Overview

Monitor generation health, pass rate, quality score, and data coverage.

Updated 2026-08-11 00:50:34 BJT · refresh 60s
Valid Trajectories
3
for SFT in LF format
Valid Tokens
0.000B
for SFT in LF format
Data Sources
1
SFT dataset jobs
Languages
2
JavaScript/TypeScript, Python · across converted SFT data
Models
1
glm-5.2-urg · across converted SFT data
Scaffolds
1
claude_code · across converted SFT data
Instances Processed
80
from 1 jobs
Pass Rate
30.00%
for all instances processed

Programming Language Distribution

Based on valid SFT/LF trajectories with task language labels.

Trajectory Quality Score Distribution

Score histogram from exported SFT quality facts.

Job

Harbor job status and progress with SFT/pass metrics.

JobScaffoldStatusProgressInstancesPassedErrorsPass RateTask DifficultyValid TrajectoriesTokens
merged_swe_tasks_copy-custom-claude-code-2.1.118-glm-5.2-urg-20260729223211
id: 6f742da5-d927-44e1-8eb8-93e3384cbdf5
claude_codedone
80 / 80 (100.0%)
80241830.00%2.59
80 scored
3valid -
total 1.66M
Programming languagepass rate, difficulty and token totals per programming language
SegmentInstancesPassedErrorsPass RateTask DifficultyValid TrajectoriesTokens
Modelpass rate, difficulty and token totals per model
SegmentInstancesPassedErrorsPass RateTask DifficultyValid TrajectoriesTokens
Scaffoldpass rate, difficulty and token totals per scaffold
SegmentInstancesPassedErrorsPass RateTask DifficultyValid TrajectoriesTokens
Domainpass rate, difficulty and token totals per domain
SegmentInstancesPassedErrorsPass RateTask DifficultyValid TrajectoriesTokens
Categorypass rate, difficulty and token totals per category
SegmentInstancesPassedErrorsPass RateTask DifficultyValid TrajectoriesTokens
Difficultypass rate, difficulty and token totals per difficulty
SegmentInstancesPassedErrorsPass RateValid TrajectoriesTokens
Sourcepass rate, difficulty and token totals per source
SegmentInstancesPassedErrorsPass RateTask DifficultyValid TrajectoriesTokens

Valid Trajectories

Valid reward=1 trajectories converted for SFT, grouped by source job.

Data Source n Scaffold Score Turns Calls/Traj Error Rate Tokens COT Turn Ratio COT Chars/Turn Task Difficulty

Trajectory Quality Score Matrix

Click column headers to sort. Only non-zero trajectory-score weights (Σw=1.00).

Trajectory Sampler

Embedded full trajectories load locally from data/traj_embedded shards; other cards fall back to /api/traj and the local path.

Sampled Trajectories
Status unavailable. /gpufs/haoli/code/SWE-Lego-Live/.claude/worktrees/dashboard-refact/blocks/tracer/artifacts/index.yaml not found

SFT Datasets

Stats from lf.stats.json with quick artifact path copy actions.

JobScaffoldRecordsToken len (min/mean/max)Turns (min/mean/max)Scores (min/mean/max)Tool-call errorsim.jsonllf.json
merged_swe_tasks_copy-custom-claude-code-2.1.118-glm-5.2-urg-20260729223211
claude_code329,320 / 35,873 / 48,862
gt_128k: 0 · total: 107,619
10 / 15 / 23
gte_100: 0
0.8311 / 0.8762 / 0.93100.00%
0 / 54 calls · traj: 0.00%
438.5 KB453.7 KB