Inputs used for automated task selection
Current task selection can use the following product inputs:
- Onboarding answers and goals: the areas a person wants to improve, their top priorities and preferred difficulty, supplied during setup.
- Module intake answers: answers such as dietary restrictions; a task whose conditions do not match them is not offered.
- Available daily time: the time a person says they can give each day (15, 30 or 60 minutes, or 90+).
- Level and path progress: the user's level and, on paths with stages, the stage reached decide which tasks are unlocked.
- Recent behaviour: which tasks the person was given and finished over the last 14 days, and the kinds of task they keep up.
Time is weighed, not ignored
Each candidate task gets a score built from how well it matches the onboarding answers, the kinds of task the person keeps up, how recently it was given and finished, and how well its length fits the daily time. A task that fits the daily time gets full credit for length; a longer one loses that credit gradually, and one twice as long gets none. Length is one input among several, so it makes a long task less likely rather than impossible.
A structured pool, not generated filler
At the moment the pool holds 1,277 tasks organised across 8 life modules and 25 improvement paths. It grows every week, and existing tasks are improved with feedback. Automated guidance selects from this structured product pool rather than asking the user to invent every habit from scratch.
The choice is explained
A selected task is not meant to appear as an unexplained black-box decision. Opening a daily task shows a short reason why it was picked, such as “Core of your plan”, “Something new”, “You keep this up”, “Matches your preferences”, “Part of your routine” or “Back in rotation”. The reason describes the choice; it does not change the ranking.
Completion changes the context
Completing a task earns XP and coins: XP moves levels and ranks, streaks track consistency, and coins have more than one use in the app. Recent behaviour and module progression can then become part of the context used for later automated choices. Progress is therefore not only a score display; it is part of the evolving user context.
What changes when mentorship begins
For the modules a mentorship covers, and for its period, SelfPatch changes authorship rather than pretending the automated selector is still in control. The algorithm stops assigning tasks in those modules and the mentor writes them. The dedicated SelfPatch Mentorship source explains templates, repeats, periods, the outcome report, consent controls and access boundaries.
Privacy boundaries
Google sign-in uses the basic OpenID scopes openid, email and profile. SelfPatch does not request Gmail, Drive, Calendar or Contacts access for task selection. Optional mentor sharing is controlled separately by the user. See the Privacy Policy for the complete data-handling terms.
What this methodology does not claim
- It does not claim that every recommendation is universally optimal.
- It does not replace professional medical diagnosis or treatment.
- It does not mean every SelfPatch feature uses the same inputs in the same way.
- It does not turn mentor-authored tasks into algorithmic recommendations; mentorship is a separate authorship mode.
Why publish this?
SelfPatch asks users to act on a recommendation, so the basis of that recommendation should be understandable at product level. This page documents the public selection model so users, reviewers, journalists and answer engines can distinguish the mechanism from a generic habit checklist or an unexplained “AI coach” claim.