Assessment Crash TestRed-team structured Python assessments before students take them.
Offline
Audit idle.
Assessment editor

Nested Python Loop Trace

Golden demo

Expected answers are required so scoring remains deterministic. GPT-5.6 attacks the assessment; it does not grade itself.

Assessment cases12 deterministically graded cases
Case 1
Case 2
Case 3
Case 4
Case 5
Case 6
Case 7
Case 8
Case 9
Case 10
Case 11
Case 12
UNSUPPORTEDView evidence
Not scored

Edit or load an assessment, then run a crash test.

Edit or load an assessment, then run Attack -> Verify -> Repair.

AI attack scoreNot testedNo AI attempt has been graded.AI solvabilityNot testedNo solvability calculation is available.Student decisions requiredNot testedNo evidence checkpoint requirements have been verified.
Deterministic rule

Evidence checks have not run

Run the crash test to grade cases and verify checkpoint requirements.

Audit workflow

Attack, verify, repair, re-test

Attack

GPT-5.6 attempts the current assessment definition.

Before and after

State change from original to revised protocol

Generated evidence checkpoint

Checkpoint unavailable

Run the crash test to generate a task-specific checkpoint.
Technical evidence (not tested)

Case-by-case grading and rule checks appear after an audit finishes.