Mastering Wings & Prayers

The numbers told a story the ears had only suspected. The core album — tracks 01 to 08 — was remarkably consistent: every integrated reading inside a 0.8 LU window, with 3.5 to 5.1 dB of true-peak headroom everywhere. Whoever mixed these eight songs was hearing the same record all the way through.

The four remixes were a different country. Their loudness spans 4.6 LU — from −18.2 to −13.6 LUFS — and their peaks ride as high as −1.4 dBTP. Put together, the album arrived spanning 4.6 LU: quiet lullaby next to hot dance mix, with no shared ceiling.

Track 09, Tailwind (EDM Remix), was the hardest case on the record, and the measurements say exactly why: the quietest track of the twelve at -18.2 LUFS, but with only 1.6 dB of peak headroom. It had the longest distance to travel to a release level, and the least room to travel it in — the limiting problem of the album, stated in two numbers.

None of this required taste to see. It required a meter that reports honestly, across a whole album at once — which is precisely the part of the engine that existed.

Here the honesty matters most, and the honest split is between deciding and doing. The engineer chose every target: the album level, the ceiling, and the five tracks that deliberately sit away from the album standard to protect a lullaby, a reflection, a homecoming and a remix. Nothing inferred those. WPAudio Engine's Decision Engine does not exist.

The processing itself was the engine's. Each master was produced by aurora master running a deterministic, published recipe — DC correction where it was measured, gain to the stated target, look-ahead true-peak limiting, then micro edge fades — with every step recording what it changed and which measurement prompted it. The twelve files this page serves are that command's output, unedited.

1. Analyse all twelve pre-masters. Read the reports, not the waveforms.
2. Decide targets per track — level, ceiling, and what to protect. The engineer.
3. Process. A fixed recipe, applied identically to every track. The engine.
4. Re-measure every candidate against its original.
5. Keep only versions whose measured change matches the stated intent.

That loop — measure, decide, act, measure again — is the workflow the software is being built to serve. A recipe an engineer aims is not the same thing as an engine that decides; running the loop on a real record is how we learn what the Decision Engine should eventually propose, and what it must never decide on its own.

The album gave back as much as it took. Album-level analysis — cross-track statistics, outlier detection, one report for a whole record — exists in WPAudio Engine because this record needed it. Thirty-three minutes of audio, analysed end to end in seventy-nine seconds.

A real record also breaks software in ways a synthetic signal never will. The clearest catch: digital silence was reporting an integrated loudness of 0.0 LUFS — maximal loudness, the worst sentinel a loudness metric could pick. It now reports negative infinity, and the JSON exporter serialises non-finite values as null instead of emitting invalid JSON.

One project's engine, mastering another project's music, each improving the other. That is the whole Studio thesis inside a single working directory.