Run a tiny shell command through run_capped to obtain a real CappedOutput with a process-supplied ExitStatus for decision testing.

Return all trials as CompletedTrial with ranking and scoring.

Get trials on a specific Pareto front.

Get all feasible trials with NSGA-II ranking information.

Convert a Rust CompletedTrial to a Python CompletedTrial.

Sample from this Gaussian component. Uses the reparameterization: x = μ + L * z, where z ~ N(0, I).

Save a checkpoint to disk.

Save checkpoint as JSON to a file. Uses atomic write (write-to-temp + fsync + rename) to prevent data loss.

Select top-k feasible trials using NSGA-II criteria. Selection is based on: 1. **Rank** (lower is better) - trials from earlier fronts are preferred

Apply an optional retention cap to the underlying leaderboard. `None` leaves the leaderboard unbounded (default). `Some(n)` caps it at `n` trials usi

Set or clear the optional cap on the number of stored trials. # Eviction policy When a cap is set and a `push` would exceed it, the leaderboard evic

Set the monotonic completed counter, clamping up to the number of currently stored trials. This is the mechanism for carrying the completed count acr

Return all trials sorted by observation, including infeasible.

Return the k trials with the lowest observations, including infeasible.

Scalarize multi-objective observations and return top-k feasible trials. The scalarizer function converts `BTreeMap<String, f64>` to a single `f64`.

Get the number of completed trials.

Build a per-write-unique temp path in the same directory as `path`. A shared, deterministic temp name (e.g. `path.with_extension("tmp")`) lets two co

Evaluate a WFG problem via pymoo, rescaling our [0, 1] inputs to pymoo's ``[0, 2i]`` decision-variable domain.

Generate the true Pareto front for a WFG problem via pymoo.

Apply matplotlib rcParams for publication-quality figures.

Return the trial with the best (lowest) observation. Returns `None` if no feasible trials exist.

Compute per-objective TLP scores φ_i from raw metrics. Each score is P_i × normalized_distance, matching the paper's formula.

Create a diagonal Gaussian component. Non-positive or non-finite variances are clamped to a small positive value so the covariance is always valid an

Get color for an optimizer, with fallback.

Load a checkpoint, preferring full checkpoints and falling back to legacy leaderboard-only files. This is used by CLI config `checkpoint.load_from`,

Compute TLP score for a single value (shared by scalarize_raw and vectorize_raw).

Append an existing trial record, preserving its ID and timestamp. This supports rebuilding a leaderboard with a different observation type while pres

Append a trial with raw metrics preserved for lazy re-scalarization.

Append a trial with an externally assigned trial ID and raw metrics preserved for lazy re-scalarization.

Append a trial using an externally assigned trial ID. This is useful when trial IDs are issued before completion. The internal auto-assignment counte

Run a child command, capturing stdout/stderr with a per-stream byte cap.

Sum of group costs → single scalar for strategy updates. Each group's cost already includes per-objective priority weights, so this is a plain sum: F

Start the server on the given port. Blocks until the server is shut down. If `dashboard_dir` is provided, the dashboard UI is served at `/`.

Set the refit configuration for this strategy instance.

Return feasible trials sorted by observation (ascending).

Update objectives mid-run, re-scalarizing all trials.

Compute a single trial's 0-indexed NSGA-II global rank from a cheap snapshot of `(trial_id, observation)` pairs, without cloning trials or building th

Return a likely-available TCP port. There is an inherent TOCTOU race: the port may be claimed by another process between this call and the su

Return True if dashboard/tests/node_modules/jsdom is present. Installs via `npm install` in dashboard/tests if missing. Returns False if npm/

Deterministic seed from run parameters.

Append a multi-objective result row.

Append a single-objective result row.

Return the k trials with the highest (worst) observations. Excludes infeasible trials. Use `bottom_k_all()` to include them.

Build a CompletedTrial from a scalar leaderboard trial.

Build a CompletedTrial from a vector leaderboard trial.

Build a fresh `trial_id -> index` map from a slice of trials. On duplicate trial IDs the last occurrence wins, matching how `get` would observe the m

Cancel a pending trial.

Enforce the bearer token when one is configured. Shared by the mutation and read authorization checks.

Compare Sobol vs GMM on a function.

Get a single completed trial by ID with scoring and ranking.

Single-objective config with a deterministically seeded auto strategy, used by the concurrency stress tests so refit/exploration paths are exercised t

Create the Axum router with the dashboard and explicit server options. Returns an error if any configured CORS origin is not a valid HTTP header valu

Validate the format version from raw bytes, then fully deserialize them. Used by the capped file/reader load paths so the cheap version gate runs bef