# The hardest parts of web testing, each one run for real

> Iframes, new tabs, downloads, uploads, drag and drop, slow pages, video, email codes, two sites at once. Why each is hard to automate, and a real run of each.

By Jason Nguyen, Founder, Testsy. Published 2026-11-20. https://testsy.dev/blog/the-hardest-parts-of-web-testing/

## A browser test is easy until it is not

Open a page, click a button, read some text. That part of web testing has been solved for years, and it is the part every tool demonstrates.

The trouble starts at the edges. The control is inside another document. The answer is in a file, not on the screen. The action is a gesture with a path. The page is not ready yet, or never will be. These are the tests teams leave to a person, and Selenium's own documentation goes as far as listing some of them, [file downloads](https://www.selenium.dev/documentation/test_practices/discouraged/file_downloads/) included, under practices it discourages automating through the browser.

We sorted the hard parts by **why** they are hard, and ran each one as a plain-language test on Web Lab, our practice site. Twenty runs in all. Every section links to a post with the test text, screens from the run, the result page and the review video.

## 1. The page is more than one page

A script looks for a control in the document it is holding. These put the control somewhere else.

- **[An iframe, Shadow DOM and a new tab](https://testsy.dev/blog/test-iframes-shadow-dom-and-new-tabs/).** A form embedded in a frame, a counter inside a component's private tree, and a link that opens a second tab. 7 checks, 47 seconds.
- **[Alert, confirm, prompt and permission prompts](https://testsy.dev/blog/test-browser-dialogs-and-permission-prompts/).** Boxes the browser draws over the page: accepted, cancelled, answered with text, allowed and blocked. 7 checks, 45 seconds.
- **[Sign-in and saved data after a reload](https://testsy.dev/blog/test-that-sign-in-survives-a-reload/).** What the page remembers is not what was stored. The test reloads twice and compares. 11 checks, 1 min 31 s.

## 2. The answer is not on the screen

- **[What is inside a download](https://testsy.dev/blog/test-what-is-inside-a-downloaded-file/).** A CSV, an Excel workbook, a PDF, a photo and a ZIP, each opened and compared with the page: row counts, a sheet name, a total on page 2, pixel dimensions, archive entries. 15 checks across two runs, 41 and 56 seconds.
- **[File uploads](https://testsy.dev/blog/test-file-uploads-good-fake-and-too-big/).** A real photo, a text file named `.jpg`, a 52 MB image, a good CSV, a broken one, and two documents dropped onto a drop zone. We attached no files. 17 checks across two runs, 41 and 52 seconds.
- **[A video that really plays](https://testsy.dev/blog/test-that-a-video-really-plays/).** Press play and measure whether the time moves. Check the captions. Find the video that cannot play and read what the page says. 8 checks, 49 seconds.

## 3. It is a gesture, not a click

- **[Drag and drop, resizing and a canvas](https://testsy.dev/blog/test-drag-and-drop-resize-and-canvas/).** Four kinds of drag and two strokes on a canvas. The slowest run of the series, and the result shows the drags that needed a second try. 15 checks, 3 min 33 s.
- **[Date pickers, hover, right-click and the keyboard](https://testsy.dev/blog/test-date-pickers-menus-and-keyboard-shortcuts/).** A custom calendar, a tooltip that only exists under the pointer, a context menu, Tab, Ctrl+K and Escape. 24 checks across two runs, 1 min 18 s and 2 min 42 s.

## 4. It is not ready yet

- **[Slow pages and failed requests](https://testsy.dev/blog/test-slow-pages-without-sleeps/).** A dialog that mounts late, a report that loads, a button that only works once its animation ends, a request that returns 404. No wait times in the test. Every action landed once. 19 checks, 1 min 57 s, and 5 checks, 33 seconds.
- **The page that never becomes ready.** The same post has the run we wanted to end badly: a control that stays on "Loading…" for ever. The verdict came back **Blocked** in 37 seconds, with what was expected and what was seen. It did not wait for a timeout and it did not report a pass.

## 5. All of it in one sitting

- **[A workflow across a frame, an upload, a second tab and a download](https://testsy.dev/blog/test-a-workflow-across-tabs-uploads-and-downloads/).** Upload a file inside a sandboxed frame, validate, approve in another tab, refresh the first, download the manifest and read it. The 5 units that went in at the start were in the file at the end. 6 checks, 1 min 55 s.

## 6. It leaves the page, or the site

- **[An email code, a wrong code and a resend](https://testsy.dev/blog/test-email-verification-codes-wrong-code-and-resend/).** Each run has its own mailbox. The test signs up, is refused with a wrong code, reads the real one from the email, then asks for a second code and checks the first has stopped working. 15 checks across two runs, 2 min 53 s and 57 seconds.
- **[An admin site and a public site, as two people](https://testsy.dev/blog/test-an-admin-site-and-a-public-site-as-two-people/).** An editor publishes one article and keeps another as a draft. A visitor in a separate browser must see the first and not the second. Then the same journey across two web addresses. 18 checks across two runs, 2 min 30 s and 2 min 29 s, and one earlier run that came back Inconclusive because its steps were done out of order.

## What the runs have in common

Every test is a paragraph of plain text with no selectors, no waits and no code. "The embedded form", "the new tab", "the downloaded file" are written the way you would say them to a colleague.

Every result has the same parts: each check with the value it was decided on, a replay of every step, a review video, and for a problem, what was expected, what was seen and how to reproduce it.

The results are also honest about effort. Where a drag had to be repeated, a shortcut took more than one try or a paste was refused by the browser, the check says so. You can open every run: each post links to its own, read-only, until 18 February 2027.

## What this series does not show

- **An SMS code on a website.** A run can have its own phone number on paid plans, and [Test sign-ups that send an email or SMS code](https://testsy.dev/blog/test-sign-ups-that-send-an-email-or-sms-code/) explains it, but our practice site sends no text messages, so there is no run of it here. The phone version has one: [Test SMS code sign-in on Android](https://testsy.dev/blog/test-sms-code-sign-in-on-android/).
- **A real payment form.** The iframe post explains the idea. A run against a payment provider's test mode needs a site that has one.
- **Bot walls.** A site that challenges cloud browsers stops the run, and the verdict says Blocked. See [Network and allow-listing](https://testsy.dev/docs/network/).
- **Other browsers.** Every run here is Chrome on a 1280 × 800 screen.

The same exercise for phones is in [The hardest parts of mobile app testing, each one run for real](https://testsy.dev/blog/the-hardest-parts-of-mobile-app-testing/).

## Try one

Pick the item on this list that your team tests by hand. Paste your site's address into **New test**, write three sentences about it, and run it. The [quickstart](https://testsy.dev/docs/quickstart/) takes a few minutes.
