Journal — StormNet
Prime resonance as an exploratory detector
A scoring rule that rewards clean, prime-indexed peaks in a spectrum. It was run on the storm data as an exploration, and that is all it is.
Alongside the centroid analysis, the January work tried a second scoring rule on the storm audio. It deserves a short entry because it was used, and because the Lab has a rule that anything used on real data gets written down with its status.
The rule
Take a window of audio, compute its spectrum, and rank the peaks. Score the window higher when the strongest peaks fall at indices that are prime numbers and are cleanly separated, and lower when peaks fall at indices that share divisors and crowd together. The intuition is that a signal produced by one physical process rings at a few clean frequencies, while a mixture of processes produces a muddy tangle of harmonically related peaks. The name "prime resonance" describes that preference.
What was done
On a synthetic sixty-second signal, thirty seconds of low-frequency rumble followed by thirty seconds with added high-frequency crackle, the rule produced a neutral transition score, meaning it did not clearly separate the two halves. The centroid did separate them. On the real storm data, a cross-assistant run reported clean hops in the late-storm window. That report was not reproduced independently.
Status
The rule has no theoretical justification beyond the intuition above, no calibration, and one synthetic run that did not favor it. It stays in the archive as an exploration, with its code, so that a later reader can see what was tried and why it was set aside. The Lab's working detector for the published dataset is the plain statistical event detector described in the open-source entry, not this one.