Repository navigation
TEST 112
Feed validated per-input assessments to aggregation without invoking a model. Include all-pass, fail plus pass, fail plus inconclusive, only inconclusive, empty/missing required inputs, and malformed verdicts. Put more than 32 distinct informational findings in earlier frame results and a blocking finding in a later result. Repeat with blockers exceeding the output capacity, duplicate prose on different frame identities, repeated identical findings, and decision-relevant uncertainties. Re-run identical inputs to check deterministic output; separately supply a passing model decision after an independent deterministic check failed.
A required failing input never becomes pass or loses all supporting blockers. Non-passing incomplete input cannot count as completed coverage. The complete aggregate passes only with all required inputs passed and no contradictory blocker or coverage gap. Stable ordering/deduplication preserves distinct evidence references; truncation of optional detail is disclosed. Too much required support causes an incomplete/non-success result rather than an unjustified verdict. Final schema and decision consistency validation run on the aggregate, not just its individual inputs. A model cannot override the deterministic failure. Regression artifacts and executed results are recorded in REQ-110 for merged PR #238; tests preserve currently available evidence-category references, while frame-level attribution remains planned in REQ-112. No inference, native-build, or live qualification result follows from these offline cases.