Add the wide_return_session wide event for app returns

[YoussefKeyrouz]

Task/Issue URL: https://app.asana.com/1/137249556945/project/1212810093780571/task/1217573106920334?focus=true
Tech Design URL (if applicable): https://app.asana.com/1/137249556945/project/481882893211075/task/1217382844057496?focus=true
API Proposals URL(s) (if applicable):

• https://app.asana.com/1/137249556945/project/1212810093780571/task/1217532302535421?focus=true
• https://app.asana.com/1/137249556945/project/1212810093780571/task/1217532302535422?focus=true

Description

This adds wide_return_session. wide_return_session starts after the app came to the foreground.

Data

• feature.data.ext.after_idle: true only when the inactivity treatment was applied.
• feature.data.ext.landed_on: ntp, ntp_user_initiated, web, serp, or duck_ai. LUT and
  Specific Page returns are classified by their actual destination.
• feature.data.ext.time_away_ms_bucketed: lower-bound bucket 0, 60000, 300000, 900000,
  1800000, or 3600000.
• feature.data.ext.focused: one snapshot of whether the visible landing input had focus immediately
  after the landing fragment resumed. Later focus changes do not alter it.
• feature.data.ext.page_engaged, back_pressed, opening_screen_changed, close_tab_tapped, and
  burn_tab_tapped: sticky booleans, finalized as true or false.
• feature.data.ext.session_duration_ms_bucketed and
  feature.data.ext.time_to_first_interaction_ms_bucketed: lower-bound duration bucket 0, 1000,
  5000, 10000, 30000, 60000, 300000, or 600000. Time to first interaction is omitted when
  the only terminal is app backgrounding and no interaction was recorded.

Successful terminals set feature.status=SUCCESS and feature.data.ext.status_reason to search_submitted, url_submitted, ai_prompt_submitted, return_to_page_tapped, tab_switcher_selected, favorite_selected, or chat_selected.

Backgrounding sets feature.status=CANCELLED and feature.data.ext.status_reason=app_backgrounded. If the process dies with an open flow, next-process cleanup emits feature.status=UNKNOWN and feature.data.ext.status_reason=app_terminated.

The event is gated by returnSessionWideEvent FF and the global wideEvents framework flag.

Steps to test this PR

Prerequisites

• Install an internal build and complete this setup:
  • Set SAMPLING_PROBABILITY = 1.0f in RealReturnSessionWideEvent.kt for testing only. This will allow you to see every pixel and not a sampling of it.
  • In Internal Settings > Feature Flags, make sure wideEvents, enqueueWideEventPixels, sendWideEventsViaPixels, and returnSessionWideEvent.self are ENABLED.
  • Force-stop and relaunch if you change FF for them to take effect.

How to monitor

• Filter Android studio log cat on Pixel sent: wide_return_session to view all the pixels sent at the end of the session (you won’t see a pixel until you finish the session)
• You can also filter on RealReturnSessionWideEvent to view individual events as the session record them
• Alternatively you can run this comment (requires rg installed) in a terminal to view all events
  adb logcat | rg 'RealReturnSessionWideEvent|Pixel sent:.*wide_(return|post_idle)_session_[cd]'.

Instructions

1. Set Settings > General > After Inactivity to 1 minute, unless a row says Always.
2. To start with no active session. If the previous case left the app backgrounded, reopen it, submit a
   throwaway search or click on switch tab to finish that session, wait for its pixel confirming the session ended.
3. When you perform the row's action. Wait for Pixel sent: ... wide_return_session_c; the first event of the
   day also sends wide_return_session_d. Inspect the encoded feature.data.ext.* values.
4. Short keys in the tables, such as landed_on, mean feature.data.ext.landed_on.
5. When the instructions say: “Finish the session” it means perform any action that would trigger the session to finish. Easiest ones are submitting a throway search or clicking on the tab switcher.

Landing, focus, submission, and coexistence

• Run each row with the instructions. Inspect the listed fields

Coverage	Setup	Action	Expected log values
Ordinary NTP and focused input	launch app on NTP and leave its input and keyboard open	Finish the session	after_idle=false, landed_on=ntp_user_initiated, focused=true
Ordinary web, unfocused input, page engagement, and URL submission	launch app on a website and touch/scroll in the WebView	Touch the WebView, then submit a url (example.com)	after_idle=false, landed_on=web, focused=false, page_engaged=true, status_reason=url_submitted
Ordinary SERP and Back	Build Back history ending on a DuckDuckGo results page, put app in background and the start it again	Press Back once while remaining in the browser, then finish the session	after_idle=false, landed_on=serp, back_pressed=true
Ordinary Duck.ai and native AI submission	Open app on NTP, Enable open Duck.ai from the menu (hamburger menu) without submitting	Submit a real non-empty native prompt	after_idle=false, landed_on=duck_ai, status_reason=ai_prompt_submitted
After-idle NTP, search submission	use Always > New Tab Page for idle config, with hatch enabled.	Trigger a idle return then Submit a search from the address bar	Return event: feature.status=SUCCESS, after_idle=true, landed_on=ntp, status_reason=search_submitted.
After-idle LUT	Use Always > Last Opened Tab on https://example.com	Trigger an idle return then Finish the session	after_idle=true, landed_on=web
Changing opening screen	Start a session on the browser with Last Opened Tab selected in the idle menu	Open the idle menu and Change the option from Last Opened Tab to New Tab Page. Return to browser and submit a search to end the session	opening_screen_changed=true

Remaining terminals, negative case, and interaction flags

The after-idle NTP row is the primary feature.status=SUCCESS check. For successful rows below,
inspect only the unique status_reason. Rows ending mechanically use the baseline action without
re-asserting search_submitted.

