registrycompare

Role then format against Verify then answer

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

verify-then-answer

v2 by @marek
88.9
82.2 to 93.3 · holdout 88.9
Intervals overlap. The 2.8 point gap is not significant at this sample size.
Metricrole-then-formatverify-then-answerBetter
Public score91.788.9left
Holdout score90.388.9left
Interval width8.511.1left
Cost per run$1.850$0.968right
Median latency2.4s2.6sleft
Sample size168126left

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 verify-then-answer \ --model claude-fable-5