Methodology
How Wolf Score works
No black box. This page is the whole formula - including how often our predictions have been right.
Wolf Score is a 0-100 composite of five subscores computed nightly over public app-store data: chart positions across 30 countries, written reviews run through our NLP pipeline, version release history, and store listing metadata. Each input is percentile-normalized against every currently-charting game, so a score of 80 always means "stronger signals than ~80% of the charting catalog right now."
Missing data reweights - it is never faked. If a game has no analyzed reviews yet, Reception is simply excluded and the remaining weights renormalize. We never insert a fake neutral value, and the breakdown shows exactly which subscores had data. A separate confidence label (high / medium / low) reflects how much signal exists at all.
It is not machine learning - deliberately. Our history window is young; an ML model trained on it would be false precision. The weights below are published judgment, and the calibration table holds us accountable. If a model ever beats this transparent composite on calibrated precision, we'll switch and say so here.
Momentum
30%
Best-rank velocity over 7 days · market-spread change (countries charting now vs the prior week) · chart-day density over 14 days.
Reception
25%
Average written-review sentiment (60d) · sentiment↔star-rating gap (stars flattering the words is an early churn signal) · complaint pressure on fairness, monetization, and bug themes. Needs ≥5 analyzed reviews.
Execution
20%
Update cadence over 90 days · days since the last shipped version. Needs at least one observed version.
Market context
15%
Genre momentum (charting apps in the genre, week over week) · store footprint pattern (soft-launch and regional footprints carry a potential premium) · recency of first chart entry.
Monetization readiness
10%
IAP configured · price-ladder breadth · high-ticket ceiling. Google Play data only - Apple exposes none of this, so Apple-only apps aren't scored on it.
Breakout Watch - the part we can be wrong about
A composite score can hedge; a prediction can't. Breakout Watch flags games that are charting only in soft-launch or regional markets with top-20% momentum and healthy reception, and makes one falsifiable claim per flag: this game enters the top-50 free charts of a major market (US, GB, DE, FR, JP, KR) within 28 days.
Every flag is written to a ledger the moment it's made and resolved true or false against the charts - no deleting misses, no survivorship bias. The running tally:
218
Flags made
206
Still watching
12
Broke out
100%
Precision
Precision = broke out ÷ resolved. Base rate for random charting games is far below 5% - judge us against that.
Honest limits
Wolf Score sees what public store data shows: charts, reviews, listings, release cadence. It cannot see ad spend, retention, or revenue - nobody outside the developer can (our estimates carry explicit confidence intervals for the same reason). A high score means the public signals are strong, not that success is guaranteed. Not investment advice.
Method: wolf_v1 · weights 30/25/20/15/10 · thresholds: momentum ≥ P80, reception ≥ P60, horizon 28d, breakout bar top-50. Changes to any of these will be versioned and noted here.
Wolf Score rankings + Breakout Watch live on the Agency plan
Full ranked list, per-app breakdowns, and the prediction feed.