# Vasundra Srinivasan argues page loading—not planning time—is the main stale-click risk [Canonical reading](https://escapemeridian.com/readings/vasundra-srinivasan-argues-page-loading-not-planning-time-is-7369d569) Vasundra Srinivasan’s paper argues that computer-use agents’ clicks go stale mainly because pages are still loading, not because agents spend several seconds planning. In tests on 31 public web pages, 17% of clicks planned in the first second after a page opened became stale within 30 seconds, compared with 0.4% planned on fully loaded pages. Waiting about one second for network activity to stop removed most of this risk on those pages, but the wait rules of eight of ten widely used open-source agents still left more than 1% of clicks stale. The paper also reports that commercial booking and buying pages often load longer and shift their layout more than .gov pages in the HTTP Archive crawl. For checks needed before a click, such as before irreversible actions or while a page is loading, it reports that pixel comparison can verify a target’s position as accurately as a frontier vision-language model at one eighth of the latency. How well do these results generalize beyond the tested pages and the HTTP Archive crawl? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Oct 29, 2026 (2026-10-29). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T10:10:05.731Z. First observed: 2026-10-30T06:45:36.869Z. Retrieved: 2026-10-31T07:32:11.370Z. - [What Does "Now" Mean to Your Computer-Use Agent? Stale Screenshots, Page Loads, and Which Check to Run Before a Click](https://arxiv.org/abs/2610.39282) ## Limits The supplied evidence is the arXiv abstract, not the paper’s full methods or results; it does not provide enough detail to assess how the tests were conducted. The reported measurements cover 31 public web pages and an HTTP Archive crawl; the abstract does not establish how well the results generalize to other pages or settings. ## Open question How well do these results generalize beyond the tested pages and the HTTP Archive crawl? ## Source-stated Srinivasan argues that page loading, rather than the planning interval, is the main source of stale-click risk. > Prior work treats the planning interval as the source of this risk. Measuring planned clicks on 31 public web pages, we find that the risk lies in page loading - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source-stated On 31 public web pages, clicks planned in the first second after opening were more likely to become stale within 30 seconds than clicks planned on fully loaded pages. > 0.4% of clicks planned on fully loaded pages became stale within 30 seconds, against 17% of clicks planned in the first second after a page opened. - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source-stated The paper reports that waiting about one second for network activity to stop removes most of the risk on the tested pages, while eight of ten widely used open-source agents’ wait rules leave more than 1% of clicks stale. > On these pages, waiting about one second for network activity to stop removes most of this risk, yet the wait rules of eight of ten widely used open-source agents leave more than 1% of clicks stale. - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source-stated For checks before a click, the paper reports that pixel comparison verifies target position as accurately as a frontier vision-language model at one eighth of its latency. > a pixel comparison verifies the target's position as accurately as a frontier vision-language model at one eighth of the latency - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source and contributor context Srinivasan, Vasundra · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Latency The delay between sending a request and receiving the result. - [Definition reference](https://developers.google.com/machine-learning/glossary#latency) ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 7b08261a-38e0-4505-8815-e8e454bafd90. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T10:10:05.731Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: a3b7eaec9132b0090cac562bce1533ad6bf71a9467e610d4b3adedf225625841. Source text hash: 0a1b1c27253ea6f093bf2694798f543c13c342cdeca1cc4d40f3489ee13753ea. ## Revision and source-check history ### Version 1 — 2026-11-11T10:10:05.731Z Vasundra Srinivasan’s paper argues that computer-use agents’ clicks go stale mainly because pages are still loading, not because agents spend several seconds planning. In tests on 31 public web pages, 17% of clicks planned in the first second after a page opened became stale within 30 seconds, compared with 0.4% planned on fully loaded pages. Waiting about one second for network activity to stop removed most of this risk on those pages, but the wait rules of eight of ten widely used open-source agents still left more than 1% of clicks stale. The paper also reports that commercial booking and buying pages often load longer and shift their layout more than .gov pages in the HTTP Archive crawl. For checks needed before a click, such as before irreversible actions or while a page is loading, it reports that pixel comparison can verify a target’s position as accurately as a frontier vision-language model at one eighth of the latency. How well do these results generalize beyond the tested pages and the HTTP Archive crawl? ## Original source arXiv cs.AI · Article Original source date: Oct 29, 2026 (2026-10-29). