registrycompare

Role then format against Plain summary

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

plain-summary

v1 by @system
73.2
64.3 to 80.5 · holdout 72.9
Intervals do not overlap. The 18.5 point gap is significant at this sample size.role-then-format wins
Metricrole-then-formatplain-summaryBetter
Public score91.773.2left
Holdout score90.372.9left
Interval width8.516.2left
Cost per run$1.850$1.426right
Median latency2.4s2.3sright
Sample size168112left

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 plain-summary \ --model claude-fable-5