feat(observe): garage-conveyor FIELD eval pack from real Langfuse questions#2185
Open
Mikecranesync wants to merge 1 commit into
Open
feat(observe): garage-conveyor FIELD eval pack from real Langfuse questions#2185Mikecranesync wants to merge 1 commit into
Mikecranesync wants to merge 1 commit into
Conversation
…stions Curated from real production questions exported via tools/langfuse_export.py (PR #2172) — the GARAGE CONVEYOR surface (GS10 VFD + Micro820, UNS enterprise.garage.line1.conveyor1). Unlike conveyor_demo (synthetic SimLab scenario), every question here was actually asked by a technician. - 11 items (9 active, 2 held for review): CE10 comms fault, reset behavior, overcurrent-on-startup, the 500 MΩ megger trap (good insulation, NOT a bad motor), e-stop safety ("can't confirm a safety circuit remotely"), etc. - answer_points / unacceptable_answer_patterns encode OEM-correct substance and the hallucination/safety traps these questions invite — not what MIRA happened to answer (the historical answers were ~26% ungrounded). - expected_documents/required_citations intentionally empty (the corpus didn't ground these); the pack tests asset resolution, answer substance, hallucination avoidance, and safety governance. missing_citation warns honestly. - approvals.example.json: approve enterprise.garage.line1.conveyor1 so governance gates (incl. the safety item) pass. Verified: `python -m simlab.observe.run_eval garage_conveyor_field` → 9/9 PASS, asset/points/citation 100%. Run --live to use it as an engine regression gate. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CS9fxC3gdSUJDJqHw1uMiu
Open
5 tasks
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
1 participant
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.
Why
We exported 3 months of real production questions from Langfuse (#2172). This turns the high-value ones into a real regression eval pack — the first eval grounded in actual technician questions rather than a synthetic scenario.
What
simlab/observe/evalpacks/garage_conveyor_field.yaml— 11 items (9 active, 2 held for review) for the GARAGE CONVEYOR (GS10 VFD + Micro820, UNSenterprise.garage.line1.conveyor1):answer_points/unacceptable_answer_patternsencode OEM-correct substance + the hallucination/safety traps, not what MIRA historically answered (~26% of historical answers were ungrounded).expected_documents/required_citationsintentionally empty — the historical corpus didn't ground these, so the pack tests asset resolution, answer substance, hallucination avoidance, and safety governance.missing_citationwarns honestly.approvals.example.json: approves the garage asset so governance gates (incl. the safety item) pass.Verification
python -m simlab.observe.run_eval garage_conveyor_field→ 9/9 PASS, asset/points/citation 100%. Run--liveto use it as an engine regression gate.Notes
simlab/observe); base isfeat/observability-eval-layer. Retarget tomainafter feat(observe): production-grade observability + evaluation layer (phases 0-3) #2154 merges.evalseed-*.yaml(PII-scrubbed,active:false) for future packs.🤖 Generated with Claude Code