registrycompare

Signature echo against Role then format

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=154 public cases each

signature-echo

v2 by @dara
92.9
87.7 to 96.0 · holdout 92.4

role-then-format

v2 by @ilya
91.7
86.5 to 95.0 · holdout 90.3
Intervals overlap. The 1.2 point gap is not significant at this sample size.
Metricsignature-echorole-then-formatBetter
Public score92.991.7left
Holdout score92.490.3left
Interval width8.38.5left
Cost per run$1.850$1.850same
Median latency2.8s2.4sright
Sample size154168right

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 \ signature-echo role-then-format \ --model claude-fable-5