Strategy

EVOLVE_STRATEGY is an intent-weight preset. It does not name a specific code change; it affects how Evolver balances repair, optimize, innovate, and explore signals.

Presets

Value	Main bias	Good fit
`balanced`	Favors innovation while retaining optimization, repair, and exploration	Default day-to-day operation.
`innovate`	Strongly favors new capability while retaining small optimization, repair, and exploration shares	Stable system with a clear need for new capability.
`harden`	Favors repair and optimization while reducing innovation and exploration	After a major change, when stability and robustness matter most.
`repair-only`	Almost entirely repair-focused, with no innovation and only small optimization and exploration shares	Incident or regression recovery without feature expansion.
`auto`	Resolves strategy dynamically from signals, idle cycles, and lifecycle state	Managed environments with reliable signals and observation.

The default is balanced.

Internal weights can change between Evolver releases, and current implementations reserve a separate share for Explore. This page therefore documents stable intent rather than fixed percentages. For an exact audit, use the strategy definition and startup diagnostics from the installed package.

Set a strategy

bash

EVOLVE_STRATEGY=harden evolver --review

Use review mode first and inspect the proposed intent before running or looping. Do not use FORCE_INNOVATION or EVOLVE_FORCE_INNOVATION as a production substitute for valid signals; they force an innovate bias.

Choose a preset

Active incident or broad test failure: repair-only.
Post-migration stabilization and missing tests: harden.
Stable baseline with a clear feature target: innovate, paired with tight change caps.
General maintenance or uncertainty: balanced.
Use auto only when signal quality and monitoring are already established.

Verify the result

The startup log reports the resolved preset. Check that:

The candidate intent matches the current operational stage.
innovate does not ignore failing validation.
repair-only does not expand product scope.
Explore runs only with allowed sources and cooldowns.

If the output is wrong, stop the loop and fix the preset or signals. Do not raise hard caps to hide a strategy problem.

EvoX Docs · Configuration · Evolution behavior

Presets

Value	Main bias	Good fit
`balanced`	Favors innovation while retaining optimization, repair, and exploration	Default day-to-day operation.
`innovate`	Strongly favors new capability while retaining small optimization, repair, and exploration shares	Stable system with a clear need for new capability.
`harden`	Favors repair and optimization while reducing innovation and exploration	After a major change, when stability and robustness matter most.
`repair-only`	Almost entirely repair-focused, with no innovation and only small optimization and exploration shares	Incident or regression recovery without feature expansion.
`auto`	Resolves strategy dynamically from signals, idle cycles, and lifecycle state	Managed environments with reliable signals and observation.

The default is balanced.

Choose a preset

Active incident or broad test failure: repair-only.

Post-migration stabilization and missing tests: harden.

Stable baseline with a clear feature target: innovate, paired with tight change caps.

General maintenance or uncertainty: balanced.

Use auto only when signal quality and monitoring are already established.

Verify the result

The startup log reports the resolved preset. Check that:

The candidate intent matches the current operational stage.

innovate does not ignore failing validation.

repair-only does not expand product scope.

Explore runs only with allowed sources and cooldowns.

If the output is wrong, stop the loop and fix the preset or signals. Do not raise hard caps to hide a strategy problem.

Strategy

Presets

Set a strategy

Choose a preset

Verify the result

Related documentation

Strategy

Presets

Set a strategy

Choose a preset

Verify the result

Related documentation