Test that exploration mode uses uniform random sampling

Test extracting diffs from a response

Test extraction from checkpoint with no programs

Test full lineage extraction

Test full lineage extraction with JSON output

Test extraction with multiple generations

Test extraction with single parent-child pair

Test that _extract_unique_features correctly handles program_type determination

Test extraction with code included

Test extraction handles corrupted files gracefully

Test extraction handles missing parents gracefully

Test extraction without code

Test the _fast_code_diversity function

Test that fast evaluations complete successfully

Test that same feature coordinates in different islands don't conflict

Test MAP-Elites feature coordinate calculation

Test that feature coordinates are formatted correctly in the prompt

Test feature coordinate calculation with new default dimensions

Test that None and empty list for feature_dimensions are handled correctly

Test per-island feature map operations for MAP-Elites

Test that feature ranges are preserved across checkpoint save/load cycles

Test min-max feature scaling

Test feature scaling when all values are identical

Test that feature_stats accumulate correctly across checkpoint cycles

Test that feature_stats remain consistent across multiple save/load cycles

Test that feature_stats ranges are preserved when loading from checkpoint

Test that feature_stats are correctly saved and loaded from checkpoints

Test feature_stats serialization handles edge cases correctly

Test feature statistics are updated correctly

Test that numpy types are correctly handled in serialization

Test that different file extensions are handled correctly

Test that final checkpoint is created regardless of interval

Test that the first program added to an island becomes the best

Test that fitness calculation is consistent across all methods

Test _format_inspirations_section directly with feature_dimensions

Test _format_inspirations_section works without feature_dimensions

Test that initial program is added when starting fresh

Test that fresh start correctly handles iteration 0 as special

Test that dict -> EvaluationResult -> dict roundtrip works

Test complete evolution with real LLM

Test getting the best program

Test getting top programs from a specific island

Test get_top_programs with specific metrics

Test that global best program is tracked correctly across all islands

Test that grid expands correctly when new programs exceed existing ranges

Attempt to import all config files

Test that initial_changes_description defaults to empty string

Test initial state of island best program tracking

Test that islands are properly initialized

Test that initial_program.py works

Test the initial program generates parameters

Test that the initial program file exists

Test that the initial program has EVOLVE-BLOCK markers

Test handling of inspirations with missing or invalid metrics

Test handling of invalid island indices

Test that NaN is invalid

Test that negative infinity is invalid

Test that None is invalid (not a number)

Test that positive infinity is invalid

Test that island best programs are maintained across operations

Test island best tracking with combined_score metric

Test island best tracking when best program is removed

Test that initial batch is distributed across islands

Test that each island gets its own feature map

Test that island feature maps are properly populated during evolution

Test that island_id wraps around correctly

Test that parallel workers maintain island isolation

Test that island best tracking is isolated between islands

Test that migration doesn't create duplicate chains with real evolution

Test basic island operations

Test island status logging functionality

Simulate evolution and verify all islands get programs. With the fix for issue #391, children are placed in the target island

Test JSON format output

Test JSONL format output

Test library API gives proper error messages when not configured

Test that LLM configuration can be created properly

Test that temperature defaults to 0.7 in LLMConfig

Test that top_p defaults to None in LLMConfig (for Anthropic compatibility)

Test that long evaluations time out properly

Test handling of malformed feature_stats during deserialization

Test that early_stopping_metric defaults to combined_score

Test that migrated programs preserve original content and metrics

Test that migration creates programs with clean UUID IDs, not _migrant suffixes

Test that migration creates proper program copies

Test that migration selects the best programs for migration

Test that migration works correctly with pinned workers

Test that migration preserves program content and metrics

Test that programs marked as migrants don't migrate again

Test that migration rate is properly applied

Test that migration follows ring topology

Test that migration skips programs if target island already has identical code

Test that programs migrate to different islands, not same island

Test that migration timing logic works correctly

Test that migration updates the last migration generation

Test that migration can update island best programs

Test that migrants go through MAP-Elites deduplication (same cell = keep better)

Test that migration validation passes after our fixes

Test that migration handles empty islands gracefully

Test that when migrants compete for same feature cell, resolution is clean

Test that missing feature dimensions raise appropriate errors