• Run each remaining unique case:

Unique case	Action	Expected log values
Favorite	From a fresh NTP return, tap a favorite	status_reason=favorite_selected
Existing chat	From a fresh Duck.ai return, select an existing chat suggestion without typing	status_reason=chat_selected
Tab switcher	From any fresh return, tap the Tabs button	status_reason=tab_switcher_selected
Return to Page	Use Always > New Tab Page from a web tab, then tap Return to Page	status_reason=return_to_page_tapped; landing treatment was already verified in the after-idle NTP row
Empty Duck.ai negative	From a fresh NTP return, open Duck.ai with empty input but do not submit; return to the NTP and finish the session using the baseline action	No event may complete before the baseline finish, especially none with status_reason=ai_prompt_submitted
Background with no interaction	Press Home without touching the landing page or input	feature.status=CANCELLED, status_reason=app_backgrounded, session_duration_ms_bucketed present, time_to_first_interaction_ms_bucketed absent
Process cleanup	While a fresh web return is foregrounded, run adb shell am force-stop <package>, then relaunch	Identify the prior flow's cleanup separately from the new return: feature.status=UNKNOWN, status_reason=app_terminated; landing and time-away classifications were already verified above
Close tab	Use Always > New Tab Page from a web tab, tap Close Tab, then finish the session using the baseline action	close_tab_tapped=true, burn_tab_tapped=false
Burn tab	Use Always > New Tab Page from a web tab, tap Burn Tab, then finish the session using the baseline action	burn_tab_tapped=true, close_tab_tapped=false

Timing

• Trigger various sessions and verify that time_to_first_interaction_ms_bucketed, time_away_ms_bucketed, and session_duration_ms_bucketed reflect the correct time buckets.

Feature gate and reset

• With no active session, turn only returnSessionWideEvent off, force-stop/relaunch,
  open an NTP, press Home, immediately reopen, and finish using any action. Confirm no
  wide_return_session.

UI changes

No UI changes

---

Note

Medium Risk
Touches first-screen and omnibar submission paths used on every app open. Changes are mostly additive telemetry behind a feature toggle, but misclassifying user vs internal navigation could skew session metrics.

Overview
Adds a return_session wide event that runs from each foreground return until a terminal action (search, URL, AI prompt, hatch actions, background, etc.), capturing landing surface, after-idle treatment, bucketed time away, session duration, and engagement flags.

RealReturnSessionWideEvent starts after FirstScreenHandler resolves landing (after_idle, landed_on as NTP, web, SERP, or Duck.ai). It listens via BrowserInteractionsPlugin / HatchInteractionsPlugin and is gated by ReturnSessionWideEventFeature. ShowOnAppLaunchOptionHandler now returns ShowOnAppLaunchResult so idle NTP/LUT treatments match telemetry.

BrowserTabViewModel splits user vs internal query submission so restoration and error recovery do not fire search/URL classifiers. Landing focus, back, Duck.ai prompts, hatch close/burn, and JS reportMetric submissions feed the same session. NtpEngagementTracker reports NTP body engagement once per app open.

^{Reviewed by Cursor Bugbot for commit 1d39ae0. Bugbot is set up for automated code reviews on this repo. Configure here.}

[YoussefKeyrouz]

• Add page_type and source to prompt submission pixels #9568
• Add duck.ai entry point pixel #9550
• Add the wide_return_session wide event for app returns #9516 👈 (View in Graphite)
• Add the m_app_return pixel, fired on every foreground return #9502
• Add toggle_visible to the native omnibar-shown pixel #9479
• Report the toggle's default side on unified input submissions #9467
• Report Duck.ai capability state on the daily feature-state pixel #9465
• develop

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

Privacy Review task: https://app.asana.com/0/69071770703008/1217579681353955

[malmstein]

Reviewed statically, stacked PR so no CI APK. CI is green, the three red checks on the PR page are a superseded cancelled run.

Genuinely well built for its size. Session start is deferred until FirstScreenHandler has applied the opening-screen decision, so after_idle and landed_on are authoritative at flowStart rather than guessed. abortIfDisabled() on every entry point aborts rather than skips, with a comment explaining that a session started while enabled would otherwise linger and get force-completed as Unknown after a kill-switch flip. The app_terminated default in the start metadata pairs correctly with OnProcessStart cleanup. Mutations are serialised through a mutex on a single-parallelism scope, matching RealPostIdleSessionWideEvent. The -api additions on BrowserInteractionsPlugin and HatchInteractionsPlugin have default bodies so existing implementers are untouched, and the API Proposal check passes. landed_on matches ReturnSessionLanding exactly, every latency is bucketed, and nothing carries PII.

Two small non-blocking things inline, neither affecting behaviour.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

There are 4 total unresolved issues (including 3 from previous reviews).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 7f795b8. Configure here.}