Trace Reconstruction on DARPA OpTC (D2)
81.5RecallHunterAgent
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HunterAgentMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 81.5 | 90.4 | 85.7 | 7 | |
| Tool-Aug. ReActMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 77.6 | 46 | 57.7 | 62.3 | |
| HunterAgent†Methodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 75.4 | 64.3 | 69.4 | 29.1 | |
| KAIROSMethodology Category=Recent learned provenance / streaming, Log-suppression rate=30%2026.05 | 62.5 | 75.4 | 68.4 | 13.4 | |
| GraphRAGMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 57.6 | 32.4 | 41.5 | 73.8 | |
| NODLINKMethodology Category=Recent learned provenance / streaming, Log-suppression rate=30%2026.05 | 56.7 | 80.9 | 66.7 | 8.1 | |
| GNN-LinkPredictMethodology Category=Statistical relational learning, Log-suppression rate=30%2026.05 | 44.2 | 69.4 | 54 | 35.1 | |
| Zero-Shot LLMMethodology Category=LLM & agentic baselines, Backbone=GPT-4o, Log-suppression rate=30%2026.05 | 34 | 13.1 | 18.9 | 91.7 | |
| MAGICMethodology Category=Deterministic provenance graphs, Log-suppression rate=30%2026.05 | 26.8 | 98.6 | 42.1 | 0.5 | |
| SLEUTHMethodology Category=Deterministic provenance graphs, Log-suppression rate=30%2026.05 | 19.6 | 99.2 | 32.7 | 0 |