neat/pruning/facade

Public pruning facade helpers for the stable Neat entrypoint.

This chapter is the public pruning-policy bridge for the stable Neat entrypoint. The actual pruning orchestration still lives in pruning/pruning.ts, but the top-level class preserves two lazy async wrappers so optional pruning behavior does not crowd the main facade or force every caller to load pruning mechanics up front.

That separation matters for both teaching and behavior. The root pruning chapter answers controller-facing questions about schedule-driven versus adaptive pruning, while this facade answers the public-API question of how a caller safely reaches those controllers from Neat itself.

Read this as the stable public contract, not as a second pruning engine. The facade keeps two entrypoints on Neat because callers often need to say "run the scheduled pruning path" or "run the adaptive pruning path" without also importing the broader pruning chapter or its policy helpers directly.

Read this file when you want the stable wrapper story:

• applyEvolutionPruning() forwards scheduled generation-time pruning,
• applyAdaptivePruning() forwards metric-driven pruning,
• both helpers preserve best-effort lazy loading instead of turning pruning into a mandatory runtime dependency.

Read the symbols in this order:

• NeatPruningFacadeHost defines the narrow host seam preserved by the public Neat methods,
• applyEvolutionPruning() covers predictable schedule-driven pruning,
• applyAdaptivePruning() covers feedback-driven pruning based on live population metrics.

Invariant: this facade decides how callers reach pruning from Neat, but it does not resolve pruning schedules, adaptive thresholds, or pruning math on its own.

neat/pruning/facade/pruning.facade.ts

applyAdaptivePruning

applyAdaptivePruning(
  host: NeatLikeForPruning,
): Promise<void>

Run adaptive pruning through the stable public Neat facade.

This is the public wrapper for metric-driven pruning. Use it when pruning should react to live population complexity signals rather than a fixed generation schedule.

The facade keeps the same lazy optional-loading behavior so callers can continue to treat adaptive pruning as an additive maintenance feature rather than a required runtime dependency.

This wrapper is the reactive pruning entrypoint: it says "inspect the live population metrics and adjust the shared prune level only if the drift is big enough to matter." That makes it the better fit for maintenance-style runs where pruning should respond to observed complexity rather than the calendar.

Parameters:

• host - - Neat instance exposing adaptive pruning state and population metrics.

Returns: Promise that resolves after the best-effort adaptive pruning attempt finishes.

Example:

await neat.applyAdaptivePruning();

applyEvolutionPruning

applyEvolutionPruning(
  host: NeatLikeForPruning,
): Promise<void>

Apply evolution-time pruning through the stable public Neat facade.

This is the public wrapper for schedule-driven pruning. Use it when the caller wants the root pruning chapter to inspect generation timing and apply pruning only when the configured evolution schedule says it is active.

The underlying pruning module is loaded lazily so pruning remains optional, and the facade preserves the same best-effort behavior as before the pruning logic was extracted out of the main Neat class.

This wrapper is the predictable pruning entrypoint: it says "consult the generation schedule and prune if the current run state says it is time." That makes it the better fit for orchestrated evolve loops and deterministic experiments where pruning should follow a known calendar.

Parameters:

• host - - Neat instance exposing pruning options, generation state, and population.

Returns: Promise that resolves after the best-effort pruning attempt finishes.

Example:

await neat.applyEvolutionPruning();

NeatPruningFacadeHost

Narrow Neat host surface required by the public pruning facade.

This stays intentionally small because the facade only forwards scheduled and adaptive pruning calls into the extracted pruning implementation.

The facade does not own the pruning math itself. It only needs the host state that the root pruning chapter already consumes, which keeps this layer from becoming a second public pruning implementation.

Read this seam as the minimum public bridge: enough controller state to reach pruning safely, but not enough to teach the internal pruning policies here.