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T10:10:05.731Z. First observed: 2026-10-30T06:45:36.869Z. Retrieved: 2026-10-31T07:32:11.370Z. - [What Does "Now" Mean to Your Computer-Use Agent? Stale Screenshots, Page Loads, and Which Check to Run Before a Click](https://arxiv.org/abs/2610.39282) ## Limits The supplied evidence is the arXiv abstract, not the paper’s full methods or results; it does not provide enough detail to assess how the tests were conducted. The reported measurements cover 31 public web pages and an HTTP Archive crawl; the abstract does not establish how well the results generalize to other pages or settings. ## Open question How well do these results generalize beyond the tested pages and the HTTP Archive crawl? ## Source-stated Srinivasan argues that page loading, rather than the planning interval, is the main source of stale-click risk. > Prior work treats the planning interval as the source of this risk. Measuring planned clicks on 31 public web pages, we find that the risk lies in page loading - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source-stated On 31 public web pages, clicks planned in the first second after opening were more likely to become stale within 30 seconds than clicks planned on fully loaded pages. > 0.4% of clicks planned on fully loaded pages became stale within 30 seconds, against 17% of clicks planned in the first second after a page opened. - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source-stated The paper reports that waiting about one second for network activity to stop removes most of the risk on the tested pages, while eight of ten widely used open-source agents’ wait rules leave more than 1% of clicks stale. > On these pages, waiting about one second for network activity to stop removes most of this risk, yet the wait rules of eight of ten widely used open-source agents leave more than 1% of clicks stale. - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source-stated For checks before a click, the paper reports that pixel comparison verifies target position as accurately as a frontier vision-language model at one eighth of its latency. > a pixel comparison verifies the target's position as accurately as a frontier vision-language model at one eighth of the latency - [Supporting source](https://arxiv.org/abs/2610.39282) ## Source and contributor context Srinivasan, Vasundra · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Latency The delay between sending a request and receiving the result. - [Definition reference](https://developers.google.com/machine-learning/glossary#latency) ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 7b08261a-38e0-4505-8815-e8e454bafd90. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T10:10:05.731Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: a3b7eaec9132b0090cac562bce1533ad6bf71a9467e610d4b3adedf225625841. Source text hash: 0a1b1c27253ea6f093bf2694798f543c13c342cdeca1cc4d40f3489ee13753ea. ### uncertain — 2026-11-11T10:10:05.731Z — anthropic/claude-opus-5.5 Publication: 7b08261a-38e0-4505-8815-e8e454bafd90 attribution — pass: The reading attributes the paper and its findings to Vasundra Srinivasan, the sole listed author on the arXiv page. It frames the results as the paper's argument and report rather than as established fact. > Authors:Vasundra Srinivasan summary — uncertain: The figures are faithful, but the central term 'stale click' is never explained, and the summary drops its consequence. Per the source, an agent plans on a screenshot for seconds, and if the page changes in the meantime, the click lands on an unintended element. A general reader cannot see why staleness matters without this. Smallest repair: add one opening sentence restating that mechanism and noting that agents act from screenshots. 'Computer-use agent' and 'frontier vision-language model' should get term-library tooltips. The question is also duplicated at the end of the summary. > Computer-use agents plan on a screenshot for several seconds and then click at a position on the screen; if the page has changed in the meantime, the click lands on an unintended element. date — pass: originalPublished is 2026-10-29, which matches the arXiv v1 submission date. The later feed date of Oct 31 was correctly not used. > \[Submitted on 29 Oct 2026\] support — pass: Each claim matches the abstract verbatim in substance: the 17% versus 0.4% figures, the one-second wait, eight of ten agents above 1%, the commercial versus .gov contrast, and pixel comparison at one eighth of the latency. The draft adds no unsupported benefits or costs. The generalization question follows fairly from the 31-page sample and the HTTP Archive crawl. > On these pages, waiting about one second for network activity to stop removes most of this risk, yet the wait rules of eight of ten widely used open-source agents leave more than 1% of clicks stale. media — pass: The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/7369d5698fa96ce17db499d7e62121b2f4d588514cf4cbfb73c70a86e62dd6e4/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/vasundra-srinivasan-argues-page-loading-not-planning-time-is-7369d569.json)