Skip to instrument
SentinelRuntime assurance research
Frozen planner. Observable intervention.
Daniel Wahnich
03 / Experimental outcomesEvidence / 6dad3b6c

NeuroNCAP · full 14-scene power run

OFF2.12
→
Released union2.91

Benchmark score · 799 / 800 episodes retained

Released union − OFF / score difference+0.783
-10+1

95% CI [+0.605, +0.928]

Simulation measurements. The scope and comparator remain attached to each result.

Sentinel / evidence inspector

Research record