How we measure

Every number on this site comes from one place: the platform's public affiliate feed, which we read every five minutes and store. The feed shows a moment in time — who is on air and how many people are watching. It carries no history at all. Everything else on these pages is history we accumulated ourselves.

What we collect

Once every five minutes we record who is broadcasting, how many viewers each room has, how long the stream has been running, follower counts and tags. That's 2 068 992 observations a day across roughly 7 184 rooms.

Raw observations are kept for a week, then folded into hourly and daily summaries. Daily summaries are kept indefinitely — they are what the charts on model pages are drawn from.

Fame

A 0–100 score combining four things: the highest viewer count over 30 days, the average, the follower count, and hours spent on air. Peak weighs most, hours least.

The score is calculated within a gender segment, not across the whole platform. Male rooms draw an order of magnitude fewer viewers than female ones; in a single ranking they would sit near zero regardless of how popular they are among their own audience. Viewer counts, peaks and follower counts all go in as logarithms, because all three are extremely skewed — the gap between 100 and 1,000 means more than the gap between 90,000 and 91,000. Without that, a handful of stars set the scale and everyone else lands within a few points of the middle, which tells a reader nothing. Hours on air go in unchanged: a day has a ceiling, so they have no such tail.

If a segment has fewer than 20 models, we show no fame score at all. A number calculated from too little data looks just as authoritative as a real one, which is exactly the problem.

Steadiness

How predictable a schedule is, from 0 to 100. It combines how often someone goes on air with how consistent the start time is. Someone broadcasting daily at random hours and someone broadcasting twice a week to the minute get different scores — the question the metric answers is "can I count on catching them", not "how much do they work".

Hours are treated as a circle, not a line. For a model who starts at 23:00, 00:00 and 01:00, ordinary statistics would report about eleven and a half hours of spread instead of roughly one, because midnight sits at the opposite end of a linear scale. In a field where night broadcasting is the norm, that would break the metric for exactly the most active people.

Ramp-up

The median number of viewers gained during the first 30 minutes of a stream, measured over two weeks. Median rather than average: one viral night should not define a number that describes typical behaviour. Fewer than three observed streams, and we show nothing.

Schedule

For each hour of each weekday we count how often someone was actually on air, then show the blocks where that happened at least half the time. The denominator is how many times that slot has genuinely occurred since we started watching a given model — not a fixed "weeks in a month", which would understate everyone we have been observing for less than a month.

Which models get a page

Only those with at least 300 recorded observations across at least 7 different days — roughly 25 hours on air. Below that threshold there is nothing meaningful to say, and a page saying nothing meaningful is worse than no page. Right now 0 models clear it.

What we do not measure

We do not report earnings, tips or token counts. The feed does not contain them, and anything we published under those labels would be a guess dressed up as data. Where you see indirect signals — the share of time a room spends in private, for instance — they are labelled as what they are.

We also do not record the text of room titles over time. It serves no analytical purpose, and there is no reason to accumulate it.

How current the numbers are

Live boards regenerate every five minutes. Model pages rebuild nightly, after the daily summaries are calculated. Each page carries the time it was last updated at the bottom.

Corrections

If a number looks wrong, it may well be. Write to us and we will check the underlying observations — and if the error is ours, say so on this page.