EXPERIMENT 51 VERIFICATION
==========================
Compilation tested successfully.
Execution tested successfully.

Fresh deterministic seed:
9783790101901

Scenarios: 30
Dataset modes per scenario: 2
Total executions: 60
Compile errors: 0
Runtime errors: 0

Local verification observations only:

A_NO_PERTURBATION
- scenarios: 10
- prediction compliance: 10/10
- improved/worse/equal: 0/0/10
- aggregate delta: 0

B_MINIMAL_LATE
- scenarios: 5
- prediction compliance: 3/5
- improved/worse/equal: 2/0/3
- aggregate delta: +14

C_BROAD_EARLY
- scenarios: 5
- prediction compliance: 5/5
- improved/worse/equal: 1/0/4
- aggregate delta: +10

D_OTHER_PERTURBATION
- scenarios: 10
- prediction compliance: 10/10
- improved/worse/equal: 0/0/10
- aggregate delta: 0

Overall prospective rule accuracy:
28/30 = 93.33%

Rule accuracy:
- PREDICT_EQUAL: 20/20
- PREDICT_NON_IMPROVING: 3/5
- PREDICT_NON_REGRESSING: 5/5

Outcome control for fresh holdout:
- Standard total: 7916
- Dispersion total: 7940
- Aggregate delta: +24
- Improved/worse/equal: 3/0/27

Important interpretation:
The B_MINIMAL_LATE non-improving rule failed in two fresh scenarios, both of
which improved. The Experiment 50 association therefore does not fully
generalise. C_BROAD_EARLY remained non-regressing, while A and D remained
neutral in this local verification.

No thresholds or rules were changed after observing these outcomes.

Official findings require the user's saved run.