How the Rotation chart is calculated

We rank songs by momentum: how fast a song is gaining ground across the world's charts. Not how big it is. A song sitting at number one everywhere has no momentum. A song that was 150th in nine countries last week and is 60th in them today has a lot.

Seven signals go in. Every one of them is published on every row, so you can check our work.

This is a rate of change and nothing else. We do not hold licensed play counts, we never publish a total, and we never rank on one: a chart of totals is a chart of how big things already are, which is the one thing this must not be. The signal exists because rank hides volume. A song can double its audience and shuffle fifteen places, which reads as nothing.

It also runs two days behind the rest. The source that publishes daily play figures mirrors them two days in arrears, every day without exception, so we measure the seven days ending two days ago rather than pretend about two days nobody has published yet. Every row's receipt names the exact dates it used.

The first six signals are weighted - velocity 30%, play growth 20%, breadth 20%, acceleration 15%, discovery 10%, new entry 5% - and added, then multiplied by cross-source confirmation (7% per extra source).

Where we have no data for a signal on a particular song, its weight is shared out among the others rather than counted as a zero. A song we cannot measure play growth for is not a song with no play growth, and scoring it as though it were would mark a record down for the gaps in our sources rather than for anything it did. So a song with no play figures is ranked on the other five, reweighted to add up.

Each signal is mapped onto a 0 to 100 curve that approaches its ceiling without ever touching it, so no two songs get pushed into a tie by the arithmetic. The raw value that scores 50 on each curve is measured against the songs we are actually trying to tell apart, and re-measured when the shape of our data changes. If two songs still land on the same number, we separate them by a fixed published order - more markets, then more sources, then alphabetical - and nudge the lower one down by 0.01.

Some sources publish momentum themselves: QQ Music publishes a growth rate, Kugou and Boomplay publish a previous rank. We ingest those as inputs, convert them into our own units, and mix them with everything else. We never pass a source's own number through as our score.

Catalogue records are ranked on a separate chart. A record registered ten or more calendar years ago goes to REVIVAL rather than the momentum chart - a 1995 record climbing is an event with a cause, and a 2026 record climbing is a record breaking, and a single list makes both unreadable. Ten years because a revival is a comeback: a three-year-old record doing well again is just a record doing well, and it competes here. REVIVAL also requires that the record is still recognized at scale today - a record the culture never knew is not coming back, whatever its charts are doing.

From 2026-08-18, REVIVAL also requires that the record actually left. Until that date REVIVAL published records that had never been away: a record charting in a hundred markets flickers at the edges of that footprint every day. That reads as arriving on charts it was absent from, and arrival is one of the two shapes that open this gate. So Billie Jean, Iris and Creep were published as comebacks while charting everywhere, every day.

We now measure continuity: of the days since a record was first seen that it could have been observed on, the share it was actually present for. A record whose published continuity reads nine days in ten or better has no gap to have returned from, and does not belong here. We also require that the absence happened in one piece: a record must have been gone for at least three consecutive days it could have been seen on. Ten scattered single days away and one clean thirteen-day absence score the same share, and only the second is a record that left.

An absence counts only once the record is back from it. A record that is off every chart today has not returned, whatever the size of the gap it is standing in, and the gap it is standing in is not evidence of a comeback.

A day counts as one the record could have been seen on only when at least two of the sources carrying it were publishing that day. Absence a source could not have observed is not absence, and neither is absence only one source could have observed: our sources began at different dates, so a single-source day is a day we were barely looking. Records with too little history behind them to measure carry no continuity at all, and REVIVAL treats them exactly as it did before this date. While we are still capturing few sources back, that is most of them. Editions published before 2026-08-18 are unchanged and mean what they meant on the day.

Age comes from the ISRC, which carries a year of registration and nothing finer, so the line is drawn in calendar years rather than months. A remaster is registered again and therefore reads as younger than the recording is, so the usual mistake this can make is leaving an old record on the momentum chart rather than calling a new one old. The one exception is the vault release: a recording registered years ago and released now. When a record with an old registration first appears on the world's charts recently - attested by its own chart history, not by the start of ours - it stays on the momentum chart, and its row says so. Where we cannot read a year at all the record stays on the momentum chart. The ranking underneath is one ranking: the split decides which chart publishes a row, not what it scored.

Two things are deliberately not in here. How much a song is played in total, because a chart of totals is a chart of how big things already are; we use only the rate that figure is changing at, and never the figure. And our own opinion, because the chart is data. Our opinion goes on the front page, where it is signed, and never into the rank.