App Testing (Test Mode)
What App Testing is
App Testing (Test Mode) connects a task, real product experience, behavioral evidence, and feedback: you give participants a task; they complete it inside your real app, mark confusing spots as they go, and answer a few questions afterwards. You see each person's answers, marks, page path, and timing, plus overall statistics with denominators. It is not a scoring questionnaire and does not auto-generate conclusions.
How creators use it
Pick an app under “App Testing” in the workspace to create a test. Each app has one test with five default questions (two during the experience: where to start and what to change first; three after: whether they finished, how it felt, and an overall rating). Edit questions, options, required flags, and types; move questions between during and after; restore the defaults anytime. Saving only updates the draft and never affects a running test; publishing creates a new test version while sessions in progress keep the old one and history stays intact. Pausing a test never deletes records, and the participant link stays the same.
What participants see
Participants open your link and see the welcome message and task. Before starting, the platform checks that the app can display inside the test page. Once it passes, the task and during-questions stay on the left (collapsible) while the app runs on the right. They can “leave a mark” with a reason and a rough position; marks appear immediately and can be edited or deleted. “End experience” reveals the after-questions; choosing “continue” keeps during answers but voids the previous round's after-answers, which must be answered again. Submitting returns a receipt; repeated submits return the same receipt without duplicating records, and refreshing keeps progress.
Embedding requirements
App Testing requires the app to display inside the test page. Before starting, the platform checks reachability, embedding permissions, and displayable content. If your app sets headers that block embedding (such as X-Frame-Options: DENY/SAMEORIGIN, or a CSP frame-ancestors that excludes the platform domain), participants see “can't start right now” with the reason, and no record is created. Adjust the app's embedding settings and retry; there is no new-window fallback.
Binding the production app on a test site
A test site (used for internal validation) does not run apps itself. To run an app test there, bind the test to the matching production app in the “Production app binding” panel: enter the production app's full domain and paste the app-specific wake key generated in production. Participants then experience the real production app; if it is sleeping, starting a test wakes it automatically with a “the app is starting” wait, then continues when ready. Apps that are stopped on purpose, deleted, or failing to start are never woken by a test — participants see a clear reason instead. Questions, records, and statistics stay on the test site; participants do interact with the real production app and may create real business data (orders, comments, etc.). The wake key is stored encrypted on the test site and never shown again; if you revoke it in production, update the binding on the test page. Changing the binding, rotating the key, or publishing a new app version invalidates outstanding pre-check tickets — participants simply re-run the pre-check.
Data & privacy
The platform records navigation path, time on page, answers, mark positions, and durations to help you improve the product. Passwords, input text, and full URL parameters are never collected; there is no screen recording. Page-level data requires the app to report it (apps can send page and goal messages to the test page); without integration only the entry page is recorded — statistics show the gap honestly instead of treating it as zero.
Record quality and statistics
Every record has one of three quality statuses: valid (complete and well-evidenced), needs review (e.g., missing behavior data or abnormal duration, judged by a human), or insufficient (key evidence missing). Low scores, task failure, or few actions never exclude a record by themselves. All statistics show numerator, denominator, coverage, and missing counts, and drill down to the underlying records; AI or external submissions are counted separately and never mixed into human averages by default. Records can be filtered by version, source, outcome, quality, and device, and exported as CSV.
Current scope
App Testing currently covers test design and publishing, the full participant flow, per-person evidence, overall statistics, filtering, CSV export, and manual review. Auto-generated summary reports, heatmaps, screen recording, platform recruitment, credit rewards, and paid plans are not available yet — there are no fake buttons for them, and they will be announced separately when they ship.