registry/compare
Ask when ambiguous 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
ask-when-ambiguous
v1 by @nils90.3
84.6 to 94.0 · holdout 89.4
role-then-format
v2 by @ilya91.7
86.5 to 95.0 · holdout 90.3
Intervals overlap. The 1.4 point gap is not significant at this sample size.
Metricask-when-ambiguousrole-then-formatBetter
Public score90.391.7right
Holdout score89.490.3right
Interval width9.48.5right
Cost per run$1.643$1.850left
Median latency1.5s2.4sleft
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 \ ask-when-ambiguous role-then-format \ --model claude-fable-5