Test that parent_metrics are also sanitized

Test that +inf in metrics is sanitized to None

Test that sanitized data can be JSON serialized

Test that sanitized None values work with strict JSON

Test saving extracted traces in JSON format

Test saving extracted traces in JSONL format

Test that artifacts survive database save/load cycle

Test that sequential access works without issues using safe sampling

Test SerializableResult dataclass

Test that evaluations shorter than timeout complete successfully

Test the migration timing logic

Test that single island setup doesn't attempt migration

Single-line changes should show inline format

Test sampling from island with only one program

Test statistics tracking

Test that storing artifacts for non-existent program doesn't crash

Test storing artifacts when within the limit

Test storing large artifacts to disk

Test storing more artifacts than the limit

Test storing both small and large artifacts

Test storing small artifacts in JSON

Test that _submit_iteration samples from the specified island

Test that _submit_iteration uses the safe sample_from_island method

Test that successful evaluations can also have artifacts

Test that system_message_changes_description defaults to None

Test that the timeout configuration value is actually used

Test that timeouts do not trigger retries (correct behavior)

Test conversion to dictionary

Test creating an evolution trace entry

Test that disabled tracer does nothing

Test that temperature and top_p can be set to float values

Test that temperature and top_p can be set to None

Test setting unlimited artifacts with None

Test that unlimited artifacts config is None

Test that unsanitized infinity fails strict JSON serialization

Test that unsanitized NaN fails strict JSON serialization

Test that integers are valid

Test that valid metrics are not changed

Test that negative floats are valid

Test that positive floats are valid

Test that zero is valid

Test that weighted mode favors programs with higher fitness

Verify that WITH explicit target_island, child goes to target island. This is the FIXED behavior implemented in process_parallel.py.

Verify that without explicit target_island, child inherits parent's island. This is the OLD buggy behavior that we need to avoid in p

Test that worker can access file suffix from config

Test that changes description works when diff_based_evolution=True

Test that a worse program does not replace the island best

Test loading evolution trace config from YAML

Test loading api_key ${VAR} from actual YAML file

Test loading reasoning_effort from actual YAML file

Test setting max_snapshot_artifacts to 0

Test that patience=0 effectively disables early stopping

Handle timeout signal

Handle timeout signal

Backward compatibility - return just metrics

Save configuration to a YAML file

Performs a single training step.

Performs a single training step.

Performs a single training step.

Example transformation: - Validate input (2D, integer values 0-9). - Rotate the grid 90 degrees clockwise. - Increment every cell by

Transformation logic: 1. Identify the bounding box of the '8' values (mask) in the grid. 2. The output grid has the same dimensions as th

Transformation logic: 1. Identify the bounding box of the '8' values (mask) in the grid. 2. The output grid has the same dimensions as th

Example transformation: - Validate input (2D, integer values 0-9). - Upsample each cell to a 2x2 block (doubling both dimensions). -

Similar to attempt 1 but with an additional fallback strategy. Strategies: 1. Vertical reflection: grid[31-r, c] 2. Diagonal transpos

Similar to attempt 1 but with an additional fallback strategy. Strategies: 1. Vertical reflection: grid[31-r, c] 2. Diagonal transpos

Visualize the circle packing Args: centers: np.array of shape (n, 2) with (x, y) coordinates radii: np.array of shape (n) wi

Wait for all tasks in the pool to complete