Repository navigation
feat: prototype hosted Spark evals - #471
Draft
mattrossman wants to merge 3 commits into
Draft
mattrossman wants to merge 3 commits into
mattrossman wants to merge 3 commits into
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
mattrossman
added a commit
that referenced
this pull request
Nov 5, 2026
Adds a Spark eval suite and a minimal sample scenario for adding due dates to a seeded task tracker. Follow-up from #471 which was a larger attempt at adding MCP / CLI integration and adapting existing scenarios. ## Demo See [sample trace](https://www.braintrust.dev/app/supabase.io/p/Evals/experiments/spark-codex-gpt-6-luna%4026af6cd-20261104T0022Z) from a local run. <img width="3836" height="2160" alt="CleanShot 2026-11-03 at 7 27 20 PM@2x" src="https://github.com/user-attachments/assets/ba03b4bd-42f5-4dee-96aa-2be4e7f8df49" /> ## How to use - **To add a scenario** → Put `PROMPT.md`, `EVAL.ts`, and optional seeded `remote/project.sql` under `evals/spark/<id>/`. - **To run scenarios** → Set `SPARK_API_URL` in `.env` (find it [here](https://grid-supabase.enterprise.slack.com/archives/C0C7XC4AV34/p1793135179096979?thread_ts=1793135043.970229&cid=C0C7XC4AV34)), connect to Supabase WARP, then run `pnpm eval:upload -- --suite spark`. - **To view results** → Open the Braintrust experiment link printed by the command, or hand it to [Braintrust MCP](https://docs.braintrust.dev/integrations/developer-tools/mcp). The expected workflow is to add a scenario, verify it fails, update Spark API, and verify the eval passes. The Spark team owns the suite, skill, and runtime so they can run that process self-serve. ## Notes Spark interface is explained to agent in a `/spark` skill derived from [spark-server/API.md](https://github.com/supabase/spark-platform/blob/8c21e34203f58ad4cc234b4d8c1986b1fbcecef7/spark-server/API.md). Currently this can only run locally with WARP access. CI support will need Sandboxes to have WARP access. You can find existing scenarios from our benchmark and regression suites in [evals/](https://github.com/supabase/evals/tree/main/evals), though these will need adaptation to make sense for Spark due to known incompatibilities (e.g. many rely on RLS during setup). Closes AI-1399
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This POC runs Supabase evals against hosted Spark projects through a Management API-compatible adapter. The POC forwards SQL queries and migrations to Spark and seeds SQL fixtures before the agent starts. It also includes experimental Spark adaptations of benchmark and regression scenarios.
The task-tracker trace shows an MCP run against hosted Spark. The benchmark and regression experiments ran the existing scenarios unchanged against Spark at 158d8df. The scenario inventory summarizes the results and potential adaptations.
The adapter implements only part of the Management API. The scenario adaptations are exploratory and need review.
Ref AI-1390