Trace Reconstruction on DARPA TC E3 (D1)
83.1RecallHunterAgent
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HunterAgentMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 83.1 | 91.6 | 87.1 | 6.2 | |
| Tool-Aug. ReActMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 80.1 | 48.5 | 60.4 | 58.7 | |
| HunterAgent†Methodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 77.8 | 66.1 | 71.5 | 26.4 | |
| KAIROSMethodology Category=Recent learned provenance / streaming, Log-suppression rate=30%2026.05 | 64 | 76.2 | 69.6 | 12 | |
| GraphRAGMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 60.4 | 33.2 | 42.9 | 71.6 | |
| NODLINKMethodology Category=Recent learned provenance / streaming, Log-suppression rate=30%2026.05 | 58.4 | 81.5 | 68 | 7.2 | |
| GNN-LinkPredictMethodology Category=Statistical relational learning, Log-suppression rate=30%2026.05 | 46.8 | 70.4 | 56.2 | 33.1 | |
| Zero-Shot LLMMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 36.2 | 14 | 20.2 | 90.4 | |
| MAGICMethodology Category=Deterministic provenance graphs, Log-suppression rate=30%2026.05 | 35.4 | 98.6 | 52.1 | 0.4 | |
| SLEUTHMethodology Category=Deterministic provenance graphs, Log-suppression rate=30%2026.05 | 27.8 | 99.1 | 43.5 | 0 |