                              │
   │   1. lookup intent → SkillRouter → SkillManifest             │
   │   2. assemble Capsule (TaskSpec + Constraints + ExpectedOut) │
   │   3. choose model via ModelManager (with CircuitBreaker)     │
   │   4. PipelineOrchestrator.run(capsule):                      │
   │        phase 0 Reception                                     │
   │        phase 1 Reality Anchor                                │
   │        phase 2 Planning (SDD check)                          │
   │        phase 3 Gates (21 absolute rules)                     │
   │        phase 4 Execution (acquires locks, invokes skill)     │
   │        phase 5 Review (auditor-haiku — reviewer ≠ producer)  │
   │        phase 6 Handoff (final Checkpoint + capsule emission) │
   │                                                              │
   │   5. AgentMetrics emit + AuditChain.append                   │
   │   6. MirrorLearner observes trace                            │
   └─────────────────────────────────────────────────────────────┘
                              │
                              ▼
   ┌─────────────────────────────────────────────────────────────┐
   │ Body (hormones) → Baroreceptor / BackpressureController      │
   │ feeds capacity adjustments back into WaveScheduler           │
   └─────────────────────────────────────────────────────────────┘

Cross-cutting:

Every memory write goes through privacy.classifier → privacy.audit_chain.
Every artifact triggers privacy.retention TTL scheduling.
Every failed circuit fires a ReflexEngine signal.
Every successful cycle increments myelin strength for the skill chain.