08/16/2026
If you put the best singer in the world in a noisy room with a bad microphone, you’ve already lost. You can pay the best sound engineer on earth to work on the recording afterward, but that engineer is never going to recover what wasn’t captured or was garbled in the first place.
It feels to me like quantitative finance starts with a bad recording.
Crypto volume is scattered across God knows how many exchanges, venues and DeFi networks. With equities, dark-pool activity can move a market without appearing on the live tape, and then get reported later…so the historical tape you’re training your AI on is not the tape you would have seen live.
And then we take these imperfect recordings, turn them into bars, returns, correlations and indicators, and ask increasingly sophisticated AI to figure out what happened.
We are spending too much time on the sound engineer and not enough on the microphone.
That’s a big part of why we built Cotes around spectral decomposition. Capture the best data you can. Clean it before you destroy its structure. Separate the noise from the underlying frequency, phase, amplitude and coherence relationships.
Then train the AI.
You can’t get the singer’s pitch perfect voice back onto the tape after you’ve botched the recording.
