Policy model
Core policy set
| Policy | Evaluator | Intervention |
|---|---|---|
POL-TERRITORY-01 |
ScopeEnforcementEvaluator |
Denies targets outside the resolved workspace |
POL-LINEAGE-01 |
LineageEvaluator |
Denies missing or mutable revision identities |
POL-STEWARDSHIP-01 |
StewardshipEvaluator |
Denies declared ownership mismatches |
POL-CONCORD-01 |
ConcordEvaluator |
Adds a glossary-review obligation |
POL-TRANSPARENCY-01 |
TransparencyEvaluator |
Denies declared side effects without audit configuration |
POL-CONSERVATISM-01 |
ConservatismEvaluator |
Requires approval for declared irreversible effects |
POL-VERIFICATION-01 |
VerificationEvaluator |
Requests repair when claims lack evidence |
POL-SOVEREIGNTY-01 |
SovereigntyEvaluator |
Denies actor-type scope mismatches |
POL-MUTATION-01 |
ProtectedPolicyMutationEvaluator |
Denies writes to protected runtime and doctrine paths |
POL-LABELING-01 |
GenericLabelsEvaluator |
Requests a precise artifact name |
POL-IDENTITY-01 |
IdentityDeclarationEvaluator |
Denies undeclared actor or task identity |
Aggregation
The PDP uses the following precedence:
DENY > ESCALATE > REPAIR > ALLOW_WITH_OBLIGATIONS > ALLOW
If an evaluator raises an exception, the PDP returns a SYS-FAIL-CLOSED denial.
If no evaluators run, the PDP also denies.
Policy claims
Each policy claim should map to:
- a concrete evaluator;
- at least one intervention test;
- at least one compliant-path test;
- a gold case when the behavior is part of the public benchmark.
Text that describes a roadmap policy is not evidence that the runtime enforces it.