Run multi-agent method improvement cycle to identify bottlenecks and generate proposals
Run the multi-agent method improvement cycle to analyze solver performance, identify bottlenecks, and generate improvement proposals.
# Run 1 improvement cycle
python docs/analysis/method_improvement_cycle.py --cycles 1
# Run 3 cycles (recommended)
python docs/analysis/method_improvement_cycle.py --cycles 3
# Resume from previous state
python docs/analysis/method_improvement_cycle.py --resume
# Interactive mode (human review of proposals)
python docs/analysis/method_improvement_cycle.py --interactive --cycles 1
# Generate summary report only
python docs/analysis/method_improvement_cycle.py --report
# Fresh start (clear history)
python docs/analysis/method_improvement_cycle.py --fresh --cycles 3
# Analyze specific solvers only
python docs/analysis/method_improvement_cycle.py --solvers implicit_fdm,spectral_cosine
data/pareto_fronts/: Pareto front JSON files per solverdata/cycle_reports/: Markdown reports per cycledata/improvement_history.json: Cycle history and proposals| Category | Description | Typical Action |
|---|---|---|
| stability | Solver fails at certain params | Reduce dt, adaptive stepping |
| accuracy_gap | Large error variation | Increase resolution |
| speed_gap | Large time variation | Optimize implementation |
| coverage_gap | Single solver dominates | Find solver niches |
| Type | Description | Auto-implementable |
|---|---|---|
| parameter_tuning | Adjust dt/nr | Yes |
| algorithm_tweak | Modify algorithm | Sketch provided |
| new_solver | New implementation | Sketch provided |