ArchBench Traceability Link Recovery

Architecture Traceability Link Recovery: Recover traceability links between software architecture documentation and source code.

10rows
weighted_avg_f1primary metric
2026-05-06sampled

Metadata

Metrics

Precision, Recall, F1, Weighted Average F1

Latest Results

Rows ranked by weighted_avg_f1. ArchBench paper: https://arxiv.org/abs/2603.17833. CLI: https://github.com/sa4s-serc/archbench-cli.

Rank Subject Weighted Average F1 Model Match Provenance Sampled
1 TransArC (Manual SAM) 0.87 Imported 2026-05-06
2 GPT-4o (SAD-extracted SAM) 0.86 Imported 2026-05-06
3 GPT-4 Turbo (SAD-extracted SAM) 0.86 Imported 2026-05-06
4 GPT-4 (SAD-extracted SAM) 0.86 Imported 2026-05-06
5 GPT-4o mini (SAD-extracted SAM) 0.85 Imported 2026-05-06
6 GPT-3.5 Turbo (SAD-extracted SAM) 0.85 Imported 2026-05-06
7 Llama 3.1 70b (Code-extracted SAM) 0.81 Imported 2026-05-06
8 Codellama 13b (SAD-extracted SAM) 0.71 Imported 2026-05-06
9 Llama 3.1 8b (SAD-extracted SAM) 0.68 Imported 2026-05-06
10 ArDoCode (Direct SAD-to-Code) 0.62 Imported 2026-05-06