---
type: "post"
id: "our-harnesstax-analysis-examines-how-harness-choice-changes-80954d"
slug: "our-harnesstax-analysis-examines-how-harness-choice-changes-80954d"
title: "Our HarnessTax analysis examines how harness choice changes cost and performance."
description: "Thesis: For production coding agents, harness design should be treated as a model- and task-dependent control surface, not a fixed wrapper around a model."
retrieval_nugget: "Thesis: For production coding agents, harness design should be treated as a model- and task-dependent control surface, not a fixed wrapper around a model."
published_at: "2026-10-17"
updated_at: "2026-10-17"
record_date: "2026-10-17"
date_kind: "published_at"
topics: ["coding-agents"]
source_urls: ["https://arxiv.org/abs/2609.20804"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:our-harnesstax-analysis-examines-how-harness-choice-changes-80954d"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:299ffa33df02d5f1cee8ec1491068662baf4ba6a489ceb26a8cdd83903736ae0","reviewed_at":"2026-10-17T23:01:32.381Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"For production coding agents, harness design should be treated as a model- and task-dependent control surface, not a fixed wrapper around a model.","observed_facts":[{"text":"The authors report 176 matched settings across four models, evaluated on SWE-Bench Verified and Terminal-Bench 2.1.","source_urls":["https://arxiv.org/abs/2609.20804"]},{"text":"They find context management more valuable as the context-window budget tightens, with most benefit coming from preventing context-overflow failures.","source_urls":["https://arxiv.org/abs/2609.20804"]},{"text":"They also report that planning becomes a cost saver for stronger models with little accuracy change, while predefined tools help models with weaker bash proficiency.","source_urls":["https://arxiv.org/abs/2609.20804"]}],"mechanism":"Context management prevents overflow failures; planning changes where execution trajectories stop; and action-space design changes the granularity at which code is written.","why_now":"As coding agents are compared across models and tasks, aggregate scores can obscure whether gains come from model capability or the harness choices surrounding it.","implications":["Teams should test harness components against their actual model and task mix rather than assume a single tool interface is universally best.","Cost and accuracy should be read together, especially when selecting a shell-only interface or adding planning and context-management steps."],"evidence_boundary":"These are findings reported in the abstract of an arXiv preprint; they do not establish independent replication or general production performance beyond the tested configurations and benchmarks.","watch_conditions":["Watch for replications that test whether the reported model-by-harness interactions persist on real engineering workloads and newer coding models."],"related_records":[],"new_branch_reason":"No prior New Runtime analytical record was supplied; this preprint warrants a distinct record because it isolates harness components in matched empirical settings."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/our-harnesstax-analysis-examines-how-harness-choice-changes-80954d/","markdown":"https://newruntime.com/posts/our-harnesstax-analysis-examines-how-harness-choice-changes-80954d.md","json":"https://newruntime.com/posts/our-harnesstax-analysis-examines-how-harness-choice-changes-80954d.json"}
related_patterns: ["harness-architecture-outlives-model-choice"]
---

# Our HarnessTax analysis examines how harness choice changes cost and performance.

## Retrieval answer

Thesis: For production coding agents, harness design should be treated as a model- and task-dependent control surface, not a fixed wrapper around a model.

Thesis: For production coding agents, harness design should be treated as a model- and task-dependent control surface, not a fixed wrapper around a model. The arXiv study varies planning, action space, and context management while holding the execution loop constant, making component-level comparison possible.

Observed facts: The authors report 176 matched settings across four models, evaluated on SWE-Bench Verified and Terminal-Bench 2.1. They find context management more valuable as the context-window budget tightens, with most benefit coming from preventing context-overflow failures. They also report that planning becomes a cost saver for stronger models with little accuracy change, while predefined tools help models with weaker bash proficiency.

Mechanism: Context management prevents overflow failures; planning changes where execution trajectories stop; and action-space design changes the granularity at which code is written. The abstract says rule-based elision before LLM summarization produced the strongest overall efficiency among tested strategies, while recoverable elision added machinery without accuracy gain.

Why now: As coding agents are compared across models and tasks, aggregate scores can obscure whether gains come from model capability or the harness choices surrounding it. This study offers a modular way to ask that question.

Implications: Teams should test harness components against their actual model and task mix rather than assume a single tool interface is universally best. Cost and accuracy should be read together, especially when selecting a shell-only interface or adding planning and context-management steps.

Evidence boundary: These are findings reported in the abstract of an arXiv preprint; they do not establish independent replication or general production performance beyond the tested configurations and benchmarks.

Watch condition: Watch for replications that test whether the reported model-by-harness interactions persist on real engineering workloads and newer coding models.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Coding Agents](https://newruntime.com/topics/coding-agents.md): Explore the coding-agents topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)