FE-1505: Time each interviewer turn by purpose

[lunelson]

🌟 What is the purpose of this PR?

Condition 5 measured about 145 seconds per interviewer turn without saying where the time goes. This branch attributes each model call, compaction, and repair to a harness purpose, emits privacy-safe server telemetry spans, and makes baseline timing runs kill-safe and bounded so the per-purpose split can be read from a short run or from the Proof 0 panel run.

Draft: opened so the stack is visible end to end. The description is completed when the branch is tied off.

🔗 Related links

• FE-1505 (internal; owning issue)

🚫 Blocked by

• Stacks on ln/fe-1506-legible-panel-status; lands after it.

🔍 What does this change?

Per-purpose turn timing in the Flue binding and the baseline runner; OpenTelemetry spans in the server; BRUNCH_BASELINE_HARD_STOP and incremental condition-5.timings.jsonl writes; run-specific timing evidence preserved.

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

This PR:

• does not modify any publishable blocks or libraries, or modifications do not need publishing

📜 Does this require a change to the docs?

The changes in this PR:

• are internal and do not require a docs change

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
hash	Ready	Preview	Aug 27, 2026 5:56pm
petrinaut	Ready	Preview	Aug 27, 2026 5:56pm
petrinaut-docs	Ready	Preview	Aug 27, 2026 5:56pm

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
hashdotdesign-tokens	Ignored	Preview	Aug 27, 2026 5:56pm

[lunelson]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• FE-1507: Move the repertoire into core behind a guarded prompts subpath #9349 : 2 dependent PRs (#9355 , #9381 )
• FE-1431: Build the plugin authoring surface (cycle two) #9348
• FE-1505: Time each interviewer turn by purpose #9347 👈 (View in Graphite)
• FE-1506: Show captures and completion while a panel elicitation runs #9346
• FE-1504: Run the process-model elicitor from the Petrinaut panel on one owned document #9345
• FE-1431: Build the plugin authoring surface #9337
• FE-1406: Audit the harness-teaching lineage and propose the fixed-key plugin contract (ADR-0007) #9327
• FE-1497: Build the controller read path and the SDCPN plugin package #9325
• FE-1404: Run condition 3 as the walking skeleton #9322
• FE-1403: Assemble and test the CPS interview guidance #9321
• FE-1402: Define and rehearse the elicitation completion contract #9320
• FE-1407: Catalogue elicitor failures that published measures miss #9319
• FE-1480: Project and realize a CPS model as a Petrinaut SDCPN #9313
• FE-1495: Prevent Brunch TypeScript contracts from drifting #9312
• FE-1494: Keep Brunch strategy and control records coherent #9311
• FE-1437: Move brunch-agent into hashintel/hash with its history #9274
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cursor]

PR Summary

Low Risk
Changes are confined to evaluation harness output, dev-server telemetry (content suppressed), and tests; no auth, persistence, or production interview logic paths are altered.

Overview
Condition 5 baseline runs can now break down wall-clock per Flue model call into interview, sweep, and repair instead of only reporting coarse per-turn cost.

A new createTurnTimingRecorder listens to Flue observe() events (prompt nesting, brunch_sweep refused/applied, <settlement-check / <sweep-repair signals) and records TurnTimingRecord rows. The harness runner attaches timings to each turn and condition-5.raw.json, appends condition-5.timings.jsonl after every interviewer turn (so short runs still leave evidence), and prints per-purpose ms in transcript turn headers. BRUNCH_BASELINE_HARD_STOP caps interviewer turns (default 24); BRUNCH_BASELINE_OUTPUT_DIR can isolate production evidence.

The brunch-agent dev app registers OpenTelemetry via @flue/opentelemetry with content: false; the walking-skeleton integration asserts expected agent/chat/tool spans and that sensitive content attributes stay off spans.

Tests cover the purpose splitter, incremental timing persistence under hard stop, and a baseline fixture path that refuses the first sweep then repairs so repair timings are exercised.

^{Reviewed by Cursor Bugbot for commit 45ab0cd. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 35549b9. Configure here.}