LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management

A support or data assistant repeats work across related customer requests.

What you would change

Try the paper's memory change in a replay of past tasks, with separate memory per customer.

Try the paper's memory change in a replay of past tasks, with separate memory per customer.

Compare against

Your current memory pipeline plus a no-memory control using the same agent.

1. Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.
2. Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.
3. Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.
4. Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test.

What to measure

Task success, stale memories, retrieval misses, response latency, and all memory maintenance costs.

When to keep it

Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation.

recommendation

test

research area

memory and context

technical change

Explicit evidence-conclusion state management for agentic long-context reasoning

published

2026-10-09

The method is model-agnostic and runs on API-accessible models, so it is feasible under constrained infrastructure, but its large benchmark gains are reported on specific corpus-reasoning benchmarks and must be validated on the reader's own document collections and query mix before production adoption.

1. claim 01Paperraft inference
   Read the brief interpretation

   LongHarness replaces implicit evidence tracking in a growing interaction history with an explicit Evidence-Conclusion Coevolution loop, in which unresolved conclusions direct retrieval and new evidence revises conclusions, plus lifecycle management that keeps evidence current and traceable. The cost is additional orchestration complexity and more model calls per query, since state must be maintained, updated, and consulted throughout the interaction, increasing latency and API spend relative to single-pass retrieval. It can fail when the conclusion state itself is wrong and steers exploration away from relevant evidence, when lifecycle updates drop or mislabel evidence, and on workloads whose questions do not require cross-document evidence integration, where the added machinery yields little over simpler RAG.

   This is a Paperraft inference. It is not pinned to a verified PDF passage.

   Inspect the source
2. claim 02Paperraft inference

   Exact match on OfficeQA Pro rises from 15.8% to 60.2% (V1) and 4.4% to 61.1% (V2) with GLM-5.3-Flash versus direct reasoning; surpasses Claude Code by 12.0 points on V1 and Codex by 18.9 points on V2.

   This is a Paperraft inference. It is not pinned to a verified PDF passage.

   Inspect the source