C.3.A:5 - Evaluation semantics and order (normative)

E-01 (Order). Recover exact declarations and kind compatibility first; check Scope coverage second; when the receiving action is candidate-bearing, check admissibility and evaluate the exact candidate judgment only for an admissible request; then apply R consequences, freshness, and policy thresholds before the separate action disposition.

E-02 (Determinism). With fixed candidates when any, kind/signature editions, slices, bridge/assertion editions, dependencies, and time selectors, the judgments and guard predicates MUST be reproducible. Implicit “latest” is forbidden.

E-03 (Admissibility and three values). Every current C.3.2 or C.3.4 classification consumed by a guard MUST retain true, false, or unknown. Keep not-applicable outside the classification range. Missing evidence or an unavailable declared dependency in an admissible evaluation MUST NOT be coerced to false.

E-04 (Fail-closed without truth rewrite). A required check returning false, unknown, or not-applicable, a missing declaration, non-obtaining relation, unavailable bridge assertion, or uncovered Scope causes refusal. The refusal is not itself a classification value or an assertion that the relevant world-side relation fails to obtain.

E-05 (Weakest link and bridge consequence). Chained bridge assessments use the governed weakest-link rule. The receiving R path records each relied-on bridge/assertion; neither F nor G nor a judgment value is modified.

E-06 (Predicate separation). Declaration compatibility, candidate admissibility, candidate classification, Scope coverage, evidence freshness, bridge applicability, capability fit, and action disposition SHALL remain separately inspectable predicates.