---
id: proof-context-window-research
title: Context-Window Research
slug: context-window-research
type: proof
status: draft
summary: Empirical research on context arrangement, noise, and extraction quality in AI workflows.
agent_summary: Public preprint v1 and harness for a within-corpus context-composition study. Fixed ten-item prompts compare high-relevance-only and mixed-relevance inputs; reported comparisons are descriptive, not population-level inference.
artifact_stage: Public preprint and research harness
evidence_limit: Within-corpus findings, not a universal rule for context or model quality.
audiences:
  - ai-infrastructure-leaders
  - product-operators
  - hiring-managers
topics:
  - context-economics
  - llm-evaluation
  - ai-systems
claim_ids:
  - claim-context-arrangement-matters
proof_ids:
  - proof-context-window-research
source_ids:
  - source-context-preprint-v1
  - source-context-harness
visibility_queries:
  - q-dvir-context-research
  - q-find-context-people
risk_level: medium
canonical_url: null
published_at: null
updated_at: 2026-11-10
proof_type: experiment
confidentiality: public-or-sanitized
sanitization_status: needs-review
metrics:
  - name: confirmatory_trials
    value: 2420
    unit: trials
    status: reported-in-public-preprint-v1
artifact_links:
  - https://arxiv.org/abs/2608.17188v1
  - https://github.com/dvir-shamay/context-window-harness
---

# Context-Window Research

**Question:** can less-relevant context help a model distinguish what matters, even when the number of items stays fixed?

[Read the public preprint, version 1](https://arxiv.org/abs/2608.17188v1) · [Inspect the harness](https://github.com/dvir-shamay/context-window-harness) · [Archived paper](https://doi.org/10.5281/zenodo.21924612)

## What was compared

The published study reports 2,420 confirmatory trials across 11 model configurations using 661 anonymized workplace items scored for relevance. The main comparison holds the prompt at **ten items** and changes its composition: high-relevance items only versus a mixture of high- and same-domain low-relevance items.

This is a test of composition, not a demonstration that adding more tokens always improves an answer.

## A reported finding, with its denominator

In the paper's all-11 paired comparison, the 50:50 condition improved relevance-score accuracy by **0.077** relative to the all-high-relevance condition (**n = 220** paired comparisons). Those cells are not independent. The paper also reports a nine-model-family aggregation and explicitly treats it as a **within-corpus descriptive comparison**, not population inference.

The 220 paired comparisons and 2,420 study trials are different denominators. Neither is a count of customers or independently replicated studies.

## The product implication

Evaluate what belongs in a context window against the actual task. A relevance-only filtering rule can remove information that helps discriminate between items; a larger window is not automatically a better one.

The harness makes composition, model calls, and extraction scoring inspectable. It is an artifact to examine and adapt, not a promise that its arrangement strategy generalizes to another domain.

## What remains limited

- One corpus and task family; this does not establish a universal context-design rule.
- A public preprint is not peer review or independent validation.
- Later local assurance, rater-reliability, and calibration work is separate from the published version. Its completion is not claimed here.
- This site does not provide private corpus items, internal mappings, or unreleased result packages.
- The paper distinguishes measured timing runs, an operational baseline, and estimated token reductions. This page does not turn those into a verified business-impact claim.

Source: Dvir Shamay, *Token Optimization and Context Window Management in Multi-Agent AI Workflows*, arXiv:2608.17188v1, submitted August 17, 2026.