fix(run_orchestrator): preserve multiple_results in MAD_CONTAINER_IMAGE synthetic manifest

[coketaste]

Summary

• _create_manifest_from_local_image() builds a synthetic manifest for MAD_CONTAINER_IMAGE (local-image) runs, but the built_models entry it constructs omitted the multiple_results field
  from models.json.
• Without it, ContainerRunner never sets MAD_OUTPUT_CSV, never copies the perf CSV out of the container, and falls back to regex-scraping the run log for a performance: NUMBER METRIC
  line — reporting FAILURE (no performance metrics) even when the model produced a fully valid perf CSV.
• Discovered while debugging a MAD Uber-storefront run that completed a full correctness + perf sweep successfully but was marked FAILURE in perf.csv/perf_super.json solely because it
  ran via local-image mode.

Fix

• run_orchestrator.py: add "multiple_results": model.get("multiple_results", "") to the synthetic built_models entry, matching how every other models.json field is already carried
  through.

Test plan

• Added TestCreateManifestFromLocalImage in tests/unit/test_orchestration.py:
  • test_multiple_results_field_is_preserved — a model with multiple_results set retains it in the generated manifest.
  • test_multiple_results_defaults_to_empty_string — a model without it still gets the key defaulted to "", matching existing field-default conventions.
• pytest tests/unit tests/integration/test_orchestrator_workflows.py — 562 passed.

[Copilot]

Pull request overview

This PR fixes local-image (MAD_CONTAINER_IMAGE) execution by ensuring the synthetic build_manifest.json produced by RunOrchestrator._create_manifest_from_local_image() preserves the multiple_results field from models.json, which ContainerRunner relies on to set MAD_OUTPUT_CSV and correctly export perf CSV results.

Changes:

• Preserve multiple_results when constructing synthetic built_models entries for local-image runs.
• Add unit tests verifying multiple_results is preserved when provided and defaults to "" when absent.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/madengine/orchestration/run_orchestrator.py	Adds multiple_results to the synthetic manifest’s built_models entries so local-image runs follow the same manifest schema as built-image runs.
tests/unit/test_orchestration.py	Adds unit coverage to prevent regressions for multiple_results propagation and defaulting behavior.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.