Core Principle: IFA (Intelligence From Architecture). AI enriches and informs. Humans decide. Everything is auditable. Non-compliant execution is structurally impossible.
TauDIL is runtime constitutional infrastructure positioned above probabilistic intelligence systems. It evaluates security, governance, admissibility, authority, and integrity before an AI provider is called and before operational consequence propagates.
Intelligence may generate possibilities.
Architecture determines which possibilities are allowed to execute.
Most systems remain probabilistic, behaviorally aligned, and observationally monitored. ⌬TauDIL is different: governance is a deterministic runtime architecture positioned above intelligence.
Seven enforcement layers plus a persistence boundary. AI is invoked only after deterministic security and governance evaluation succeeds.
┌──────────────────────────────────────────────┐
│ L1 — USER BOUNDARY │
│ Admin UI · User Panel · JWT Auth │
└──────────────────────────────┬───────────────┘
↓
┌──────────────────────────────────────────────┐
│ L2 — SECURITY LAYER │
│ USE · DriftAligner · TRCP-Phi κ · X-Domain │
└──────────────────────────────┬───────────────┘
↓
┌──────────────────────────────────────────────┐
│ L3 — GOVERNANCE PIPELINE (DETERMINISTIC) │
│ RulesEngine · DAL · Session · UAE · PACE │
└──────────────────────────────┬───────────────┘
↓
┌──────────────────────────────────────────────┐
│ L4 — KNOWLEDGE LAYER (CKG-RAG) │
│ Ontology · CKGs · Registry · SYGON · CHRONOS │
└──────────────────────────────┬───────────────┘
↓
┌──────────────────────────────────────────────┐
│ L5 — ESCALATION & ADMISSIBILITY (HITL) │
│ Triggers · Timeouts · Staff Dashboard │
└──────────────────────────────┬───────────────┘
↓
┌──────────────────────────────────────────────┐
│ L6 — AI PROVIDER BOUNDARY │
│ Claude · Mistral · Cohere · Gemini · ... │
└──────────────────────────────┬───────────────┘
↓
┌──────────────────────────────────────────────┐
│ L7 — AUDIT BOUNDARY │
│ Mirror · Chronicles · TRVC · Compliance │
└──────────────────────────────────────────────┘Entry points into the system:
authority_level embedded in tokenEvery message passes through this before anything else.
block_threshold=0.35, critical_floor=0.56; skipped for AI outputdomain_access_rules lookup, default deny on sensitive domainsNo AI in this path. Deterministic decision layer and authority gating.
< 0.6 → human review (no AI call) · 0.6–0.8 → uncertain flag · ≥ 0.8 → full authority path
{id}_full_memory.json_pace_enrichedGrounds all AI responses in verified facts.
CompanyKnowledgeGraph, instance: default)domain_registry.db; ALLOW/DENY per domain pair/api/v1/ckg/reasonHuman enforcement is architectural.
AI is called only after all governance checks pass.
skip_manipulation=TrueEvery decision permanently and immutably recorded.
aelthered.db; 8 event categories; SHA-256 chain; watchdog; key rotation; tamper responseConcrete execution sequence for User Panel chat messages. Block paths are structural outcomes of security and governance evaluation — not “errors.”
blocked:true · Cross-domain → 200 blocked:true · Rules BLOCK → status:BLOCKED · Rules REDACT → data replaced · TRVC NO_CONSENSUS → Chronicle
⌬TauDIL separates operational state, audit lineage, and knowledge snapshots into explicit persistence boundaries.
taudil.db — users, domains, rules, settings, sessions
aelthered.db — audit chronicles, hash chain
ckg/*.json — graph snapshots with versioning
sygon state — semantic engine, drift history
domain_registry.db — cross-domain access rules
{id}_full_memory.json — per-user session memory
We review your domain authority model, admissibility conditions, cross-domain constraints, and audit obligations — then map them to executable TauDIL invariants.