Spanda

Cognitive & Resilience — Capability Maturity

Promotion criteria and current tier for each functional domain. Canonical architecture: cognitive-resilience-architecture.md.

Current tiers (v1.0+)

Domain	Tier	Rationale
Reflex & Safety	Stable	Reflex catalog, runtime traces, kill-switch integration, REST+gRPC
Operational Coordination	Stable	Fleet, mission runtime, distributed decisions (separate architecture)
Strategic Planning	Stable	Mission planner, governance, deployment profiles (platform services)
Homeostasis Engine	Stable	Entity + scheduler telemetry, drift detection, corrections
Platform Immunity	Stable	Trust/tamper quarantine, registry overlay
Sensory Fusion	Stable	Rule-based multi-source fusion; conflicts feed readiness partial scoring; live sensors Stable-with-env-gate via SPANDA_LIVE_FUSION_SENSORS
Attention Engine	Stable	Health-mapped prioritization, habituation CLI
Operational Memory	Stable	Category refs, persistent episodic store ↔ replay index, browse-by-category UI
Adaptive Learning	Stable	Recovery confidence + strategy preference on mission abort/replan; ML out of scope
Maintenance & Optimization	Stable	OTA Stable; spanda maintenance window + Control Center schedule panel
Control Center tab	Stable	Live REST panels for all domains; promotion gate script

Promotion checklist

Sensory Fusion → Stable

• Rule-based fuse_observations with conflict detection
• Entity-derived sensor bundle (health_status, readiness_status, trust_status)
• REST /v1/autonomy/fusion + gRPC GetAutonomyFusion
• Live multi-sensor pipeline — Stable-with-env-gate via SPANDA_LIVE_FUSION_SENSORS=1 + registered supplier (CLI registers automotive/GPS/IMU/camera proxy); full hardware pipelines via packages remain optional
• Readiness partial scoring wired from fusion conflicts in evaluate_entity_readiness

Operational Memory → Stable

• Five memory categories on Entity.memory_refs
• Registry enrichment (reflex, procedural, episodic, working, semantic)
• REST /v1/autonomy/memory returns refs + OperationalMemoryModel + by_category
• Persistent episodic store linked to replay index (.spanda/autonomy-episodic-memory.json)
• Control Center browse-by-category (not just JSON dump)

Adaptive Learning → Stable

• RecoveryConfidence from orchestrator history
• POST /v1/recovery/recommend + recovery metrics
• Strategy preference surfaced in mission planner abort/replan (verify_mission_assurance_with_recovery, spanda mission verify)
• Documented accuracy thresholds (below); field soak remains organizational

Field accuracy thresholds

Threshold	Value	Use
min_attempts	3	Prefer a strategy only after enough history
escalate_below_rate	0.30	Abort/replan when preferred rate or overall score is below this
Stable field success rate	≥ 0.70	Preferred strategy should hold ≥70% over soak before production claims
Field soak	30 days	Organizational gate — organizational-gates.md

Defaults live in AdaptiveRecoveryPolicy::platform_defaults() and spanda_autonomy::adaptive_recovery::field_accuracy_thresholds.

Maintenance & Optimization → Stable

• OTA deploy Stable
• MaintenanceWindow, SleepMode types in spanda-autonomy
• CLI spanda maintenance window list/set
• Control Center maintenance schedule panel

Control Center Cognitive & Resilience → Stable

• All functional domain panels live
• Strategic Planning summary (governance + deployment profiles)
• Cross-interface REST/gRPC smoke
• RBAC audit for autonomy routes — GETs are read-only (sensitive-read prefix); POST /v1/autonomy/maintenance/windows requires Operate
• Field soak sign-off per organizational-gates.md (pending — organizational)

CI

./scripts/cognitive_resilience_smoke.sh
./scripts/cross_interface_consistency.sh   # includes fusion/memory + SDK probes
./scripts/cognitive_resilience_stable_promotion_gate.sh  # smoke + cross-interface + RBAC notes

Legacy alias: ./scripts/bio_inspired_autonomy_smoke.sh (same checks as smoke).

Related

• feature-status.md — platform-wide matrix
• responsibility-matrix.md — ownership
• known-limitations.md — fusion/recovery caveats

This site is open source. Improve this page.