Skip to content

spike(guardian): replace Luna with Jev - #1883

Draft
sentry-junior[bot] wants to merge 3 commits into
mainfrom
spike/jev-classifier-evals
Draft

sentry-junior[bot] wants to merge 3 commits into
mainfrom
spike/jev-classifier-evals

Conversation

@sentry-junior

@sentry-junior sentry-junior Bot commented Sep 17, 2026

Copy link
Copy Markdown
Contributor

Replace Luna with typesafe-ai/jev in the production Guardian path and use the existing Guardian eval suite to compare its pass rate.

  • make Jev the default Guardian model
  • add an AI SDK 7 evaluation call in the production Guardian reviewer
  • keep the existing structured completion path for configured non-Jev models
  • set the Guardian eval suite default to Jev
  • remove the prior eval-only router adapter
  • add exact, temporary release-age exceptions for AI SDK 7 dependencies

Checks:

  • pnpm --filter @sentry/junior typecheck
  • pnpm --filter @sentry/junior-evals typecheck
  • pnpm --filter @sentry/junior lint
  • git diff --check

Live Guardian evals need AI Gateway credentials. This checkout has no Vercel project link. The focused test command also cannot start its test database at 127.0.0.1:54322.

via David Cramer.

--

View Junior Session [Sentry]

@vercel

vercel Bot commented Sep 17, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
junior-docs Ready Ready Preview Sep 17, 2026 10:17pm UTC

Request Review

Co-Authored-By: David Cramer <david@sentry.io>
@sentry-junior
sentry-junior Bot force-pushed the spike/jev-classifier-evals branch from eb2d1be to a9062be Compare September 17, 2026 21:59
@sentry-junior sentry-junior Bot changed the title spike(evals): evaluate Guardian and router with Jev spike(guardian): replace Luna with Jev Sep 17, 2026
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.

0 participants