CurveSpotter Methodology / version 0
Rules before results.
Entries are observations, not predictions or financial advice. The criteria are recorded before outcomes so the ledger can be audited on the same terms over time.
Pre-registered resolution criteria
The bars above are inclusive: ≥ includes the number itself, so a count that
lands exactly on a bar clears it. A call misses when it finishes below the bar, or when it
satisfies only one half of the D+21 condition. Misses are published in the same format as
hits.
Criteria changes are versioned and never applied retroactively. A call keeps the criteria version recorded when it was opened.
What the thresholds imply by size (v0.3)
Early, small apps
The absolute floor of +50 ratings is not size-neutral, and we would rather say so than let it flatter us. A call opened on an app with a handful of ratings has to find fifty new ones inside three weeks, which for an app that small is not a growth curve — it is a discovery event: a feature, a video, a link that lands. Those calls are lottery-grade, and we mark them as such when we open them.
Growth-range calls
A call opened in the range of roughly 500 to 2,000 ratings can clear the same bar on sustained growth alone, without needing luck. That is where the ledger's hit rate is actually being measured, and from the second batch onward it is where most calls come from. The early, tiny, mechanically-interesting apps keep one slot per batch, openly labelled as the lottery ticket it is.
The D+42 branch has no absolute floor
The two branches are not shaped alike. D+21 requires ≥2.0× the baseline and at least fifty more ratings; D+42 requires ≥3.0× the baseline and nothing else. Below a baseline of fifty, that makes the later bar the lower one. A call opened at five ratings needs 55 by D+21 and only 15 by D+42 — it can miss the first check and still resolve HIT on the second, at a number smaller than the one it missed. Three of the five calls open today sit in that region; two do not.
We are saying this now, five weeks before it could decide anything, because a bar published in advance is only worth what its least flattering reading is worth. A hit on that branch is the bar working as written, and when one is announced it will carry the baseline, the final count, both bars, and the earlier miss in the same breath. The thresholds themselves do not move: the five open calls resolve under v0 exactly as signed. A future criteria version, binding only calls opened after it is published, is expected to give the D+42 branch a floor of its own so the longer window can never ask for less.
One night, several readings
A single night can measure the same app more than once — the chart sweep, the discovery stage and the nightly tracking pass each look apps up independently, and every observation is kept. Apple serves those requests from edge caches, so a later one can return an older count than an earlier one did. The night's number is the highest count it observed. A rating count does not fall, so each observation is a floor on the true count, and a cached answer arriving late cannot unmake what an earlier request established.
These versions change no threshold. They record what the existing thresholds mean, and how the instrument underneath them reads. An open call always keeps the criteria version it was opened under.
How a call gets opened
From the second batch, a growth-based nomination requires a positive interval in two consecutive observations — three timestamps, not two. One interval cannot tell a burst apart from a trend, and we would rather miss a fast start than log a call we could not distinguish from noise. A nomination made on novelty rather than measured growth says so on its own page, and carries no velocity claim at all.
Corrections and source record
Corrections are appended with timestamps, never silently edited. The source record uses Apple's public iTunes and RSS endpoints, with snapshots recorded daily. Every velocity claim names both timestamps that define its interval.
The measured population is the US App Store, iOS only: apps released on or after 2025-01-01 in all App Store genres except Games. Each nightly tracking work list is capped at 4,000 apps, discovered through Apple's new-applications feed, per-genre top-free and top-grossing charts, and a fixed search-term sweep.
Criteria version history
Every criteria version ever used by a live call is listed here, with the date range it was in effect and what changed from the version before it. A call keeps the version it was opened under even after this table adds a new row.
- v0 – Initial version. rating count reaches ≥2.0× baseline AND absolute gain ≥50 by D+21, OR ≥3.0× baseline by D+42
- v0.1 – No threshold change; publishes what the v0 thresholds imply at different app sizes.
- v0.2 – No threshold change; publishes that the D+42 branch carries no absolute floor, so below a baseline of 50 its bar is lower than the D+21 bar.
- v0.3 – No threshold change. Instrument calibration: where one night measured an app more than once, the night's number is now the highest count observed rather than the last one received. See /corrections for the measured effect.
Timestamps
A call's open time is recorded at day granularity — 00:00Z of the day it is published — and its resolution dates are counted from that day. The baseline it is measured against carries the exact timestamp of the snapshot it came from, which is the figure the outcome is checked against. An entry is never opened with a date later than the day it is published.
Conflicts
We may hold interests in competing apps. Conflicts are disclosed on the relevant entry so readers can evaluate the observation with that context.
Selection bias, first batch — Two of the three calls in the first batch, Sentence and Normal, are screen-time control apps. The operator of this site ships a competing app in that category. The overlap is not accidental: the novelty watch that produced the first batch began in the category the operator knows best, which is also where a mechanism can be judged fastest. From the second batch onward the primary selection path is mechanical — candidates are drawn from the observed velocity band across all tracked niches rather than from the operator's familiarity. Entries that touch the operator's interests carry a conflict-of-interest notice at the top of the entry. Nothing above this line was edited to accommodate it.