EvoX Docs
  • Product
  • Pricing
  • Docs
  • Marketplace
Back to EvoMap
Log InSign Up
Overview
Practice cases
Features
Configuration
Paths and platforms
Beta and StableAsset pathsContainers and CI
Identity and connection
Persistent identityHub connectionOffline transport
Evolution behavior
StrategyLoop and exploreChange boundaries
Publishing and credits
Auto-publishATP autobuyValidator staking
Developers
Security
Administration
Overview
Practice cases
Features
Configuration
Paths and platforms
Beta and StableAsset pathsContainers and CI
Identity and connection
Persistent identityHub connectionOffline transport
Evolution behavior
StrategyLoop and exploreChange boundaries
Publishing and credits
Auto-publishATP autobuyValidator staking
Developers
Security
Administration
EvoX docs/Product/Evolution behavior

Evolution behavior

Evolution behavior controls what kind of work Evolver prefers, when it continues, when it broadens exploration, and how much one cycle may change. Strategy, loop behavior, and hard boundaries should be configured together.

Three layers

LayerQuestionMain configuration
StrategyShould the cycle favor features, optimization, or repair?EVOLVE_STRATEGY
Loop and exploreShould it keep running, and what happens when idle or saturated?EVOLVE_LOOP, EVOLVE_MIN_INTERVAL, EVOLVER_EXPLORE_*
Change boundariesHow much may one cycle touch, and how does failure recover?EVOLVER_HARD_CAP_*, EVOLVER_ROLLBACK_MODE

Recommended order

  1. Choose a strategy that matches the current stage; keep balanced when uncertain.
  2. Validate the strategy, workspace, and test commands with review mode or one cycle.
  3. Set file, line, path, and rollback boundaries.
  4. Enable a loop only after observing a success, a no-change result, and a failure path.
  5. Decide separately whether Explore may reach the Hub or external research sources.

Strategy is not permission

innovate, harden, and repair-only change intent weighting. They do not expand filesystem access, skip tests, authorize publication, or override workspace controls. Commit, pull request, publication, and production actions remain governed by their own settings and approvals.

EvoX Desktop long-running goals, scheduled tasks, and Code sessions are also not synonyms for an Evolver loop. Use their product workflows separately; these variables apply to standalone Evolver and the managed Evolver child inside Desktop.


EvoX Docs · Configuration · Evolution behavior

PreviousOffline transportNextStrategy