GSP connected workflow: first run
Run an exact archived planning workflow, inspect independently checked results and export a reproducible evidence bundle.
Start with the known-answer scenario
Python 3.13 and the pinned requirements are the tested environment. The workflow executes the unchanged archived DieWeaver and ModelNative 1.2.0 engines through a bounded adapter. It is not a full native GSP or physical design run.
python -m pip install -r requirements.txt
python tools/gsp_cli.py sample --output scenario.json
python tools/gsp_cli.py run --input scenario.json --output evidence.json --bundle evidence.zip
python tools/gsp_cli.py verify evidence.jsonExpected answer
The supplied scenario selects resources b and c, spans two failure domains, uses 18 scenario power units and returns DieWeaver objective 394. ModelNative plans only over those selected resources. All independent checks must pass. Model estimates are not measured chip performance.
A blocked result is an answer
Raise requirements.min_failure_domains to 3 without adding a third domain. The adapter must return blocked and must not claim a successful model plan. Set prefill memory_mb to 20000 to exercise insufficient selected memory.
Preserve provenance
The bundle contains the input, placement, model plan, independent checks, lineage and execution envelope. Its hashes establish consistency and identity, not a third-party certificate. Save the input revision in the engineering workspace before comparing runs.