Experiment blocks: stopping + external archive

This project supports method-level early stopping and external archive tracking. These are not feature toggles: they define explicit contracts (artifacts + metadata) and are evaluated experimentally.

stopping.hv_convergence

Enable HV-based convergence stopping driven by a hypervolume trace sampled during the run.

Example:

stopping:
  hv_convergence:
    enabled: true
    every_k: 200
    window: 10
    patience: 5
    epsilon: 1e-4
    epsilon_mode: rel     # abs|rel
    statistic: median     # mean|median|min
    min_points: 25
    confidence: null      # e.g. 0.95 to enable bootstrap CI
    bootstrap_samples: 300
    ref_point: [2.0, 2.0] # must match n_obj (or use "auto")

Canonical record:

• manifest.json outcome metrics under hooks.stopping, including the bounded trace

Notes: - For 2 objectives, HV is computed exactly. - For >2 objectives, HV may be unavailable unless a backend provides it; trace rows log reason codes. - Use ref_point: "auto" to let the runner derive a reference point from current data.

archive.external

Enable external archive maintenance with explicit pruning policies.

Example:

archive:
  external:
    enabled: true
    capacity: 200
    truncate_size: 200
    pruning: crowding          # crowding|hv|mc_hv|knn|maxmin|ref_dirs
    hv_ref_point: null         # optional; required for hv-based policies
    rng_seed: 0
    objective_tolerance: 1.0e-10
    deduplicate_in: objective  # objective|decision|both
    decision_tolerance: 1.0e-32

Canonical record:

• manifest.json outcome metrics under hooks.archive, including the bounded trace

Notes: - In the tuning spaces, external archives use the population size as their default capacity. - When an algorithm is configured with an external archive, top-level results come from that archive by default unless result_mode="population" is requested. - pruning: hv uses exact HV contributions in 2D and, when moocore is installed, exact higher-dimensional contributions as well. mc_hv always uses the Monte Carlo proxy.

Reproducibility

Runs should be launched with fixed seeds and fixed budgets. Early stopping changes executed evaluations, but the run still reports the original max budget in the resolved spec. Use the stopping payload and trace in manifest outcome metrics for analysis.