neat/multiobjective/archive

Compact Pareto archive helpers for the NEAT controller.

This chapter keeps the write-side archive mechanics together: when multi-objective mode is enabled, it snapshots the leading Pareto fronts into a small archive that downstream telemetry and visualization helpers can inspect without retaining whole genomes.

The archive boundary is intentionally smaller than the main ranking pass. objectives/, dominance/, fronts/, and crowding/ decide which genomes belong on the frontier and how spread out they are. This file runs only after that work is complete. Its job is to preserve a lightweight historical trace of the leading fronts so later reads can answer questions such as "which genomes were on the frontier a few generations ago?" without copying the full population into long-lived telemetry state.

The write path stays compact on purpose:

• keep only the top fronts that are most useful for inspection,
• store stable genome identifiers instead of genome objects,
• label each snapshot with the current generation when available,
• trim the archive so long runs do not grow without bound.

flowchart TD
  A[Ordered Pareto fronts] --> B{Archive enabled?}
  B -->|No| C[Skip archive write]
  B -->|Yes| D[Take top fronts only]
  D --> E[Convert genomes to compact _id lists]
  E --> F[Push generation-labelled snapshot]
  F --> G{Archive too long?}
  G -->|No| H[Keep snapshot]
  G -->|Yes| I[Drop oldest snapshot]

neat/multiobjective/archive/multiobjective.archive.ts

archiveParetoFrontsIfEnabled

archiveParetoFrontsIfEnabled(
  neatInstance: NeatLikeWithMultiObjective,
  fronts: default[][],
): void

Archives a compact snapshot of the current Pareto fronts when multi-objective mode is enabled.

This helper converts the frontier result of the current ranking pass into a compact historical record. It does not recompute ranks, crowding, or dominance; it trusts the incoming ordered fronts array and only decides whether and how to persist a smaller snapshot.

The saved snapshot is intentionally lightweight:

• stores only genome _id values, not full genomes,
• keeps only the top MAX_PARETO_ARCHIVE_FRONTS fronts,
• labels the snapshot with generation when the caller tracks one,
• maintains a bounded archive by dropping the oldest entry when necessary.

Read this as an inspection aid rather than a replay format. The archive is useful for telemetry, charts, and quick frontier-history questions, but it is not designed to reconstruct full genome state later.

Parameters:

• neatInstance - - Neat instance.
• fronts - - Ordered Pareto fronts from the current ranking pass.

MAX_PARETO_ARCHIVE_FRONTS

Maximum number of top Pareto fronts to retain per archive snapshot.

This keeps each saved snapshot focused on the most competitively relevant layers instead of persisting every dominated tail front from the population.

MAX_PARETO_ARCHIVE_LENGTH

Maximum number of archive snapshots to retain.

The archive is meant to be a rolling inspection aid, not an unbounded event log. Older snapshots are discarded once this limit is exceeded.