Android · fork
A fork of Mihon, the open-source Android manga reader, in Kotlin. I forked it to add what the reader I use every day was missing. It installs next to the stock app, updates itself, and has shipped 21 releases. The newest feature shows where readers quit a series, measured honestly.
In one line: the series page shows how many readers dropped a title, whether that is low or high for its kind, and the chapters where they quit.
Popularity counts who started. It says nothing about who stayed. A small service on my homelab reads AniList's public reading-status counts and the latest 400 public "Dropped" lists for each title, then corrects three things a plain count gets wrong (below the chart). Pick a title, then drag to your chapter.
Solo Leveling finished, 201 chapters
3.2% low for finished manhwa
Latest 400 droppers: 289 quit at a chapter (median 87) and 110 dropped before chapter 1. A finished series has no newest chapter to catch up to.
The Divine Twilight's Return ongoing
8.3% typical for ongoing manhwa
Latest 400 droppers: 265 quit at a chapter (median 57), 76 dropped before chapter 1, and 59 had caught up with the newest release (18.2% of those checked), so they are not counted as quits.
The Legendary Moonlight Sculptor ongoing
10.0% high for ongoing manhwa
Latest 400 droppers: 165 quit at a chapter (median 40), 125 dropped before chapter 1, and 110 had caught up with the newest release (40% of those checked), so they are not counted as quits.
01 · caught up
On an ongoing series, readers who reach the newest chapter often mark it Dropped.
Counted as quits, they pile up at the newest chapter and fake a cliff. The service checks each dropper's last chapter against the series' release history on the day they dropped. Within one chapter of the newest release means caught up, not quit. Median share across 138 ongoing titles: 11.4%.
02 · peers
A 10% drop rate is below the median for finished manhua and in the top 13% for ongoing manhwa.
| Group | Titles | Median |
|---|---|---|
| finished manhua | 45 | 11.9% |
| finished manhwa | 117 | 9.4% |
| ongoing manhua | 45 | 8.6% |
| ongoing manhwa | 116 | 6.2% |
Labels are shrunk toward the group median (empirical Bayes, K = 100), so a small title cannot earn an extreme label. Under 100 readers: no label (12 of 342).
03 · cliffs
A busy chapter range is not automatically a cliff.
A range is marked only when it is a local peak and a Poisson test says the excess over its neighbours is unlikely to be noise. That cut titles with a "cliff" from 268 of 340 to 127. Today: 157 cliffs across 127 of 342 titles.
In the app, the drops server is optional, set by the user, used only on home Wi-Fi, and its address stays on the device. Readers' notes from past your chapter stay hidden until you ask. This page shows counts only: no reader names, no notes.
In one line: a push becomes a signed test build in about six minutes, and nothing ships until it passes on a real phone.
In one line: webtoons are one long vertical strip, so the reader scrolls itself at a speed you set and stops at the end of a chapter.
press play, or tap the screen.
Placeholder panels. The engine is the real one: speed changes apply mid-scroll, a tap pauses, and it stops at the chapter seam instead of sliding through. In the app the same controller also binds to the volume keys.
In one line: a title that several independent services recommend outranks one that a single service pushes.
Turn sources off and watch the ranking move.
Sample titles; the ranking rule is the app's. Each independent source that recommends a title adds one point, popularity only breaks ties. Hidden gems drops the most popular titles to surface what big lists bury.
The app source is private; the releases are public.
what it demonstrates
Kotlin and Jetpack Compose at feature scale; release engineering (signed APKs, CI, in-app updates); statistics applied with care (peer baselines, empirical Bayes shrinkage, a Poisson test for cliffs, intervals on every rate); rate-limited integration with five APIs; testing on a real phone against written expectations; and keeping a fork current with an active upstream.