| InfrastructureIntent |
Abstract base class for any infrastructure request (provision, grant access, deploy config). It contains service name, environment, requester, and provenance metadata. |
Core |
core_concepts.md |
| HealingIntent |
Immutable container for a governance recommendation. Contains action (approve/deny/escalate), risk score, justification, confidence, and full decision trace. |
Core |
governance.md |
| RiskScore |
Bayesian posterior mean (0–1) computed by the RiskEngine. Blends conjugate prior, HMC prediction, and optional hyperprior. |
Mathematics |
mathematics.md |
| GovernanceLoop |
Orchestrator that processes an InfrastructureIntent and returns a HealingIntent. Integrates cost, policy, risk, epistemic uncertainty, predictive foresight, and memory. |
Core |
governance.md |
| RecommendedAction |
Enum: APPROVE, DENY, ESCALATE, DEFER. Determined by expected loss minimisation, overridden by policy violations or high epistemic uncertainty. |
Decision |
governance.md |
| Expected Loss Minimisation |
Decision rule that selects the action with smallest expected loss, using cost constants (COST_FP, COST_FN, etc.) and optional CVaR for tail risk. |
Decision |
governance.md |
| Escalation Gates |
Mechanical validations in the enterprise layer: license, confidence, risk, rollback, causal. All required gates must pass for execution. |
Enterprise |
enterprise.md |
| Execution Ladder |
Hierarchy of autonomy levels: Advisory → Operator Review → Supervised → Autonomous (Low) → Autonomous (High) → Novel Execution. Enforced by enterprise gates. |
Enterprise |
enterprise.md |
| Conjugate Beta |
Online Bayesian model with Beta priors per action category. Posterior parameters (α,β) updated after each outcome. Provides fast, interpretable risk updates. |
Mathematics |
mathematics.md |
| HMC Logistic Regression |
Offline Hamiltonian Monte Carlo (NUTS) model that captures complex patterns (time of day, user role, environment) from historical data. |
Mathematics |
mathematics.md |
| Hyperprior Shrinkage |
Hierarchical Beta model (α₀,β₀ ∼ Gamma(2,1)) that shares statistical strength across categories. Fit via SVI in Pyro. |
Mathematics |
mathematics.md |
| Epistemic Uncertainty |
Composite score ψ = 1 - (1‑hallucination)(1‑forecast_uncertainty)(1‑data_sparsity). When ψ > threshold, forces escalation. |
Decision |
governance.md |
| Lyapunov Stability |
Discrete‑time stability guarantee for healing actions. Quadratic Lyapunov function V(x,r) decreases along trajectories if a stabilising action exists. |
Mathematics |
mathematics.md |
| Temporal Reliability |
Optional external layer for cross‑session reliability aggregation. Uses session_id, observed_at, time windows, decay. Does not affect in‑session scoring. |
Extension |
temporal_reliability.md |
| RAGGraphMemory |
Semantic memory using FAISS index (IndexFlatL2 in OSS) to retrieve similar past incidents. Enhances confidence via similarity‑weighted boosting. |
Core |
design.md |
| CVaR |
Conditional Value at Risk. When enabled (USE_CVAR=true), approve loss uses average of worst α=0.05 fraction of loss samples – penalises tail risk. |
Decision |
constants.py |
| IntentSource |
Enum indicating origin of a HealingIntent: OSS_ANALYSIS, HUMAN_OVERRIDE, RAG_SIMILARITY, INFRASTRUCTURE_ANALYSIS, etc. |
Core |
healing_intent.py |