fix: resolve false-positive run failures

[coketaste]

Summary

• Execution (ROCM-27774): the post-run log scanner greps the entire run
  log for error substrings, including a model's own generated stdout. A
  generative benchmark (e.g. an LLM writing ValueError: inside a code
  sample it generated) could flip an otherwise fully-passing run to
  FAILURE even though valid performance metrics were already recorded.
  Since the scan can't distinguish framework/harness diagnostics from model
  output, valid extracted performance metrics now take priority over a
  pattern match. The match is still printed (in yellow) for triage
  visibility instead of being silently dropped.

Test plan

• tests/unit/test_container_runner_helpers.py — new TestResolveRunStatus
  cases covering the performance-vs-error-pattern priority (ROCM-27774)
• Full unit suite passes (pytest tests/unit -q)

[Copilot]

Pull request overview

This PR adjusts post-run status determination so that when valid performance metrics are extracted, a log error-pattern match no longer flips the run to FAILURE (the match is still surfaced for triage). This mitigates false failures when generative model output includes error-like substrings.

Changes:

• Introduces resolve_run_status() to centralize status/reason selection with the new “performance wins” priority.
• Updates ContainerRunner status logic to use resolve_run_status() and print SUCCESS-with-error-match in yellow.
• Adds unit tests and updates documentation to reflect the revised log-scan behavior.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
tests/unit/test_container_runner_helpers.py	Adds unit tests covering the new status resolution rules.
src/madengine/execution/container_runner.py	Refactors status assignment/printing to use resolve_run_status() and implement the new priority.
src/madengine/execution/container_runner_helpers.py	Adds resolve_run_status() helper with documented precedence rules.
docs/configuration.md	Updates documentation for log error pattern scan behavior when performance metrics are present.

Comments suppressed due to low confidence (1)

src/madengine/execution/container_runner.py:1876

• has_performance calls performance_value.strip() twice and will raise if performance_value is not a string. Computing the stripped value once (and guarding type) avoids duplicate work and makes status determination more robust.

                                has_performance = bool(
                                    performance_value
                                    and performance_value.strip()
                                    and performance_value.strip() != "N/A"
                                )

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.