synthetic-disentangled-Mar19 Results
synthetic-disentangled-Mar19 Results
Collection State
- planned jobs:
63 - completed jobs:
62 - missing job:
sorting_algorithms + beta0p1_base
Across the collection, controlled_test_final/mean_overall_final_answer_accuracy stays in a narrow 0.9853 to 1.0 range, while controlled_test_final/alignment_one_to_one ranges from 0.1624 to 0.6930.
Best Results By Task
grid_pathfinding
- best setting:
beta0p1_twr2 alignment_one_to_one = 0.6930local_strategy_coverage_mean = 1.0000local_strategy_full_coverage_rate = 1.0000mean_overall_final_answer_accuracy = 1.0000
Representative non-token-weighted comparison:
beta0p5_basealignment_one_to_one = 0.224233local_strategy_coverage_mean = 0.442500mean_overall_final_answer_accuracy = 1.0000
sorting_algorithms
- best setting:
beta0p1_twr2 alignment_one_to_one = 0.270660local_strategy_coverage_mean = 0.397800mean_overall_final_answer_accuracy = 0.993040
Representative non-token-weighted comparison:
beta0p1_pe0p5_rmk0p5_rsw0p1alignment_one_to_one = 0.175460local_strategy_coverage_mean = 0.223340mean_overall_final_answer_accuracy = 0.999720
multidigit_addition
- best setting:
beta0p1_rsw1 alignment_one_to_one = 0.332367local_strategy_coverage_mean = 0.332400mean_overall_final_answer_accuracy = 0.996600
Comparison with token-weighted reconstruction:
beta0p1_twr2alignment_one_to_one = 0.3280local_strategy_coverage_mean = 0.3281mean_overall_final_answer_accuracy = 0.991267
Figures
Caption: Token-weighted reconstruction is the only block that changes the best achievable alignment on grid_pathfinding and sorting_algorithms.
Caption: Alignment varies sharply with token-weighted reconstruction weight compared with the flatter occupancy-only trends seen elsewhere in the aggregate bundle.
Caption: The standout grid_pathfinding token-weighted run reaches full local strategy coverage in a way the non-token-weighted baselines do not.