Strategy
A strategy is a weighted preset for selecting evolution intent. It is neither execution permission nor a fixed four-step procedure. The presets and their weights are listed below.
Presets and weights
| Preset | Repair | Optimize | Innovate | Explore | Use |
|---|---|---|---|---|---|
balanced | 20% | 20% | 50% | 10% | Everyday progress with maintenance. |
innovate | 5% | 10% | 80% | 5% | A stable baseline and a clear new capability. |
harden | 40% | 35% | 20% | 5% | Stability, performance, and failure paths after a large change. |
repair-only | 80% | 18% | 0% | 2% | Stop feature expansion during incidents and regressions. |
early-stabilize | 60% | 22% | 15% | 3% | Address existing issues in early cycles. |
steady-state | 55% | 25% | 5% | 15% | Maintain capabilities and find new directions after saturation. |
Weights influence selection. They guarantee neither fixed counts across ten cycles nor execution of a selected candidate.
Explicit and automatic selection
Autoexec reads EVOLVE_STRATEGY as a selection input. An explicit preset expresses the desired bias. When no preset is fixed, or auto is used, the resolver also considers execution stage, saturation, and pivot signals. auto is not a seventh fixed weight table.
Configure the runner and repository in Loop configuration, then set the strategy environment variable, such as EVOLVE_STRATEGY=harden in the service environment. The variable neither starts a process nor enables a disabled execution path.
Evaluate the selection
- Compare the problem with the candidate category; regression repair must not expand product scope.
- Inspect strategy and selection records, distinguishing an explicit preset from runtime resolution.
- Compare outcomes using the same validation criteria, including failed and no-improvement cycles.
- Correct signals or narrow the objective before retrying; higher file caps do not fix poor selection.
Strategy changes no publication, spending, path, or verification requirement. Zero innovation weight in repair-only does not replace review of the actual diff.