TealTiger Runtime Governance Reference Architecture (v1.2.x)
This page describes the canonical runtime governance architecture for TealTiger v1.2.x. It is intentionally single-page and conceptual, similar to a cloud security reference architecture: it shows what exists, where enforcement happens, and how evidence is produced.Architectural Goals
TealTiger’s reference architecture is designed to:- Enforce policy at execution boundaries
- Produce deterministic governance decisions
- Emit audit‑ready evidence by default
- Avoid centralized control planes or gateways
- Preserve developer velocity
High‑Level Components
1) Policy Authoring & Compilation
- Policies are authored in a locked CEL subset
- CEL is authoring‑only
- Policies compile to canonical JSON rules
- A stable
policy_hashis computed over canonical rules
Canonical rules, not source policy text, are evaluated at runtime.
2) Runtime Enforcement Points (PEP)
TealTiger enforces policy at runtime boundaries:- Model invocation
- Tool invocation
- Output‑to‑action transitions
3) Deterministic Decision Engine (PDP)
At each enforcement point, TealTiger evaluates:- Canonical policy rules
- Execution context
- Classification signals
- Cost metadata (when available)
ALLOWDENYREQUIRE_APPROVALREDACTTRANSFORMDEGRADE
4) Audit Events (Internal)
Each decision produces structured AuditEvents containing:- Decision outcome
- Reason codes
- Risk score
- Execution identity
- Policy hash
5) Evidence Artifacts
AuditEvents are deterministically aggregated into:EvidenceBundle
- Run‑level governance summary
- Decision counts
- Top reason codes
- Cost rollups
LineageGraph
- Execution provenance
- Span relationships
- Safe attributes only
6) TEEC Export Layer (Option A)
TealTiger supports TEEC v0.1 as an export contract:- Canonical reason‑code registry
- Deterministic mapping from AuditEvents
- Vendor‑neutral evidence semantics
Key Invariants
- No runtime CEL evaluation
- No probabilistic enforcement
- No control plane dependency
- Evidence is produced, not reconstructed

