eval-harness-first

eval-harness-first · tier mission · source repo · MIT

verified

Build the evaluation harness that gates every fine-tuning run — golden sets, per-failure-mode graders, judge calibration, and base-model…

Test results

— jury mean

quality

—

specificity

—

maintenance

—

Jury agreement: —
Scored: 2026-10-04
Score expires: 2026-11-03

Scores are valid for 30 days (the same TTL as the pipeline's expensive scoring layer). After that this page marks them expired until the catalog is rebuilt.

Security scan

Level: verified
Scanned at: 2026-10-04
Findings: 0
Upstream commit: 156b7a5e7a8b
Description length: 278 chars
Files fingerprinted: 3

Capabilities

agent-evaluation

Stacks

Pairs with

llm-calls

Community experiences

What actually happened when you used this skill. No stars, no ratings — just experience reports.

Local prototype: your comment is stored in this browser only — other visitors cannot see it. The queue below is a moderation simulation (try it with ?demo=moderation in the URL); in production a human moderator approves via signed entries shipped with the site build.

Loading…

Share your experience

Moderation queue (local prototype)

Empty.