Do DJs Actually Mix in Key? We Measured 7,055 Real Transitions

Every so often the argument restarts. Someone posts the Camelot wheel with a caption implying that anyone who needs it is a fraud, a few hundred DJs pile in, and for about a week the timeline splits into people who say harmonic mixing is the foundation of a listenable set and people who say it's a crutch that makes every set sound the same. It flared up loudly enough in 2024 that the music press covered it. It has never once been settled with data, because nobody arguing has had any: the whole debate runs on taste, memory, and what people believe they do.

We can actually check. DJs build sets in Mixgraph's Flow Builder, track by track, and every one of those running orders is a record of a decision — this track, then that one. Both tracks have an analysed key. So the question "do DJs actually mix in key?" stops being a matter of opinion and becomes a matter of counting.

We counted 7,055 adjacent transitions, from 703 sets built by 508 different DJs, measured on 10 August 2026. Here is what came back.

The short answer: mostly, yes — and by a much wider margin than chance

70.2% of transitions were a textbook Camelot move — same key, one step up, one step down, or the switch between relative major and minor. The remaining 29.8% were not.

On its own that number is close to meaningless, and this is where most stats about DJing go wrong. You need to know what 70.2% would look like by accident. If you shuffled the same tracks into a random order, the textbook moves would still come up 18.4% of the time, purely because there are only 24 keys and four of them count as a match from any given position. So the real finding is not "70%". It is that harmonic selection runs at 3.8× chance, and same-key mixing alone runs at 8.2× chance. Whatever DJs say about the wheel, their hands are doing something extremely non-random.

Every move, against what chance would give

Breaking it out by move is where it gets interesting, because two of these numbers are not what the rulebook would predict.

The move nobody actually makes

The +2 "energy boost" — jumping two steps clockwise, 8A to 10A — is taught almost everywhere as the fifth legal move, the one you reach for when you want a lift without leaving harmonic territory. In 7,055 real transitions it accounts for 2.3%, against a chance rate of 4.8%.

That is not DJs neglecting a useful tool. That is DJs actively avoiding it at half the rate you'd hit it by throwing darts. It is the only move on the chart that appears less often than randomness would produce. Whatever the +2 sounds like in theory, in practice it is being rejected — which is worth knowing before you build a set around it because a chart told you to.

The move that isn't in the rulebook

Now the opposite case. Move one step around the wheel and switch mode at the same time — 8A to 9B, or 8A to 7B — and no standard Camelot chart calls that a legal move. It doesn't appear in the three rules. DJs did it 12.1% of the time, at 1.6× chance, which makes it the single most common thing DJs do that the wheel says they shouldn't. Put another way: 41% of all "rule-breaking" is this one move.

It is not an accident, and the music theory explains why. A minor to G major differ by exactly one note. So do A minor and F major. That is the same distance as the +1 the rulebook does endorse — the chart just doesn't have a cell for it. The diagonal is a real harmonic move that the standard teaching aid quietly omits, and DJs found it by ear anyway. If you take one practical thing from this piece, take that one.

What the other 29.8% were actually doing

The obvious assumption about the rule-breaks is that DJs compensate — that if you take a harmonic risk you tighten everything else up, matching tempo hard so the clash has somewhere to hide. The data says the exact opposite, and it says it consistently.

A harmonic break comes with a bigger tempo jump, a bigger energy swing, and a much higher chance that the two tracks aren't even the same genre. Everything moves at once. That is not the signature of a DJ getting a transition slightly wrong — a mistake would look like one variable out of place, not four. It is the signature of a deliberate reset: the moment in a set where you stop developing an idea and start a different one. Key is simply one of the things being thrown out, along with tempo, energy and genre.

Which reframes the whole argument. The 29.8% isn't sloppiness, and it isn't evidence that the wheel doesn't matter. It's the punctuation. There's a fuller treatment of when that's the right call in when to break Camelot rules on purpose.

Tempo is where DJs are genuinely strict

If harmonic discipline is strong, tempo discipline is close to absolute. The median absolute BPM change across every transition is 1.0 BPM. 27.5% of transitions don't change tempo at all, 72.9% stay within 2 BPM, and 89.8% stay within 6.

Energy, by contrast, is close to a coin flip: the next track is more energetic than the last 54.9% of the time. Sets don't climb steadily. They climb slightly more often than they fall, which over a couple of hours is enough to build an arc without it ever feeling like a straight line — the shape covered in the energy flow guide.

What to take from this if you're building a set

Mixing in key is real, and it is what most DJs do most of the time. Not because a chart told them to — the rate holds up even among DJs who explicitly turned harmonic matching off, which is the closest thing this data has to a control group. It sits at 67.0% for them against 71.3% for everyone else. Switching the assistance off moves the needle by four points. The behaviour is the DJ's, not the tool's.

Treat the diagonal as a legal move. It's the biggest gap between what the wheel teaches and what DJs do, it's one note away from the moves you already trust, and it roughly doubles the pool of tracks available to you at any point in a set.

Be suspicious of the +2. Hundreds of DJs are avoiding it at half the rate of chance. That's a stronger signal than the diagram that put it on the chart.

Break key and tempo together, or not at all. The rule-breaks that occur in real sets are wholesale changes of direction. A harmonic clash on its own, with everything else held constant, isn't a move anyone here is making.

If you want to see how a specific pair scores across all six dimensions rather than key alone, Score My Transition will do it, and the mixing guide covers the mechanics per genre.

How this was measured, and what it can't tell you

The corpus is 7,055 adjacent track pairs from 703 saved sets by 508 DJs, where both tracks had an analysed Camelot key. It is not concentrated in a handful of power users: the single most prolific DJ accounts for 3.7% of transitions and the top ten together for 20.4%, with a median of 8 transitions each. Autosave drafts are excluded — those are created from the first track someone drops and are sets nobody chose to keep.

The limits, stated plainly, because they matter more than the headline:

• These are sets built in a tool that suggests compatible tracks. That is the biggest caveat and it cuts in an unhelpful direction — it should push the harmonic rate up. The key-matching-off comparison above is the best available correction, not a perfect one. Read 70.2% as an upper bound on unaided human behaviour.
• Machine-sequenced sets were removed, and they behave differently. The 449 transitions our own generator produced land 83.1% textbook-legal. Leaving them in would have inflated every figure here. Note also that the obvious column for this, is_ai_generated, is useless for it: it defaults to true and the builder never overrides it, so it reads true on 93.8% of rows regardless of who did the sequencing.
• Sets built are not sets played. A planned running order is a considered decision, which is arguably a better read on intent than a live set — but it is not a record of a room.
• Some tempo jumps are analysis artefacts. 5.2% of transitions jump more than 20 BPM, and 22.1% of those are a clean ~2× ratio — a detector locking onto the half-time pulse rather than a DJ doubling tempo. Excluding every such pair leaves the median at 1.0 BPM, so the headline tempo figures are reported unadjusted.
• This does not survive a genre or per-key slice, so we haven't published one. 7,055 rows spread across 24 keys and dozens of genres leaves cells too small to say anything honest about. Any sentence of the form "in tech house, DJs do X" is not supportable from this dataset, and we'd rather say so than publish a table that looks authoritative and isn't.

The full SQL is committed alongside the site source, in seo/tools/human-transitions.sql, block-numbered to match the sections above. Figures will drift as more sets are built; this page states the date it was measured and we re-derive rather than update in place.