{ "model": "openai/gpt-oss-20b", "selected_checkpoint": "proof-v5-learned-epoch-4", "primary_metric": "certificate_pass_rate", "improved_over_base": true, "rows_sha256": "3051cd3b9dfdb597a5be54873968908491d056f39ec6dd875f0bae847a20675c", "results": { "overall": { "base": { "n": 768, "certificate_pass_rate": 0.049479166666666664, "claim_accuracy": 0.5690104166666666, "mean_generated_tokens": 1375.0546875, "token_caps": 60, "failure_types": { "fact_index": 5, "not_closed": 317, "redundant_step": 32, "schema": 78, "unavailable_key": 238, "unclean_termination": 60, "valid": 38 } }, "learned": { "n": 768, "certificate_pass_rate": 0.5572916666666666, "claim_accuracy": 0.9752604166666666, "mean_generated_tokens": 492.2161458333333, "token_caps": 4, "failure_types": { "fact_index": 6, "not_closed": 36, "query_mismatch": 1, "redundant_step": 38, "unavailable_key": 255, "unclean_termination": 4, "valid": 428 } }, "paired": { "groups": 384, "adapted_minus_base": 0.5078125, "paired_group_bootstrap_95": [ 0.4609375, 0.5533854166666666 ] } }, "test_depth": { "base": { "n": 256, "certificate_pass_rate": 0.03125, "claim_accuracy": 0.5625, "mean_generated_tokens": 1391.953125, "token_caps": 18, "failure_types": { "fact_index": 3, "not_closed": 126, "redundant_step": 1, "schema": 21, "unavailable_key": 79, "unclean_termination": 18, "valid": 8 } }, "learned": { "n": 256, "certificate_pass_rate": 0.4296875, "claim_accuracy": 0.9609375, "mean_generated_tokens": 646.69140625, "token_caps": 4, "failure_types": { "fact_index": 1, "not_closed": 7, "redundant_step": 6, "unavailable_key": 128, "unclean_termination": 4, "valid": 110 } }, "paired": { "groups": 128, "adapted_minus_base": 0.3984375, "paired_group_bootstrap_95": [ 0.3203125, 0.48046875 ] } }, "test_iid": { "base": { "n": 256, "certificate_pass_rate": 0.11328125, "claim_accuracy": 0.62109375, "mean_generated_tokens": 1256.66015625, "token_caps": 18, "failure_types": { "fact_index": 1, "not_closed": 94, "redundant_step": 10, "schema": 27, "unavailable_key": 77, "unclean_termination": 18, "valid": 29 } }, "learned": { "n": 256, "certificate_pass_rate": 0.8046875, "claim_accuracy": 0.99609375, "mean_generated_tokens": 389.2890625, "token_caps": 0, "failure_types": { "not_closed": 6, "query_mismatch": 1, "redundant_step": 4, "unavailable_key": 39, "valid": 206 } }, "paired": { "groups": 128, "adapted_minus_base": 0.69140625, "paired_group_bootstrap_95": [ 0.6171875, 0.76171875 ] } }, "test_topology": { "base": { "n": 256, "certificate_pass_rate": 0.00390625, "claim_accuracy": 0.5234375, "mean_generated_tokens": 1476.55078125, "token_caps": 24, "failure_types": { "fact_index": 1, "not_closed": 97, "redundant_step": 21, "schema": 30, "unavailable_key": 82, "unclean_termination": 24, "valid": 1 } }, "learned": { "n": 256, "certificate_pass_rate": 0.4375, "claim_accuracy": 0.96875, "mean_generated_tokens": 440.66796875, "token_caps": 0, "failure_types": { "fact_index": 5, "not_closed": 23, "redundant_step": 28, "unavailable_key": 88, "valid": 112 } }, "paired": { "groups": 128, "adapted_minus_base": 0.43359375, "paired_group_bootstrap_95": [ 0.35546875, 0.51171875 ] } } } }