| name | autoacad-run |
| description | Use when implementing and running paper experiments that need pilot timing, numerical safety, iterative repair, and reproducible outputs. |
AutoAcad Run
Overview
Use this stage to execute experiments under strict anti-fabrication rules.
Hard Rules
- Run a pilot first and print a concrete time estimate before scaling.
- Use real algorithms, real objectives, and real metrics. Never fake curves or report synthetic-looking numbers as findings.
- Implement real convergence checks. Do not rely on fixed loops alone.
- If NaN, Inf, or runtime warnings appear, fix the root cause. Do not patch over it with silent coercion.
- Save partial results before a time guard triggers.
Workflow
- Generate or update runnable code.
- Run a minimal pilot.
- Inspect logs, warnings, and intermediate metrics.
- Repair failures at the source.
- Re-run until the pilot is stable.
- Scale only within the compute budget.
- Save structured outputs in
results/.
Required Artifacts
- Executable code under
code/.
- Structured run outputs in
results/.
- A run log or summary sufficient to reproduce the final numbers.