Show HN: Gravity doesn't track mass, it tracks waveform complexity

A bandwidth-limited neural network (that I've named Erebus) trained on public LIGO gravitational wave data finds that persistence — how much temporal accumulation helps predict structure — correlates with spectral entropy of the waveform (r = +0.69, p = 2×10⁻⁵) and not with mass (ρ = −0.05, p = 0.80). Two black hole mergers at identical mass differ by 100× in persistence. The difference is entirely explained by the spectral complexity of the signal, not the mass of the source. The pattern reproduces across four architecturally distinct observers (GRU, LSTM, Transformer, ViT-Small).

A companion paper establishes the framework: fix an observation pipeline and measure persistence across 13 real-data domains from independent instruments (LIGO, EHT, CMB, sunspots, quasars, supernovae, neutrinos). A single temporal axis organizes all domains, with electromagnetism at one pole and gravity at the other. A periodic signal through the same pipeline produces zero positive persistence across 180+ runs — accumulation destroys the wave when observation boundaries don't divide the period.

First Paper: https://zenodo.org/records/19323952 Second Paper: https://zenodo.org/records/19341889

Companion piece to the first paper: Light and gravity are opposite poles of observation (less technical) https://www.wvrk.org/works/the-structure

Erebus (the underlying system): https://erebus.org