Paper 2606.02373
Harness-1: Reinforcement Learning for Search Agents with State-Externalizing Harnesses
- Published
- Jun 2026
- Research lab
- Independent
- Citations
- 2
- GitHub
- 941 stars
01 In brief
Summary
Harness-1 is a 20B search agent trained with reinforcement learning inside a stateful search harness.
The harness maintains environment-side working memory (candidate pool, importance-tagged curated set, evidence graph, verification records, compressed/deduplicated observations, budget-aware rendering), while the policy retains semantic decisions: what to search, which documents to keep, what to verify, and when to stop.
Across eight retrieval benchmarks (web, finance, patents, multi-hop QA), Harness-1 achieves 0.730 average curated recall, outperforming the next strongest open search subagent by +11.4 points and remaining competitive with larger frontier models.
Gains are strongest on held-out transfer benchmarks (+17.0 points mean vs.
+7.9 on source-family), suggesting RL over explicit search state generalizes beyond training domains.
Training uses SFT (899 filtered trajectories from GPT-5.4 teacher) followed by on-policy CISPO RL on SEC data (3,453 queries), with a reward combining set quality, trajectory coverage, answer evidence, tool diversity, and turn penalties.
Ablations show importance tags, sentence compression, auto-seeding, and evidence graph are key mechanisms; disabling all harness mechanisms drops recall by 12.2%.
The harness contributes a +4.2 point recall gain for GPT-5.4 alone, validating the stateful cognitive offloading principle.
02 From the paper
Abstract
Search agents are often trained as policies over growing transcripts: the model must decide how to search while also remembering what it has seen, which evidence is useful, which constraints remain open, and which claims have actually been checked. We argue that this formulation puts too much routine state management inside the policy: reinforcement learning is forced to optimize both semantic search decisions and recoverable bookkeeping that the environment can maintain more reliably. We introduce Harness-1, a 20B search agent (retrieval subagent) trained with reinforcement learning inside a stateful search harness. The harness maintains environment-side working memory, including a candidate pool, an importance-tagged curated set, compact evidence links, verification records, compressed and deduplicated observations, and budget-aware context rendering. The policy retains the semantic decisions: what to search, which documents to keep or discard, what to verify, and when to stop. Across eight retrieval benchmarks spanning web, finance, patents, and multi-hop QA, Harness-1 achieves 0.730 average curated recall, outperforming the next strongest open search subagent by +11.4 points and remaining competitive with much larger frontier-model searchers. Its gains are especially strong on held-out transfer benchmarks, suggesting that reinforcement learning over explicit search state can produce retrieval behaviors that generalize beyond the training domains. Our code is available at https://github.com/pat-jj/harness-1.