Public task

dc-legacy-alloc-full-stack

The whole retired engine at once: every config switch, raw ERP data, locks, and pins.

expertsoftware-engineering

Model configurations: 3
Evaluation attempts: 16
Public examples: 16

Current task results

Evaluation comparison

Published evaluation results for this task version. Each bar shows a model configuration’s average score.

Model averages for this task version
Model	Average score	Score range	Attempts
Muse Spark 1.3 Contributoropencode · opencode · 1.18.11	56.5%± 9.1 pp SE	0.0% – 100.0%	1111 scored
gemini-3.1-pro-previewgoogle-vertex · opencode · 1.18.11	39.0%± 0.0 pp SE	39.0% – 39.0%	22 scored
GPT 5.6 Solopenai · opencode · 1.18.11	39.0%± 22.5 pp SE	0.0% – 78.0%	33 scored

Averages and ranges use scored attempts on this exact version. Attempts without a score are counted separately. SE measures uncertainty across attempt scores.

The task

What the agent receives

# Rebuild the whole retired engine

The old planning module was retired last quarter and its allocation engine's source did not survive the migration. Its captured production runs are in `audit/` — request/plan pairs, exactly as the engine consumed and produced them. The decision policy was never written down; the archive is the only spec.

The live engine never saw switches one at a time. This is a full production cycle: four campuses, a raw ERP extract, supplier pins, `enforceSingleSupplier`, and a locked phase at ALB01 with committed assignments in the request.

Implement the engine at `allocator/solve.py` (entry point in
`allocator/README.md`) so it reproduces the legacy engine's planning
behavior on any request, and use it to plan `data/scenario.json`. The engine
re-runs as the world changes — new lots land and phases lock between runs —
and grading is outcome equivalence with the legacy engine, not matching its
exact lot choices.

Check plan structure with
`python3 contracts/validate.py <request.json> <response.json>`.