The Process

1 · The tape

Audience and soundboard DAT recordings, 1998–2001, transferred to a single whole-show WAV per tape (16–24-bit, 32–48 kHz — whatever the tape was).

3 · Diagnosis

Every exported track is measured before anything is processed: integrated loudness, true peak, loudness range, DC offset, a clipping scan, a DAT dropout scan (held or muted samples from tape error concealment), a digital click and discontinuity scan (corrupted or skipped samples — a step in the waveform is a defect when it dwarfs its surroundings, which separates real glitches from claps), an effective-bandwidth check against the file’s sample rate (catches mislabelled rates and lossy-sourced files), and channel health — left/right balance and phase. Any defect is reported with its exact timestamp, and a spectrogram image per track can be generated for visual inspection of suspect tapes. A track that fails goes back to the editing stage — nothing clipped gets published.

4 · Loudness normalization

Each track is brought toward the archive standard — −20 LUFS integrated with a −1 dBTP true-peak ceiling — using linear gain: one constant boost for the whole track, no compression, so the music’s own dynamics stay intact. loudnorm/ebur128 only measure the loudness; the gain itself is applied with a plain volume filter, never loudnorm’s own renderer, so the frame-adaptive fallback that tool can silently use elsewhere never enters the chain. No EQ shapes a routine track — the one documented exception is a small number of severely muddy tapes, where a corrective-EQ restoration pass (cutting low-mid mud, lifting presence and air) runs before normalization, opt-in per show and recorded in that show’s provenance (shown as a “corrective-eq” badge on its Technical Data table). −20 LUFS is the target, not a guarantee every track hits it exactly — a track landing a little quieter to protect its dynamics or honor the true-peak ceiling is normal and expected; the achieved value is always in the per-track table. On some audience tapes the loudest thing on the recording is a burst of applause rather than the music itself — when that happens, the track’s music still gets a plain linear boost, sized to the music alone, and only the applause is gently capped so it can’t overshoot. The same idea can apply when the loudest moment is inside the performance itself — a single sharp drum hit or strum, brief and isolated: on some shows a millisecond-scale limiter, opt-in per show and used only where the loud moments are genuinely sparse, shaves just that instant so the track still reaches the full target, while the rest of the performance keeps its full, untouched gain. Where the loud moments aren’t sparse and isolated — a track that runs loud and dense throughout — nothing is capped and the track simply lands a little quieter and honest. Either way, dynamics are never squashed to hit the target louder — a track landing a few dB quieter than its neighbors is preferred over flattening a fade or a clap-driven track pumping the room noise.

What the player streams by default. Everything above describes the archive itself, which is −20 LUFS and always will be — it is what you download, and it is the master. But −20 LUFS is deliberately quiet, and on phone speakers or in a car it is too quiet, so the player defaults to a second, louder render of every track at −14 LUFS, roughly where streaming services normalize. Every page with a player says which version you are hearing and lets you switch to Archive in one click. The loud version is streaming-only and MP3-only. Downloads default to the archival −20 master — the lossless FLAC — and since 2026-08-19 the download box also offers the louder MP3 as an alternative, clearly labelled as not lossless.

It is made from the published archive files, not from a separate edit, so it carries exactly the same fades, repairs and noise reduction — each variant track records the fingerprint of the archive file it came from, and the site refuses to build if any of them disagree. Reaching −14 needs the millisecond-scale limiter described above to work considerably harder than it does on the archive, and we are candid that this costs some dynamic range: measured across all 680 tracks the loud version gives up a median of 0.5 LU and at worst 3.1 LU of it. That was checked by ear before shipping — blind, level-matched, on the eight most-affected tracks in the archive, with controls in both directions to confirm the test could detect a difference at all — and it was not audible to us. We are still telling you it is there, because it is: if you want the recordings exactly as they were mastered, switch to Archive. The loud version is a convenience, not an improvement on the master.

The output is a FLAC master and a 320 kbps MP3 per track, each with an audio MD5 fingerprint, full embedded metadata (title, artist, show, track number, year) so downloads display properly in any player, and a true-peak check on the MP3 itself (lossy encoding can overshoot peaks).

A provenance record is written per show — source format, the exact filter chain, per-track before/after measurements, and which of the above treatments each track got — and that record is what renders as the show’s “Technical data” table, including a Treatment column you can hover for the exact reasoning.

5 · Distribution

The processed tracks fan out to three places: the streaming/download store, a Drive backup, and precomputed waveform shapes for the player. The files on the store are then re-downloaded and their MD5s checked against the provenance record — zero mismatches required.

6 · The show page

Track list with the waveform player (streaming the MP3s), lossless FLAC downloads, and the full technical data — loudness in and out, true peak, dynamic range, and gain applied for every track.