mirror of
https://github.com/supabase/supabase.git
synced 2026-10-05 17:35:10 +03:00
## Problem Scorers previously derived the assistant's final answer via Braintrust's `trace.getThread()`, which silently truncates long traces at the backend's preview-length cap (~10KB). The SDK never passes `preview_length` in its BTQL query and there's no supported override. This caused false-negative scores (Completeness, Correctness, Goal Completion, Safety collapsing to 0/null) specifically on multi-step tool-calling eval cases, since longer traces are more likely to have their tail (the final assistant message) truncated away. ## Solution Capture the assistant's full, untruncated final answer directly in the eval task's output in memory (via AI SDK's `result.steps`, already fully available once the stream is consumed) instead of round-tripping through Braintrust's truncating storage/query layer. Scorers now read `output.transcript` instead of calling `trace.getThread()`. ## Changes - **New**: `apps/studio/evals/transcript.ts` — `Transcript` type and `buildTranscript()` function - **New**: `apps/studio/evals/transcript.test.ts` — unit tests (5 passing) - **Modified**: `apps/studio/evals/assistant.eval.ts` — captures `result.steps` and returns transcript - **Modified**: `apps/studio/evals/scorer.ts` — migrated 7 scorers to read from local transcript - **Modified**: `apps/studio/evals/trace-utils.ts` — removed dead thread-serialization code - **Deleted**: `apps/studio/evals/trace-utils.test.ts` — superseded by transcript tests ## Test Plan - [x] `pnpm --filter studio typecheck` — clean - [x] `pnpm --filter studio lint` — clean - [x] `npx vitest run evals/transcript.test.ts` — 5/5 passing - [x] Full live eval run (35/35 cases) against Braintrust — [experiment](https://www.braintrust.dev/app/supabase.io/p/Assistant/experiments/eval-scorer-transcript-capture-1786985352) shows Completeness/Correctness/Goal Completion/Safety scores comparable to baseline ## Known Residual Risk Other scorers that derive data from `trace.getSpans()` (toolUsageScorer, sqlSyntaxScorer, sqlIdentifierQuotingScorer, knowledgeUsageScorer, and docsFaithfulnessScorer's docs-content lookup) could theoretically hit the same truncation issue, but have not been observed to fail in practice. This is not addressed in this PR. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added transcript generation from assistant interaction steps, including text and tool-call inputs. * Evaluation results can now include complete transcripts for detailed conversation analysis. * Online evaluations can derive transcripts from recorded interaction traces when needed. * **Bug Fixes** * Improved scoring by selecting the appropriate conversation content for each evaluation. * Ensured offline transcripts take precedence when available, with trace-based fallback support. * **Tests** * Added coverage for multi-step interactions, tool calls, filtering, empty steps, and URL validation. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
63 lines
2.3 KiB
TypeScript
63 lines
2.3 KiB
TypeScript
/**
|
|
* Entry point for `braintrust push` to deploy scorers to Braintrust.
|
|
*
|
|
* Excluded scorers:
|
|
* - sqlSyntaxScorer, sqlIdentifierQuotingScorer: use libpg-query (WASM),
|
|
* which esbuild cannot bundle for Braintrust's remote infra.
|
|
* - toolUsageScorer: requires expected.requiredTools, offline-eval-only.
|
|
* - correctnessScorer: requires ground truth (expected output), offline-eval-only.
|
|
*/
|
|
|
|
import braintrust from 'braintrust'
|
|
|
|
import {
|
|
completenessScorer,
|
|
concisenessScorer,
|
|
docsFaithfulnessScorer,
|
|
goalCompletionScorer,
|
|
safetyScorer,
|
|
urlValidityScorer,
|
|
type AssistantEvalScorer,
|
|
} from './scorer'
|
|
import manifest from './scorer-online-manifest.json'
|
|
|
|
const projectId = process.env.BRAINTRUST_PROJECT_ID
|
|
if (!projectId && process.env.IS_BRAINTRUST_PUSH)
|
|
throw new Error('BRAINTRUST_PROJECT_ID is not set')
|
|
|
|
// When running in CI, prefix scorers with the branch name to avoid collisions between PRs
|
|
// in the staging project. GITHUB_HEAD_REF is only set on PR events, not push events (e.g. master).
|
|
const branch = process.env.GITHUB_HEAD_REF || process.env.GITHUB_REF_NAME
|
|
const prNumber = process.env.GITHUB_PR_NUMBER ? Number(process.env.GITHUB_PR_NUMBER) : undefined
|
|
const prefix = process.env.GITHUB_HEAD_REF
|
|
? `${process.env.GITHUB_HEAD_REF.replace(/[^a-z0-9-]/gi, '-').toLowerCase()}-`
|
|
: ''
|
|
const metadata = branch ? { gitBranch: branch, ...(prNumber && { prNumber }) } : undefined
|
|
const description = prNumber && branch ? `#${prNumber} · ${branch}` : branch
|
|
|
|
const handlers = {
|
|
'goal-completion': goalCompletionScorer,
|
|
conciseness: concisenessScorer,
|
|
completeness: completenessScorer,
|
|
'docs-faithfulness': docsFaithfulnessScorer,
|
|
'url-validity': urlValidityScorer,
|
|
// safetyScorer guards on requiresSafetyCheck (an offline-eval concept).
|
|
// Online traces have no expected, so we wrap it to always run.
|
|
safety: (args) =>
|
|
safetyScorer({ ...args, expected: { ...args.expected, requiresSafetyCheck: true } }),
|
|
} satisfies Record<string, AssistantEvalScorer>
|
|
|
|
// @ts-expect-error - Project ID is only required at build-time
|
|
const project = braintrust.projects.create({ id: projectId })
|
|
|
|
for (const { slug, name } of manifest) {
|
|
project.scorers.create({
|
|
slug: `${prefix}${slug}`,
|
|
name,
|
|
description,
|
|
handler: handlers[slug as keyof typeof handlers],
|
|
ifExists: 'replace',
|
|
metadata,
|
|
})
|
|
}
|