Skip to content

feat: surface custom-metric verdicts on the session read path - #183

Merged
vijaykrishna-plivo merged 2 commits into
mainfrom
feat/ao-run-detail-metric-verdicts-main
Sep 10, 2026
Merged

feat: surface custom-metric verdicts on the session read path#183
vijaykrishna-plivo merged 2 commits into
mainfrom
feat/ao-run-detail-metric-verdicts-main

Conversation

@vijaykrishna-plivo

Copy link
Copy Markdown
Collaborator

Change size

files +lines -lines
Production code 1 28 1
Total 1 28 1

No test added — verified with a local API run + confirmed working on dev (dev PR #182).

What

GET /api/sessions/:id now surfaces each call's custom-metric verdicts (metric:<slug>) under the existing eval.verdicts.goal_evaluation shape — latest per metric, joined to the judge registry for the display name, account-scoped, pass/fail only.

Why

Conversation goals were retired in favour of custom metrics. Downstream consumers that read goal_evaluation.goals render metrics where goals used to appear, with no schema change on their side.

Prod promotion of dev #182

Same change already merged + verified on dev (renders as the run-detail "Metrics" section). This is the main promotion.

🤖 Generated with Claude Code

vijaykrishna-plivo and others added 2 commits September 10, 2026 14:23
The session detail endpoint (/api/sessions/:id) now includes each call's
custom-metric verdicts (metric:<slug>) under the existing goal_evaluation
shape — latest verdict per metric, joined to the judge registry for the
display name and scoped to the session's account. Conversation goals were
retired in favour of custom metrics, so goal-shape consumers render metrics
with no schema change.

Verified locally: GET /api/sessions/<id> returns goal_evaluation.goals built
from the session's metric verdicts (pass -> achieved).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGEdzjcwwatWYDkx87GRs9
The session read path now includes metric:<slug> verdicts with verdict
'unknown' (was pass/fail only) and adds a verdict field to each
goal_evaluation entry, so consumers can render 'unknown' (metric did not
apply / insufficient evidence) distinctly instead of a misleading Fail.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CGEdzjcwwatWYDkx87GRs9
@vijaykrishna-plivo
vijaykrishna-plivo merged commit e2fee3d into main Sep 10, 2026
7 of 8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants