Harness-G: Reformulating Search Agent Retrieval as Finite Action Selection on Graphs
_reachsumit · x · 2026-07-31
Harness-G is a graph-structured framework designed for reinforcement learning (RL) search agents. The paper identifies a phenomenon called retrieval-equivalence collapse during the training of models like Search-R1: agents continuously generate distinct query strings for the same question, but their accumulated evidence sets increasingly overlap, leaving trajectories with little effective retrieval contrast.
To address this, Harness-G redesigns the policy-environment interface by reformulating free-form query generation into finite action selection over a graph. The policy selects an evidence sentence, an entity, or chooses to answer, while the environment constructs the action menu, tracks retrieval state, and validates execution. The framework also introduces Structured Non-myopic Credit (SNC), which uses a frozen answer scorer to compare selected actions against their alternatives, optimizing the agent's multi-step retrieval decisions.
Related event: Harness-G Tackles Retrieval Aliasing in Search Agents(2 posts)→
More from coding & agent
- Practical Tip: Use AI to Generate Controllable Structures Instead of Final Products — jmugan · 2026-07-31
- Giving AI Agents Modal Compute Tokens: Train Models, Don't Hack the Pentagon — drscotthawley · 2026-07-31
- External AI Agent Connects to Game via MCP to Generate Card Match Replays — tristanbob · 2026-07-31
- Developer Shares Round 2 Progress of ZDC Agent Workflow Experiment — doodlestein · 2026-07-31
- LLM Agent Autonomously Conducts CT Reconstruction Research, Matching SOTA with 969 Parameters — maier_ak · 2026-07-31
- New Benchmark: 200+ Sokoban Rooms to Test Agent Planning Skills — generativist · 2026-07-31