# DynaRev: Benchmarking Web Agents under Dynamic Goal Revision

Source: https://arxiv.org/abs/2610.40133
Published: 2026-10-29
Status: unreviewed source report

## Source excerpt (up to 25 words)

> Web agents are typically evaluated with goals fully specified before execution and fixed thereafter, whereas real users often revise their requirements during interaction. To study

## Review question

Compare accepted end-to-end outcomes, tool permissions, intervention rate, recovery behavior and total cost on the same workload. A polished demo does not establish unattended reliability.

No independent proof check or claim of peer review.