sloray/mcm-26-b
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

114514

Browse Source
This commit is contained in:
Horldsence
2026-09-11 21:00:39 +08:00
parent 2a3ead5d3d
commit 1725cd0891
168 changed files with 170687 additions and 874 deletions
Download Patch File Download Diff File Expand all files Collapse all files
output/hrl/p1-multiround-q3/results.json
+263
View File
@@ -0,0 +1,263 @@
[
{
"scene_id": "scene-91000",
"variant": "hrl",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 3560.92,
"avg_locate_clear_time_s": 1186.973,
"accounted_virtual_time_s": 3560.92,
"program_run_time_s": 3.357654,
"real_elapsed_total_s": 0.072,
"total_requests": 135,
"measures": 132,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
2,
4,
14
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91000",
"variant": "hrl_heuristic",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 2613.443,
"avg_locate_clear_time_s": 871.148,
"accounted_virtual_time_s": 2613.443,
"program_run_time_s": 0.864249,
"real_elapsed_total_s": 0.06,
"total_requests": 137,
"measures": 134,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
2,
4,
14
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91000",
"variant": "bayes",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 2613.443,
"avg_locate_clear_time_s": 871.148,
"accounted_virtual_time_s": 2613.443,
"program_run_time_s": 0.734953,
"real_elapsed_total_s": 0.057,
"total_requests": 137,
"measures": 134,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
2,
4,
14
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91001",
"variant": "hrl",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 3817.466,
"avg_locate_clear_time_s": 1272.489,
"accounted_virtual_time_s": 3817.466,
"program_run_time_s": 1.084374,
"real_elapsed_total_s": 0.07,
"total_requests": 135,
"measures": 132,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
2,
3,
7
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91001",
"variant": "hrl_heuristic",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 2432.42,
"avg_locate_clear_time_s": 810.807,
"accounted_virtual_time_s": 2432.42,
"program_run_time_s": 0.84206,
"real_elapsed_total_s": 0.06,
"total_requests": 139,
"measures": 136,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
2,
3,
7
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91001",
"variant": "bayes",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 2432.42,
"avg_locate_clear_time_s": 810.807,
"accounted_virtual_time_s": 2432.42,
"program_run_time_s": 0.725561,
"real_elapsed_total_s": 0.063,
"total_requests": 139,
"measures": 136,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
2,
3,
7
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91002",
"variant": "hrl",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 3815.755,
"avg_locate_clear_time_s": 1271.918,
"accounted_virtual_time_s": 3815.755,
"program_run_time_s": 1.117844,
"real_elapsed_total_s": 0.072,
"total_requests": 135,
"measures": 132,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
3,
8,
11
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91002",
"variant": "hrl_heuristic",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 2509.25,
"avg_locate_clear_time_s": 836.417,
"accounted_virtual_time_s": 2509.25,
"program_run_time_s": 0.943166,
"real_elapsed_total_s": 0.072,
"total_requests": 141,
"measures": 138,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
3,
8,
11
],
"declared_absent_truth": [],
"anomaly_count": 0
},
{
"scene_id": "scene-91002",
"variant": "bayes",
"problem": 3,
"truth_count": 3,
"cleared_count": 3,
"virtual_time_s": 2509.25,
"avg_locate_clear_time_s": 836.417,
"accounted_virtual_time_s": 2509.25,
"program_run_time_s": 0.782048,
"real_elapsed_total_s": 0.069,
"total_requests": 141,
"measures": 138,
"clears": 3,
"clear_success": 3,
"completed": true,
"stop_reason": "all_resolved",
"exit_reason": "user_exit",
"valid": true,
"cleared_matches_truth": true,
"no_source_declared_absent": true,
"truth_channels": [
3,
8,
11
],
"declared_absent_truth": [],
"anomaly_count": 0
}
]