By Arash on Kaggle
Brought to ArtaQuest by Arash
STEEL is a record — a marching anthem and its LEGO-forge cover — published as one self-contained notebook. Everything a reader needs is in the document: the full lyric printed as a cell, the measuring instruments inlined verbatim with their selftests executing in-notebook, and the two generated inputs mounted as public Kaggle notebooks, linked in the provenance block, so a stranger can walk the entire chain.
The song is thirty lines in strict marching metre — every line opens on the downbeat with four stresses, every couplet a perfect rhyme by the CMU pronouncing dictionary — sung by a deep male voice at 128 BPM in A minor. The steel itself sings: born in fire and beat on stone, tempered so it springs back when bent, refusing rust before the work is won, and passed on to hands that wear its handle smooth as bone. The take that ships is the one the author approved by ear in an audition round, cited at its public kernel.
Every claim is measured on the delivered bytes by a deterministic, level-invariant judging chain (separation with its random shift disabled, speech recognition at temperature zero, inputs normalised by pure gain first): male register with an octave-folded choir correctly read as male, 95.1% of words understood, zero continuity holes against each moment's local musical context, and a lyric that fits its clock. The master is the approved take gained and true-peak limited — nothing else — chosen across a loudness ladder and charged for what it cost (-2.1 points, i.e. it measured MORE intelligible than the raw take): -9.7 LUFS for the open web, where no player normalises loudness, with true peak at -1.0 dBTP and zero clipped samples.
The cover is a LEGO minifigure blacksmith hammering a glowing blade, generated by Wan2.2-T2V-A14B (Apache-2.0) with a LEGO style adapter on both of its experts, its loop closed by measuring every candidate cut on the delivered frames — the wrap lands at 1.02x an ordinary frame step, where 1.00 is seamless by definition. Th
Permanent DOI short link: https://artaquest.com/gu/d/n9330
This સંગીત is a public Kaggle notebook that has been run. Before it published, a reproducibility checklist read Kaggle's public API and recorded four things anyone can check for themselves: the notebook is public, every one of its inputs is public, the run finished and produced these exact files, and whether it ran with the internet switched off on Kaggle's own record. The member who brought it here then requested publication and confirmed it from their own inbox.