Back to Reflog

/Product updates

The Reflog: launch week

Tests that learn from every pass, Scenarios with several people in them, a Scenario recorder, much bigger Discovery and Deep Scan, and a first look at Product Map.

We launched Composal on 29 October. Here is everything new in Verify since then, along with the big features we finished in the run-up to launch and haven’t written about yet.

Tests that learn from every pass

When a Scenario passes, Verify remembers how it passed. The next run replays those steps directly and only brings the AI back in when your app has changed. Runs get faster and cheaper the more you use them, and a Scenario that has fully learned your app can finish without calling the model at all.

You can see this happen. Every step is marked Learned, Healed or AI, live and in the recording, and you can open any step to see exactly what Verify did.

One Scenario, three runs
Run 1First runPassed
7 model callsEvery step worked out by the AI.
Run 2Nothing changedPassed
0 model callsReplayed 7 of 7 learned steps without the model.
Run 3Button movedPassed
1 model callReplayed 6 of 7 learned steps without the model.

Pick any step to see how Verify ran it.

An illustration on our sample app, Acme Notes. Learned steps replay straight away, so Runs 2 and 3 finish while Run 1 is still working. Pick any step to see how it ran.

Here is the real thing, a run from our demo project. The Scenario checks that sharing a note gives a read-only link, and every step was replayed from what Verify learned last time.

A Composal test run reading "Replayed 4 of 4 learned steps without the model", with both outcomes met and four passed steps, each marked with the Learned icon.
A real run on Acme Notes. The small cap icon beside each step and outcome means it was learned.

Learning is shared, too. Verify learns how each test account signs in once, then reuses it across Scenarios, Discovery, Deep Scan and every PR preview. A brand-new pull request starts with everything earlier runs already learned.

Scenarios with more than one person in them

Plenty of real flows involve more than one person. An owner shares a note, a teammate opens it, and once the owner revokes the link the teammate should be locked out. You can now test that whole story.

Scenarios can depend on each other, and each person gets their own browser. Data flows between them, so the teammate opens the exact link the owner copied. The run page shows the full timeline, which step ran when, and how the runs connect.

Execution order
share linklink seenlink revoked
OwnerBrowser 1
Waiting for step 3
Finished
TeammateBrowser 2
Waiting for step 2
Waiting for step 4

Step 5 · Teammate · Browser 2Failed

Expects "Access removed", but the note still opened. Verify reported a bug.

Depends on step 4 (Owner)Receives notes.acme.test/s/7Kq2
A connected Scenario on Acme Notes. Each person has their own browser, and a step only starts once the step it depends on has passed. Press Play, or click any step.

Record a Scenario in your browser

Some journeys are quicker to show than to write down. Run:

com verify scenarios record --guided

A private Chrome window opens on your machine. Click through the journey, and Composal writes it up as a Scenario for you to review, run and upload. Press Ctrl+Shift+M on anything that must be there, and the recording adds a check for it. Recordings can be up to 1,000 steps long, and passwords are never saved into the file.

name: Create a note
persona: Owner
steps:
  - Open Notes.
  - Choose New note.
  - Enter title Weekly sync.
  - Choose Save.
  - 'Required check for "Weekly sync": the note shows in the Notes list.'

Press Record to start.

com verify scenarios record turns what you do in Chrome into a Scenario you can review. Ctrl+Shift+M marks something that must be there, and the draft gets a check for it.

The recorder is available on Linux now, with macOS coming next.

For QA teams there is also Composal Recorder, a Chrome extension. It records across tabs, windows and sites, turns file uploads into steps, and lets you submit the Scenario and watch it run from the side panel. It’s in early access.

Discovery and Deep Scan, much bigger

Discovery explores your app and finds what’s worth testing. Deep Scan hunts for bugs. Both got a lot more room:

  • Limits doubled. A scan can now run for up to 12 hours with up to 200 testers and 96 browsers at once. On plans with a larger browser allowance, scans also default to 4 hours, 40 testers and 20 browsers.
  • Plans no longer cap how many browsers run in parallel, so a scan uses as many as its budget allows.
  • Deep Scan testers now bring QA and security know-how, and each one has its own throwaway computer where it can write and run code while it investigates.
  • Discovery keeps exploring until its time is up instead of stopping after a set number of turns.
Hours
Before: 6
Now: 12
Testers
Before: 100
Now: 200
Browsers at once
Before: 48
Now: 96
Discovery and Deep Scan limits, before and after. Plans on prepaid browser minutes start with a smaller default team, and no plan caps how many browsers run in parallel.

Every Deep Scan now ends in a proper report: a written account of what was tested, every finding, coverage charts, and a library of the screenshots and videos behind it. Each claim links straight to the moment it was seen, and you can export the whole thing as Markdown.

Coverage charts: 40 recorded test outcomes split into observed, inconclusive, issue observed and blocked, and 13 testers split by completion.

Coverage charts show what was checked, what found issues, and what is still unknown.

Real Deep Scans on Acme Notes, our sample app with a few bugs planted in it.

You can also watch a scan as it happens. The Discovery page shows what’s running now, who each run signed in as, and what it has found.

Getting started takes minutes

The easiest way to set up Composal is to let your coding agent do it. Copy one prompt, and the setup page ticks off each step as your agent finishes it. Your workspace name is filled in for you, and setup moves on to your repository as soon as your app is added.

Your first Discovery is included and starts testing as soon as your test account is ready. While it runs, every new workspace has a demo project of real test runs to look through.

A browser agent that handles more of the web

Verify now copes with more of what real apps throw at it:

  • apps on plain HTTP, IP addresses and internal hostnames
  • Google accounts with a password and two-factor sign-in
  • success messages that only show for a second
  • tabs that open on other domains
  • evidence further down the page, which it now scrolls to find
  • popups, extra windows, and search boxes inside in-app editors

Browser sessions API

You can now get a browser that stays open between requests, over HTTP or as MCP tools. Drive it directly with full Chrome DevTools control, or give it instructions in plain English. Screenshots and video come with it. See the browser API docs to get started.

Early access: Product Map and the scenario writer

Product Map turns everything Verify has seen into a map of your app. It opens as a library of your app’s flows, each with its key screens in order. Pick a screen to follow where it leads, one step at a time, or open the full map when you want the big picture.

Product Map for the Composal app: sections for Other pages, Documentation, Auth and the test workspace, each shown with a real screenshot, joined by solid lines for recorded actions and dashed lines for automatic changes.
Product Map on our own app, built from 296 screens and 1,827 states Verify has seen. Solid lines are actions a tester took; dashed lines are changes the app made by itself.

The scenario writer makes Discovery’s suggested Scenarios more dependable. Each new Scenario is tried on your real app first, fixed if it doesn’t work, and only added to your list once it has run cleanly.

Both features, and the Composal Recorder extension, are rolling out to early access customers. If you’d like to try them, get in touch.


The Reflog comes out every week. If something here helped, or you’d like to see something next, tell us.