Moodsearch / Onboarding & retention
Sign in
Use your approved dae.one email.
- Onboarding entrants
- First check-in completed
- Saved an account
- Notifications authorized
Onboarding funnel
Starts when onboarding reaches the server. Result completion does not prove the result screen was viewed.
| Stage | People | % of entrants | From prior step |
|---|
Completed-check-in retention
Day 0 is the first completed check-in. Day 1 means another completion on the next calendar day. Each person keeps their original onboarding version after upgrading.
Each cell: retention %, then returned / eligible. Blank = day not fully elapsed. Eastern time. Scroll for later days.0% 100%
Notification permission
Latest reported iOS state among saved accounts. “Undecided” means that installation reports no Allow/Don’t Allow decision; it does not tell us whether our invitation was skipped.
| Latest reported state | People | % saved |
|---|
By onboarding version
New onboarding identities in the selected date range. This is not total installs or the number of existing users who upgraded.
| Version | Entered | Completed | D1 |
|---|
D1 includes only cohorts with a fully elapsed next day. The audience buttons above filter the retention chart; this version comparison includes all permission groups.
Definitions, evidence, and what is missing
- Cohort: a new identity with a recorded pre-auth onboarding consent, attributed to 1.4.2 or 1.4.3. Known internal test identities and returning old accounts are excluded. Reinstalls or account recreation can still count as separate identities.
- Version: inferred from the first authentication-session native build near entry (158 → 1.4.2; 159 → 1.4.3), with same-onboarding diagnostic evidence for deleted accounts. This is not a version stamped directly on every funnel event.
- Date filters: select when people entered onboarding. We then follow those same people through the observation cutoff, even if they complete later. All calendar days use America/New_York, not each device’s local timezone.
- Question progression: accepted answers include explicit Skip. “Including retries” counts anyone who reached the step in an onboarding attempt. “First attempt” counts only the first guest check-in. Saved accounts and notification grants remain lifetime observed outcomes.
- Completed check-in: a stored simple check-in has a completion timestamp. Some deleted-account results are inferred as described below. Drafts, abandoned sessions, previews, and imported backfills do not count. D1/D2 are exact-day retention, not a continuous streak or “returned on or after” that day.
- Notification conversion: the funnel counts a saved account observed as authorized at least once. The state table uses its latest client report; these can differ. Undecided means
not_determined, not rejection. Our own invitation can be skipped without showing the iOS dialog. Reports can also come from different installations; an account may have an earlier grant and a later undecided report. Neither measure proves delivery or the device’s present permission state. - Push retention groups: latest reported permission on the first completion day (Day 0), including the invitation after the result. Authorized = enabled; denied or undecided = not enabled; no usable report = unknown. Later permission changes do not change groups. Today’s Day-0 classifications are partial and do not enter D1+ yet. This compares observed cohorts, not the causal effect of sending notifications.
- Deletion: probable onboarding completions are inferred from successful AI-result calls and a unique matching guest-request interval before signup. The AI call is recorded before result validation and storage, so this does not prove a result was saved. No deleted check-in text is recovered. The funnel and retention include these inferred outcomes and label them.
- Account save: an authenticated signup recorded by the observation cutoff, including an account created shortly before a guest check-in was absorbed. This is an eventual account outcome; it is not proof that every signup occurred after the result.
- Missing: downloads, pre-server onboarding screens, result-screen views, and a live permission or delivery audit. Unknown version attribution is excluded rather than guessed. Guest-to-account transitions are deduplicated where the server preserves identity or records absorption.
The September 28 preset preserves the earlier 59-person analysis. The September 30 preset adds later entrants and return check-ins. These are frozen production observations, not sample numbers.