registrycompare

Role then format against Policy quote first

Both scored on the same cases against claude-fable-5, same judge. The intervals tell you whether the gap is real or noise.

suite reasoning-chainmodel claude-fable-5n=168 public cases each

role-then-format

v2 by @ilya
91.7
86.5 to 95.0 · holdout 90.3

policy-quote-first

v2 by @dara
92.1
86.5 to 95.6 · holdout 91.7
Intervals overlap. The 0.4 point gap is not significant at this sample size.
Metricrole-then-formatpolicy-quote-firstBetter
Public score91.792.1right
Holdout score90.391.7right
Interval width8.59.1left
Cost per run$1.850$1.296right
Median latency2.4s3.0sleft
Sample size168140left

Reading this verdict

Overlap
Two entries are only meaningfully apart when their confidence intervals do not overlap. A higher number with an overlapping interval is a coin flip, not a win.
Holdout
The public score can be gamed by tuning to visible cases. A gap between public and holdout is the tell.
Reproduce
evalness compare \ role-then-format policy-quote-first \ --model claude-fable-5