Participant Experience

Join Flow

1. Researcher shares a join link (e.g., labrowser.app/join/my-study)
2. Participant opens the link in their browser
3. Page shows: study title, description, a unique start code, download links for macOS/Windows
4. Participant downloads and installs LaBrowser (if they haven't already)
5. Opens LaBrowser, enters the start code
6. LaBrowser loads the study's start URL — participant begins browsing

Install Help

Participants should always install LaBrowser from the join page they were given for the study.

• If the download link fails, refresh the join page and try the same platform button again.
• If the app says it is out of date, return to the join page and install the latest build from there before entering the start code again.
• If macOS or Windows blocks launch unexpectedly, the participant should contact the researcher or study team rather than trying ad-hoc workarounds.

The Browsing Experience

LaBrowser looks and feels like a normal browser:

• Tab bar with new tab button
• Address bar with back, forward, refresh buttons
• Full-page web content
• Everything familiar — participants shouldn't feel like they're using a special tool

If the study has an experiment pane configured, it appears as a side panel alongside the browsing area. The panel can contain markdown instructions or an embedded web page (e.g., a Qualtrics survey).

Domain Enforcement

When a participant tries to navigate to a domain not in the allowed list:

• The navigation is blocked
• A friendly message explains that the site is outside the study scope
• The blocked URL is logged as a NAVIGATION_BLOCKED event
• The participant can go back or navigate to an allowed domain

Session Timeouts

If max_session_length_seconds is configured:

• A warning appears when the session nears its limit (e.g., at 75% and 90%)
• When time expires, browsing is disabled and a completion message is shown
• All events up to that point are preserved

Data Upload

Events are captured locally first, then uploaded in batches:

• Upload happens automatically in the background
• If the connection drops, events queue locally and retry
• When the session ends, a final upload sends any remaining events
• Upload status is visible to the participant (subtle indicator)

Failure Recovery

If LaBrowser is interrupted, the usual participant experience is simple:

• Study activity is saved continuously during the session
• If the internet connection drops, uploads retry automatically when the connection returns
• If the app closes unexpectedly, reopening LaBrowser can resume the session and continue syncing

Email Anonymization

When anonymize_emails is enabled (default):

• Any email address typed by the participant is replaced with a deterministic token: [email:XXXXXX]
• Same email produces the same token within a session (preserves analytical value)
• Different sessions generate different tokens (prevents cross-session tracking)
• This covers event URLs, event payloads and client logs — replacement happens before the event is written, subject to the size limit below

What it does not cover

Anonymization is a text substitution applied as each event is written, so it does not reach everything. Please read this list before describing the guarantee to participants or to an ethics board.

• Screenshots are not anonymized. An email visible on screen is captured as pixels, and no regular expression can redact an image. This is a property of screen capture, not a gap we intend to close.
• DOM snapshots are not anonymized. The captured HTML is stored as the page served it.
• Only the first 64 KB of any single value is scanned. Anything past that is stored as typed. This matters most for captured AI responses, where max_response_bytes defaults to 512 KB — eight times the scanned window — so an address late in a long response is stored in plain text.
• The url recorded alongside each captured artifact is stored verbatim.

If a study must not hold participant email addresses at all, do not rely on anonymize_emails alone: turn off screenshots and DOM snapshots, and lower max_response_bytes below 64 KB if you capture AI responses.