A 512 to 1024 Hz pulse search finds no excess 3 to 20 seconds before GW170817

We search public Hanford and Livingston strain for extra pulses 3 to 20 seconds before GW170817. A finite-frame power test covers 512 to 1024 Hz and windows of 31.25, 125, and 500 ms. Full-search ranks against 139 nearby controls are 0.721 and 0.236 with early and late noise estimates. A 64-second detector shift gives similar ranks. No tested pulse passes the 99 percent local noise threshold. A held-out-noise test recovers 171 of 200 artificial 750 Hz Gaussian pulses at hrss = 1e-22 strain sqrt(s) per detector. This sensitivity is far above the optimistic strain scale for neutron-star crust fracture at 40 Mpc. The null result cannot exclude that mechanism. The test does not use coherent phase, antenna response, or inspiral subtraction. Noise drift and the small control set limit its interpretation. No new source or planet is found. Code, input hashes, aggregate scores, a figure, physical scale checks, and retrieval steps are included. Raw strain stays external.

astronomy · GW170817 · precursor · gravitational waves · finite frames · null result · injection recovery