Measurement data and code for "Beyond the Thirty-Second Window: Duration, Not Memory, in a Fixed-Frame TTS Model"

This record accompanies the preprint "Beyond the Thirty-Second Window: Duration, Not Memory, in a Fixed-Frame TTS Model" (preprint: https://doi.org/10.5281/zenodo.23637738). It contains every script and every table that produces a number in the paper, the protocol documents, the de-identified exports of the listening panel, and the exact audio files the panel heard. It contains no model weights and not the full generated-audio set — see "What is not here" in README.md. Contents (569 files, 268.5 MB unpacked): analysis/ (238 files), audio/ (182), benchmark/ (87), data/ (30), docs/ (6), figures/ (6), protocol/ (7), paper/ (5), numbers/ (3), plus LICENSE, LICENSE-DATA, NOTICE, README.md and MANIFEST.csv. MANIFEST.csv lists every file with its size and MD5 checksum. This is the first public release of this dataset, published as version v1 on 2026-11-02 alongside version v2 of the preprint.