All originals
HabitShipped 20245-week sprint

Lil Nudges

Small pushes, timed for the moment you can take them.

Scroll for the technical layer
Layer 01 — The world

A single amber ember drifting through an obsidian room, catching on one surface at a time.

A habit app that sends one nudge at the moment you are actually free, then gets out of the way.

Interface01
Motion02
System03
Layer 02 — The build

X-ray the interface. See the engineering.

The challenge

Reminder apps fail because they guess. A 7pm push lands in dinner, a 9am push lands in a commute, and the user learns to ignore the app within a week.

Architecture decisions

  • 01Context-aware delivery using screen-activity windows and a learned per-user quiet model.
  • 02Nudges are the only notification type, capped at two per day by design.
  • 03Habit state stored once per habit, with history as an append-only log.

Build process

Week 1Instrumented a week of real notification response times.
Week 2Prototype: one nudge, one tap, one confirmation.
Week 3Timing model and quiet-hour learning.
Week 4Cut the streak mechanic entirely.
Week 5Ship, then tune the cap from live data.

Stack

Client
React Native · Expo
Delivery
Adaptive local notifications
Model
Per-user timing window
Analytics
Completion, not opens

From the codebase

// Reward the completion, never the open
onNudgeComplete(() => scheduleNext(learnQuietWindow(user)))

Measured results

41%

week-four retention

2/day

hard notification cap

−63%

notifications ignored

The technical layer holds the challenge, the architecture decisions, the week-by-week build and the numbers. Open it when you want to judge the engineering, not the screenshots.