Skip to content

ci(ai-evals): Send each repeat's results into its own run - #387

Merged
jaydgoss merged 1 commit into
mainfrom
jaygoss/ai-evals-run-id-per-repeat
Oct 30, 2026
Merged

jaydgoss merged 1 commit into
mainfrom
jaygoss/ai-evals-run-id-per-repeat

Conversation

@jaydgoss

@jaydgoss jaydgoss commented Oct 30, 2026 •

Copy link
Copy Markdown
Member

Once getsentry/sentry-for-ai-evals#137 lands, create-runs makes a run per repeat of an :x<count> label and puts each job's run in matrix.job.run-id. The eval jobs here still look up the label and side's run, so every repeat would send into the repeat-1 run, overwriting each other (same scenario names), and the other repeat runs would stay pending until the 30-minute timeout.

  • Use matrix.job.run-id, falling back to the old lookup, so this works before and after getsentry/sentry-for-ai-evals#137 merges.
  • Part of VDY-327.

Testing: ai-evals:llms_txt:sonnet-5.5 on this PR (the fallback path, since getsentry/sentry-for-ai-evals#137 isn't merged).

🤖 Generated with Claude Code

Use the job's run-id from create-runs, falling back to the label and
side's run, so :x<count> labels keep working once sentry-for-ai-evals
makes a run per repeat.
@linear-code

linear-code Bot commented Oct 30, 2026

Copy link
Copy Markdown

VDY-327

@jaydgoss jaydgoss added the ai-evals:llms_txt:sonnet-5.5 Run the llms.txt eval on Sonnet 5.5 only, base and head label Oct 30, 2026
@sentry-evals

sentry-evals Bot commented Oct 30, 2026 •

Copy link
Copy Markdown

Sentry Evals

sentry-for-ai-evals:llms_txt:sonnet-5.5 (Comparison)

Run Status Passed Failed Errored Cost Tokens Duration
Head Passed 1 0 0 $0.28 767.6k 102.41s
Base Passed 1 0 0 $0.53 1.8M 170.58s
Diff 0 0 0 🟢 -$0.25 (-46.7%) 🟢 -986.4k (-56.2%) 🟢 -68.17s (-39.9%)

jaydgoss added a commit to getsentry/sentry-docs that referenced this pull request Oct 30, 2026
## DESCRIBE YOUR PR

Adds a `report` job to the `AI Evals` workflow that posts each label's
products matrix (test x target x product, base and head) to Slack once
the eval jobs finish, using the `post-matrix` action from
[sentry-for-ai-evals#111](getsentry/sentry-for-ai-evals#111).
This repo is public, so the matrix can't go on the PR.
[VDY-309](https://linear.app/getsentry/issue/VDY-309/onboarding-evals-post-a-pr-runs-products-matrix-to-slack)

- Same job and condition as
[getsentry/sentry-for-ai#385](getsentry/sentry-for-ai#385)
and
[sentry-for-ai-evals#115](getsentry/sentry-for-ai-evals#115).
It runs only when the eval jobs succeeded or failed, so not when a new
push cancelled them, when there's no preview, or for fork PRs.
- The log shows only whether each label was posted. The channel is set
in sentry-for-ai-evals, so this repo doesn't name it.
- Posts with the `EVALS_SLACK_BOT_TOKEN` secret, which is set on this
repo. Without it, the job logs a notice and posts nothing.
- Also sends each eval job's results into its own run
(`matrix.job.run-id`, falling back to the label and side's run), so
`:x<count>` labels keep working once
[sentry-for-ai-evals#137](getsentry/sentry-for-ai-evals#137)
makes a run per repeat. Same change as
[getsentry/sentry-for-ai#387](getsentry/sentry-for-ai#387).
[VDY-327](https://linear.app/getsentry/issue/VDY-327/onboarding-evals-upload-a-run-per-repeat-with-a-01-score-per-product)

Testing: `ai-evals:fastapi:sonnet-5.5` on this PR posted its matrix to
Slack
([run](https://github.com/getsentry/sentry-docs/actions/runs/40583152243)).

## IS YOUR CHANGE URGENT?

Help us prioritize incoming PRs by letting us know when the change needs
to go live.
Select exactly one option. For deadlines, replace `YYYY-MM-DD` with the
due date. You can update this information later by editing the PR
description.

- [ ] Urgent deadline (GA date, etc.): YYYY-MM-DD
- [x] Other deadline: 2026-10-30
- [ ] No deadline: Not urgent, can wait up to 1 week+

## SLA

- Teamwork makes the dream work, so please add a reviewer to your PRs.
- Please give the docs team up to 1 week to review your PR unless you've
supplied a deadline.

Thanks in advance for your help!

## PRE-MERGE CHECKLIST

_Make sure you've checked the following before merging your changes:_

- [ ] Checked Vercel preview for correctness, including links
- [ ] PR was reviewed and approved by any necessary SMEs (subject matter
experts)
- [ ] PR was reviewed and approved by a member of the [Sentry docs
team](https://github.com/orgs/getsentry/teams/docs)

🤖 Generated with [Claude Code](https://claude.com/claude-code)
@jaydgoss
jaydgoss merged commit 8064245 into main Oct 30, 2026
20 checks passed
@jaydgoss
jaydgoss deleted the jaygoss/ai-evals-run-id-per-repeat branch October 30, 2026 16:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ai-evals:llms_txt:sonnet-5.5 Run the llms.txt eval on Sonnet 5.5 only, base and head

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant