fix: don't prefix rlm harness metrics with rlm_

[mikasenghaas]

Summary

• The rlm harness's rlm @metric scraped rlm's session meta.json and recorded each numeric metric under an rlm_ prefix (rlm_turns, rlm_prompt_tokens, rlm_completion_tokens, rlm_compactions_count, ...).
• Drop the prefix so the metrics are recorded under their own names (turns, prompt_tokens, completion_tokens, compactions_count, ...). Non-numeric fields (e.g. stop_reason) are still skipped.

Breaking

• Metric keys emitted by the v1 RLMHarness lose their rlm_ prefix. Anything consuming these (dashboards, filters, downstream analysis) must migrate rlm_<name> → <name> (e.g. rlm_turns → turns).

---

Note

Medium Risk
Breaking change to metric key names for v1 RLM harness rollouts; consumers must update from rlm_<name> to <name>.

Overview
Breaking: The v1 RLMHarness rlm @metric no longer prefixes numeric fields from session meta.json with rlm_. Keys such as turns, prompt_tokens, and compactions_count are recorded under their original names instead of rlm_turns, rlm_prompt_tokens, etc.

Non-numeric metrics (e.g. stop_reason) are still omitted. Downstream dashboards, filters, or analysis that keyed on rlm_* names need to migrate to the unprefixed keys.

^{Reviewed by Cursor Bugbot for commit b7a7435. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Remove rlm_ prefix from RLM harness metric keys

The RLMHarness.rlm method in harness.py was prefixing each metric key with rlm_ before returning. It now returns the original key names from meta['metrics'] unchanged.

📊 Macroscope summarized b7a7435. 1 file reviewed, 0 issues evaluated, 0 issues filtered, 0 comments posted

🗂️ Filtered Issues

No issues evaluated.

[macroscopeapp]

Approvability

Verdict: Needs human review

This change modifies the naming convention for RLM harness metrics, removing the 'rlm_' prefix. While small, this is a runtime behavior change that could affect downstream metric consumers (dashboards, monitoring, alerts) expecting the prefixed names.

^{You can customize Macroscope's approvability policy. Learn more.}