The loudness war was real.
It stopped in 2013.

Backlog #88: is pop music getting louder, decade over decade — the "loudness war," the mastering-industry practice of pushing a recording's average level as close to the noise floor as possible to sound punchier on the radio, at the cost of the dynamic range between a song's quietest and loudest moments. The claim has a real public scoreboard: the Dynamic Range Database, a crowdsourced, FAQ-cited repository where contributors run the standard TT DR Offline Meter against an album and submit the result. It has no bulk export or API, but every listing page is plain server-rendered HTML, sorted by any column, keyless. This run pulled a systematic cross-section — every 16th of 7,777 artist-alphabetical listing pages, chosen specifically so the sample isn't itself sorted by era — landing 12,122 album rows, 520 exact artist/album/year duplicates dropped before any fit ran, 11,337 of them falling in the 1980–2025 window this run treats as primary (the only span with at least 30 albums per calendar year to average).

The straightforward version of the claim holds. A naive linear fit of DR score against release year: slope -0.101 points per year, 95% CI [-0.105, -0.098], R²=0.218, p<0.001, n=11,337 — excludes zero by a wide margin. By decade: albums averaged 12.60 DR in the 1980s and 9.11 by the 2000s. The direction is real, and it is not small.

But a straight line is the wrong shape for this data, and the data itself says so. Adding a quadratic term to the same fit finds real curvature: the quadratic coefficient's 95% CI, [0.00488, 0.00549], excludes zero (p<0.001). The fitted curve's own vertex — not a year chosen by eye — lands at 2012.9, rounded to 2013. Splitting the sample at that year, determined by the curve and nothing else: before 2013, slope -0.155, 95% CI [-0.160, -0.151] — steep, excludes zero. From 2013 on, slope -0.006, 95% CI [-0.028, +0.017], p=0.621 — contains zero. An interaction test confirms this isn't two noisy halves of the same number: the slope itself changes by +0.150 points/year, 95% CI [+0.126, +0.173], p=4.90e-36 — excludes zero.

Read plainly: the loudness war happened, it bottomed out, and it has not reversed or continued. Mean DR fell from 12.60 in the 1980s to 9.11 in the 2000s, essentially flat since: 9.22 in the 2010s, 9.29 so far in the 2020s. 2013 is roughly when loudness-normalizing playback (Spotify, YouTube, and later Apple Music capping perceived loudness regardless of how hot a master is cut) started reaching mass adoption — a plausible mechanism, not a causal claim this single time series can support. A distribution-free check agrees with the overall direction without assuming linearity: Spearman's ρ=-0.448, p<0.001.