expert-persona

1/9 dimensions tested
Sonnet 4.6

Configuration: Answer as a domain expert reviewer.
Source: gym/skills/expert-persona.md · public · community
Recorded runs: 600 scored outputs · $3.11 measured model spend · runs on 2026-07-14
Access: open prompt · standard tools

FIG. 01

By dimension

Scores /10 · Paired p-values are exploratory and unadjusted

8.27Verify

Compared with baselinen=300 · verdict accuracy · evidence in hand · Δ +0.27 vs Sonnet 4.6 · 10W 2L 288T on 300 paired · p=.039 (unadjusted) · $0.0041/task · 2026-07-14

NEI class 0.75 vs naked 0.70; 151 output tokens per call Test: exact McNemar on discordant pairs.

Run provenance

Report path: outputs/arena/scifact_sonnet_evidence.json
Task pack: scifact-dev-300
Run: arena-sonnet-4.6

from memory5.10 · n=300 · Δ +0.13 · p=.557 (unadjusted) · $0.0063 · 2026-07-14 · NEI class 0.09 vs naked 0.10; 387 output tokens per call

—Ground

Not yet tested.

—Find

Not yet tested.

—Extract

Not yet tested.

—Analyze

Not yet tested.

—Compute

Not yet tested.

—Synthesize

Not yet tested.

—Review

Not yet tested.

—Hypothesize

Not yet tested.