save checkpoint
This commit is contained in:
@@ -0,0 +1,42 @@
|
||||
# Policy Studio Copilot Attestation
|
||||
|
||||
## Module
|
||||
Attestor
|
||||
|
||||
## Status
|
||||
VERIFIED
|
||||
|
||||
## Description
|
||||
Policy draft attestation types for AI-generated lattice rules with test case generation and signed snapshots.
|
||||
|
||||
## Implementation Details
|
||||
- **AI Policy Draft Statement**: `src/Attestor/__Libraries/StellaOps.Attestor.ProofChain/Statements/AI/AIPolicyDraftStatement.cs` -- in-toto statement wrapping an AI-generated policy draft.
|
||||
- **AI Policy Rule Draft**: `Predicates/AI/AIPolicyRuleDraft.cs` -- individual AI-generated policy rule with condition, action, and severity.
|
||||
- **Policy Rule Type**: `Predicates/AI/PolicyRuleType.cs` -- enum of rule types (Gate, Advisory, Informational).
|
||||
- **Policy Rule Test Case**: `Predicates/AI/PolicyRuleTestCase.cs` -- auto-generated test case for validating the drafted policy rule.
|
||||
- **Policy Validation Result**: `Predicates/AI/PolicyValidationResult.cs` -- result of validating the drafted policy against test cases.
|
||||
- **AI Authority Classifier**: `Predicates/AI/AIAuthorityClassifier.cs` (with `.PolicyDraft`, `.PolicyDraftScore`) -- classifies AI-generated policy drafts by authority level (Suggestion, EvidenceBacked, AuthorityThreshold).
|
||||
- **AI Model Identifier**: `Predicates/AI/AIModelIdentifier.cs` -- identifies the AI model that generated the policy draft (provider, model, version, weights digest).
|
||||
- **AI Decoding Parameters**: `Predicates/AI/AIDecodingParameters.cs` -- captures model parameters (temperature, top-p, max tokens) for reproducibility.
|
||||
- **DSSE Signing**: `Signing/ProofChainSigner.cs` -- signs policy draft attestations for immutability.
|
||||
- **Tests**: `__Tests/StellaOps.Attestor.ProofChain.Tests/AIPolicyDraftTests.cs`
|
||||
|
||||
## E2E Test Plan
|
||||
- [ ] Create an `AIPolicyDraftStatement` with multiple `AIPolicyRuleDraft` entries and verify the statement structure
|
||||
- [ ] Generate `PolicyRuleTestCase` entries for a drafted rule and verify they cover positive and negative scenarios
|
||||
- [ ] Validate the drafted policy via `PolicyValidationResult` and verify it passes all generated test cases
|
||||
- [ ] Classify the policy draft via `AIAuthorityClassifier.PolicyDraft` and verify the authority level based on evidence quality
|
||||
- [ ] Verify `AIModelIdentifier` captures the model that generated the draft (e.g., provider="anthropic", model="claude-opus-4-6")
|
||||
- [ ] Verify `AIDecodingParameters` captures reproducibility parameters (temperature, seed)
|
||||
- [ ] Sign the policy draft statement via `ProofChainSigner` and verify the DSSE envelope is valid
|
||||
- [ ] Create policy drafts of different `PolicyRuleType` (Gate, Advisory, Informational) and verify type-specific behavior
|
||||
|
||||
## Verification
|
||||
|
||||
| Check | Result |
|
||||
|-------|--------|
|
||||
| Tier 0 - Source Verification | PASS |
|
||||
| Tier 1 - Build + Code Review | PASS |
|
||||
| Tier 2 - Behavioral Verification | PASS |
|
||||
| Verified Date | 2026-02-13 |
|
||||
| Run ID | run-001 |
|
||||
Reference in New Issue
Block a user