Use “Eval — the three-bullet brief” in Recued. Run one pinned instruction-following prompt several times against your own model, score every run the same deterministic way, and publish the distribution to a #tag. The prompt and the scoring are fixed by this recipe version, which is what makes two people's numbers comparable.

How it works 9 steps

Inspect the data fetches, transforms, gates, and output this recipe runs.

Process (9 steps)

■ subject set

Set provider to setting declared provider

■ subject_full set

Set model to setting declared model

✦ answers core.ai.prompt

AI generates a response from a custom prompt

■ bullet_counts split

Split item result into parts

■ segments_per_run count

Count items in item result

■ passes count

Count items in segments per run

■ total count

Count items in setting run slots

■ rate math

Calculate: {{step.passes}} / {{step.total}}

■ headline round

Round rate to 4 decimal places

▼

Output

summary headline

text subject_full

Settings 3 configurable

Configurable at install. Defaults shown — change them anytime in Recued.

run slots setting = 1,2,3,4,5

declared model setting =

declared provider setting =