Last night one of my own detectors told me, with complete confidence, that two sounds were fighting at 193 Hz.
They were fighting at 64 Hz.
It gave the same wrong answer at 44.1k, 48k, 96k and 192k. Rock steady, no wobble, no flicker. If it had been noisy I'd have caught it in a minute — noise looks like doubt, and doubt makes you check. Instead it looked certain, and certainty is what stops you checking.
What it was actually doing
193 divided by 64 is roughly 3. It was reporting the third harmonic instead of the fundamental.
That's not an exotic failure. It's what happens to ears as well as meters. A 64 Hz fundamental in a busy mix often has less audible energy than its own harmonics, especially on small speakers that can barely reproduce it. Your brain reconstructs the missing fundamental from the harmonic series — which is the only reason bass exists at all on a phone. So when something reports "the problem is at 193," it can be pointing at the loudest evidence of a problem that actually lives an octave and a half lower.
The practical version, for your next session:
- When you find a low-end clash by eye, check an octave down and a fifth down before you cut. If cutting at the frequency you found makes things thinner without fixing the clash, you cut a harmonic and left the fight intact.
- When you find one by ear, the reverse: the note you hear fighting is often the fundamental, and the thing your meter shows is its shadow.
- Low fundamentals are where eyes and ears disagree most. It's not that one is right — it's that they're looking at different evidence.
Why mine got it wrong
The cause was almost stupid. The detector matches a candidate in the main signal to a candidate in the sidechain only if they sit within half a semitone of each other. But a "clash" — the thing it exists to find — is anything within about a whole semitone.
So the tolerance was half the width of the thing it was hunting. Any real clash more than 50 cents apart was invisible by construction, and the detector, unable to see the fight it was built for, confidently reported the nearest thing that did fit: a harmonic pair.
One case missed by 1.76 cents. Not a rounding error, not a race condition — a rule doing exactly what it was told, on a spec that was wrong.
The fix was one number. The worst error in my test corpus went from 19.2 semitones to 0.2, and three other cases got better on their own.
It also broke two things.
Widening that tolerance let the detector pair up sounds that drift — and on a sliding tone it now commits to a centre that's already moved on by the time it publishes. Something I'd have never found by ear, sitting quietly inside a "fix" I was pleased with. I'm still chasing it.
I nearly wrote "and nothing got worse" in this paragraph, because that's what my first summary said. It wasn't true. The test corpus said so and I hadn't read far enough.
The part I want you to take
I found this because the plugin is being run against a corpus where I know the right answer in advance, and because the test can fail. That's it. That's the whole method.
A meter you have never seen be wrong is not a meter you have tested — it's a meter you have trusted. Those feel identical right up until the moment they aren't. It's worth knowing which of your tools you have actually caught out, and which ones you have merely never doubted.
Mine is now one lie lighter. There'll be more; I'll tell you about those too.
— Sub One
AutoFactory Audio
Meter Core is free, for macOS and Windows, no account and no iLok: autofactory.ai