Language learning apps.

We give each market a mark out of 100 for seven things (nobody's fixed it 20 · above the norm 15 · how angry 10 · room to build 15 · room to enter 15 · getting bigger 20 · willingness to pay 5). Then we multiply the marks together. We do not average them.

Averaging lets a market hide one bad problem behind good numbers. Multiplying means one bad mark pulls the whole score down. That is the point. A market you cannot win is still a market you cannot win, however badly it is served.

Each mark has a fixed rule. Nobody’s fixed it and willingness to pay are plain shares. Above the norm maxes out at 4 times the store’s usual rate. How angry runs from star3.0 down to star1.5. Room to enter is what the biggest app does not own. Getting bigger hits 100 at 10% growth. Room to build is different. It is a sweet spot, not a size. A market can be too small to live on, or too big to break into.

If we cannot count something, we leave it out. We never guess it. The score only sorts this list. Every other number you see on a market is counted.

Every number is arithmetic over the reviews we hold, re-counted weekly: nothing modelled, nothing scored. Themes come from a frozen pattern classifier rather than a model reading reviews, so the same review always lands in the same bucket. A market rate is pooled across every app here; the store rate is the same count across 12,728 apps. A complaint is only called a gap when at least half the market carries it, at least 20 reviews are behind it, and the market’s rate is elevated above the store’s, the same cutoff used everywhere on this site, because a problem concentrated in one app is that app’s bug, and a problem every app on the store shares is nobody’s gap. This market has no full report yet, so the per-app notes are absent rather than estimated. The anger axis is counted either way: mean star of the reviews carrying each theme, from the same classifier as the counts above.