Skip to content

Fix BEAM benchmark correctness and reproducibility - #58

Draft
ved015 wants to merge 1 commit into
BEAMfrom
codex/beam-integration-hardening
Draft

Fix BEAM benchmark correctness and reproducibility#58
ved015 wants to merge 1 commit into
BEAMfrom
codex/beam-integration-hardening

Conversation

@ved015

@ved015 ved015 commented Aug 6, 2026

Copy link
Copy Markdown
Member

Stack

This PR is stacked on #44 (BEAM). Review and merge #44 first; this PR contains the correctness and reproducibility hardening developed during its architecture review.

What changed

  • Load BEAM from a pinned, prepared public-dataset snapshot and fail closed on revision, schema, count, question-type, rubric, and haystack mismatches.
  • Make benchmark behavior explicit through versioned retrieval, answer, evaluation, and aggregation protocols instead of BEAM-specific orchestrator branches.
  • Implement the BEAM paper scoring path, including structured nugget judgments and Kendall tau-b event-ordering scoring; keep the Mem0-style nugget-only profile explicit and non-official.
  • Normalize provider search results into one text/score/session/timestamp contract and make providers honor benchmark-owned Top-K and answer cutoffs.
  • Store ingestion/indexing state once per conversation build, ingest each conversation's dated sessions in deterministic order, and support durable retry/resume without duplicating build state across questions.
  • Separate one-time build metrics from per-question query metrics and persist protocol/model/dataset identity through reports, APIs, leaderboards, comparisons, and the UI.
  • Add regression coverage and CI for dataset integrity, shared-build ownership, resume behavior, Top-K, provider normalization, scoring, reporting, and malformed-input failures.

Scope remains explicitly BEAM 1M/10M.

Validation

  • bun test — 222 tests passed, 851 assertions
  • bunx tsc --noEmit
  • bunx tsc --noEmit -p ui/tsconfig.json
  • bun run format:check
  • git diff --check

No benchmark run data or environment files are included.

@socket-security

Copy link
Copy Markdown

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff Package Supply Chain
Security
Vulnerability Quality Maintenance License
Addedhyparquet-compressors@​1.1.1981009980100
Addedjson5@​2.2.310010010082100
Addedhyparquet@​1.27.19910010097100

View full report

@socket-security

Copy link
Copy Markdown

Warning

Review the following alerts detected in dependencies.

According to your organization's Security Policy, it is recommended to resolve "Warn" alerts. Learn more about Socket for GitHub.

Action Severity Alert  (click "▶" to expand/collapse)
Warn High
Obfuscated code: npm hysnappy is 90.0% likely obfuscated

Confidence: 0.90

Location: Package overview

From: ?npm/hyparquet-compressors@1.1.1npm/hysnappy@1.0.0

ℹ Read more on: This package | This alert | What is obfuscated code?

Next steps: Take a moment to review the security alert above. Review the linked package source code to understand the potential risk. Ensure the package is not malicious before proceeding. If you're unsure how to proceed, reach out to your security team or ask the Socket team for help at support@socket.dev.

Suggestion: Packages should not obfuscate their code. Consider not using packages with obfuscated code.

Mark the package as acceptable risk. To ignore this alert only in this pull request, reply with the comment @SocketSecurity ignore npm/hysnappy@1.0.0. You can also ignore all packages with @SocketSecurity ignore-all. To ignore an alert for all future pull requests, use Socket's Dashboard to change the triage state of this alert.

View full report

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.

1 participant