Hard Blockers Override Aggregate Scores

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

A teaching audit totals 95, but a sensitive endpoint lacks a required authorization gate. An adopted blocker policy prevents delivery regardless of documentation/performance scores.

Mechanism

Establish project-owned blocker predicates and applicability, then inspect evidence. Retain raw aggregate or apply its documented cap, separately reporting readiness/findings. Confirmed blockers mean not ready; unknowns follow declared policy. State trigger, effect, repair/recheck conditions and reevaluate after repair without weakening policy.

Bad example

Approve because 95 is high despite absent authorization, or hide the score and provide no reason for failure.

Good example

Report raw teaching score 95, readiness=blocked and authorization evidence. Apply approved scoring/blocker rules and specify unauthorized rejection checks. Do not approve through compensation from unrelated scores; preserve unknowns.

Why the change matters

Hard gates define noncompensable failures. Separate score/readiness preserves overview and independently mandatory conditions.

Observable expectation

Teaching unauthorized access to sensitive data triggers blocking despite high other scores. After repair, verify rejection and valid authorized behavior before rejudging. Unrun gates remain unverified and policy must be cited, not model-invented.

Limits

Not every concern is a blocker; use actual policy/impact. Frozen 69/84 caps are a specific rubric, not universal thresholds. Model scores are not security tests, and this example probes no real sensitive system.