QuoteBench: How Matched Scores Can Hide Command-Path Failures
Abstract
QuoteBench reveals that execution-boundary parsing errors significantly reduce LLM coding agent success, and disclosing the boundary helps recover performance, showing that evaluation must account for deployment configuration rather than treating matched scores as intrinsic model properties.
LLM coding agents issue Bash commands through interfaces that may serialize, wrap, and reparse model output. Matched execution scores alone cannot distinguish command-generation errors from failures introduced after generation. QuoteBench measures this boundary with exact final-state validation on 56 one-shot tasks from 14 incident-derived families, crossing the generation contract with the execution transport around one deliberately unescaped added parser. Escaping at the interpolation point reproduces each replayed reply's raw-path outcome, so any recovery under a disclosed boundary must come from the model changing its generation. Across eight same-window configurations, replaying the same reply through the added parser lowers success by 55.4 to 73.2 percentage points; disclosure recovers 30.4 to 60.7 points for six configurations, and zero or slightly negative for the other two. Raw generation is nearly saturated at the frontier; boundary adaptation is what still separates models. GPT-5.6-sol's matched gap of -3.6 points hides -64.3 points of damage and +60.7 points of compensation. The deployment configuration reorders models: one reversal among 26 comparable pairs is unambiguous and four more sit on single-task margins. Evaluations of command-issuing agents should report the model configuration, generation contract, execution path, operating point, and final-state validator rather than treat a matched score as an intrinsic model property.
Community
QuoteBench asks a deployment question that matched scores can hide: did the model generate a bad Bash command, or did the execution interface break it afterward? Across 56 execution-verified tasks, replaying the same reply through one added parser lowers success by 55.4–73.2 percentage points. Disclosing that boundary recovers 30.4–60.7 points in six of eight configurations. The results show why coding-agent evaluations should report the generation contract, execution path, operating point, and final-state validator—not just a matched score.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- From Test-Time Scaling to Reusable Memory: Measuring Crystallization in Text-to-SQL (2026)
- Subtract, Transport, or Replay? Auditable Deletion from Language-Model Memory (2026)
- The Hidden Footprint: Making Storage a First-Class Metric for LLM Agent Evaluation (2026)
- The Recall Trap: A Recall-Maximizing Retriever Configuration Reduces Issue Resolution in Fixed-Budget Code Context (2026)
- Permission Denied: Policy-Graded Evaluation of Coding Agents in Hardened Environments (2026)
- Accuracy Hides How Language Models Fail: Measuring Failure States Under Matched Output Budgets (2026)
- Attributing Structured-Output Gains in Function Calling: Interface Alignment versus Procedural Transfer (2026)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend
Get this paper in your agent:
hf papers read 2608.13547 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Datasets citing this paper 1
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper