What a finished master can tell you, and what it can't

A record shows you where it ended up. It never shows you where it started.

Analysing a commercial release is easy. Load Abbey Road, run it through a meter, and you get its loudness, its true peak, its dynamic range, its spectral balance and its stereo width in seconds. All of it accurate.

Now try to use those numbers.

You want your track to sound like that record, so the obvious move is to copy its settings. Except you do not have its settings. You have its output — and between the mix that went into Abbey Road's mastering chain and the record that came out, there was a transformation you cannot see.

This is the central problem in reference-based mastering, and getting it wrong produces results that are not just imperfect but backwards.

Two endpoints, one observation

Every mastering decision has a before and an after. Measuring the finished record gives you the after. The before — the mix as it was delivered — was never released and in most cases no longer exists outside an archive.

So when a reference tells you "this record has a strong low end", you have learned something real about the destination and nothing about the journey. Maybe the mastering engineer added 3 dB down there. Maybe the mix already had too much and they cut 2 dB. Both produce the same measurement.

Which means the parameters split into two kinds, and they behave completely differently.

Destinations transfer. Transformations don't.

Destination parameters describe where the chain is supposed to arrive. Loudness target. Limiter ceiling. These are perfectly safe to read off a reference, because arriving there is the chain's job regardless of what came in. If a record sits at -9 LUFS with a -1.0 dBTP ceiling, aiming your chain at -9 LUFS and -1.0 dBTP genuinely reproduces that decision.

Transformation parameters describe a change applied to something. Tonal EQ. Stereo width. Compression ratio. Copying these is where it goes wrong, and the failure has a nasty shape.

Suppose your reference is a bass-heavy record, so you derive "+3 dB at 80 Hz" from it and bake that into a preset. Now feed that preset a mix that is already bass-heavy. The boost does not move it toward the reference — it compounds, and you end up further from the target than when you started.

The error is largest exactly when correction matters most. A mix that was already close would barely have needed the preset; a mix that was badly off gets pushed further off. That is the opposite of what a tool should do.

Stereo width fails the same way. The chain parameter is a multiplier, but a reference's width is an absolute ratio. Set width to a reference's absolute value and an already-wide mix ends up wider than the target, a narrow one narrower.

What engineers actually do

They listen to the gap.

No mastering engineer applies a fixed curve derived from a reference record. They put the reference on, put the mix on, level-match, and listen for the difference — this needs more weight at 100, this is harsh at 3k — and correct that. The reference is a target, not a recipe.

The measurable version of the same idea:

correction = (reference profile − mix profile) × strength
width      =  reference ratio ÷ mix ratio

Measure both sides the same way, subtract, and apply the difference. Now a bass-heavy mix aimed at a moderate reference gets cut, which is what should have happened all along. The correction is derived at mastering time from the actual audio, so the same preset does different things to different mixes — as it should, because it is describing a destination rather than a gesture.

Why not correct all the way

The strength term above is deliberately below 1.0, and that is not timidity.

A full correction drags every track onto the reference's exact spectrum. Do that and you have not mastered the mix — you have overwritten the decisions the artist made, including the deliberate ones. The guitar tone somebody spent a week on becomes whatever the reference's guitar tone implies.

Partial correction moves a mix toward a target while leaving it recognisably itself. Where exactly to sit on that dial is a taste decision, and it is the difference between mastering and transplanting.

The limits worth knowing

Even done correctly, a spectral comparison is a coarse instrument.

Split the audible range into ten octave bands and you have ten numbers describing tonal balance. That genuinely captures whether something is dark or bright, thin or heavy. It says nothing about arrangement, transient character, the flavour of the saturation, or how depth is arranged in the stereo field — and those are frequently the things that make a record sound like itself.

Two recordings can share a spectral profile and sound nothing alike. So the comparison is good for spotting extremes — "your mix is a long way from this reference and the chain will run out of room" — and poor at ranking fine differences. Any tool that reports a precise similarity percentage from ten bands is selling a resolution it does not have.

And there is a harder limit still. If you name a record we do not hold, no amount of analysis of the records we do hold produces a comparison, because the comparison needs the missing one. That question — is this substitute close to what you meant? — is not measurable at all. The only honest answer is to say what was checked, say what wasn't, and ask you.

The takeaway

A finished record is a statement of where something ended up. Treat it as a destination and it is enormously useful. Treat it as a set of instructions and it will confidently move your mix in the wrong direction — most confidently when your mix needed help the most.