Mastering Checklist

A mastering checklist is the list of technical and listening checks a track passes between "the mix is done" and "the file is released" — headroom and bounce settings on the way in, tonal balance, stereo integrity, loudness and true peak on the master itself, consistency across an album, and formats and metadata on the way out. The one below is the checklist that runs inside the Master stage of every song I track, written out in full with the 2026 streaming numbers. (I make SessionFlow, the app that Master stage lives in; it gets one mention, at the end.)

Why Another One

There are roughly forty mastering checklists on the internet and, if you've read three, you've noticed they split into two camps: the ones that boil down to "put a limiter on it and hit −14" and the ones that boil down to "trust your ears, it's an art." Both are technically true and both are useless at 1 a.m. when you've heard the song two hundred times and can no longer tell whether the low end is huge or just loud. A checklist is for exactly that moment — not to replace judgment, but to make sure the boring, verifiable things got verified while your judgment was still working.

So the items below are all things you can actually confirm, with a meter, a reference track, or a phone speaker. If an item can't be checked, it isn't on the list.

Before Mastering

Most "mastering problems" are mix-prep problems that arrive wearing a disguise. Ten minutes here saves an argument later, whether the engineer is someone else or you next Tuesday.

• Peaks land around −6 to −3 dBFS on the mix bus. No clipping, no "it's fine, the limiter will catch it." True headroom, not a mix that was pulled down 10 dB at the last second.
• Nothing on the mix bus that belongs in mastering. Limiters and loudness maximizers off. Bus compression or glue that's part of the sound stays; a brickwall used to "check how loud it can get" goes.
• Bounced at the session's native sample rate, 24-bit (or 32-bit float), no dither. Dither is a one-time operation at the final bit-depth reduction, not something to stack.
• Head and tail are clean. A short pre-roll, no chopped reverb or delay tails, and whatever noise is at the top of the file is there on purpose.
• Alternate versions bounced now, not later. Instrumental, TV mix (no lead vocal), clean/radio edit, a cappella if anyone will ever ask. Bouncing them today from the same session state is trivial; recreating that state in six months is not.
• Two or three reference tracks chosen and level-matched. Same genre, releases you actually respect, gain-staged so they sit at the same integrated loudness as your mix. Comparing to a reference 6 dB louder tells you nothing except that loud sounds better.
• Mix notes written with timestamps. "The 2:14 chorus feels harsh on the vocal," not "maybe a bit bright?" Every note tied to a time, so whoever masters it can find it in one click.
• File named like it will survive a folder full of files. Artist_Title_MIX_v3_24b48k.wav beats final_FINAL2.wav every single time.

The Master

Everything in this section gets checked twice: once on the monitors you trust, then again on something you don't — earbuds, a laptop, the car, a phone speaker at arm's length. The second pass is where you find the problems that will actually reach listeners.

• Tonal balance against the references at matched loudness. Low-end weight, low-mid mud around 200–400 Hz, presence around 2–5 kHz, air above 10 kHz. You're looking for what's different, then deciding whether the difference is a choice or a mistake.
• Mono compatibility. Sum to mono and listen for anything that vanishes or hollows out — stereo-widened bass, hard-panned doubles, chorus on the vocal. Phones, Bluetooth speakers, and club systems still play a lot of mono.
• Low end is mono below roughly 100–120 Hz. Check on a correlation meter; anything wide down there costs you punch and translates badly.
• Sibilance and harshness at loud playback. Turn it up louder than you'd listen for pleasure. If the "s" sounds and cymbals get painful before the rest of the mix gets loud, that's the thing to fix.
• Quiet playback still reads. At conversation volume, can you still hear the kick, the vocal, and the hook? Loud is easy; quiet exposes balance.
• Dynamics are a decision, not a side effect. Look at the loudness range and short-term LUFS across sections. If the verses and choruses read the same number, the limiter is doing your arrangement for you.
• Integrated loudness checked on a real LUFS meter over the whole track. Not short-term, not momentary, not the "loudest part." Integrated is what the platforms measure.
• True peak at or below −1 dBTP. Sample peaks aren't enough. (−2 dBTP if the master is louder than −14 LUFS or is going to Amazon Music; see the table.)
• Limiter isn't audibly pumping or flattening transients. A/B the limited and unlimited master at matched loudness. If the only difference is "louder," good. If snares got softer, back it off.
• Start and end at digital silence, fades intentional. No click on the first sample, no reverb chopped by an abrupt end, fade-out length actually chosen.
• Full-length listen, start to finish, no skipping. On the second system. Take a note at every time you flinch. If you flinch zero times, you weren't listening.

Loudness Numbers

The numbers people argue about, as of August 2026. Two things to hold onto before reading the table: these are playback targets — what the platform turns your track to, not a law about what you deliver — and the direction of normalization matters more than the number. Spotify and Apple Music (with Sound Check on, which it isn't by default) will boost a quiet master toward the target, but only until the true-peak ceiling stops them. YouTube, Tidal, and Amazon only turn loud masters down. A −18 LUFS master plays at −18 on those three, next to everyone else's −14.

Verified August 2026 against platform documentation and current mastering-engineer references; platforms change these quietly, so re-check before a major release.

What to do with that: one master at −14 LUFS integrated, −1 dBTP lands at full level on every service without being touched. Plenty of engineers master hotter than that — −9 to −11 is normal in electronic, hip-hop, and modern rock — because the record has to compete on the platforms that don't normalize, in DJ sets, and in every playlist where the listener turned normalization off. That's a legitimate choice. The thing that isn't legitimate is believing the loud master will play louder on Spotify. It won't. It plays at −14 with the dynamics you crushed still crushed.

Album Pass

Singles skip this section. Anything with more than one track doesn't, because the album gets judged on the transitions as much as the songs, and normalization doesn't always save you — Apple's spec describes album-mode normalization, but in practice it behaves inconsistently across clients, and Spotify applies album normalization only when you play the album as an album.

• Track order decided and written down. Not "we'll figure out sequencing." Decided.
• Relative loudness between tracks set by ear, not by number. A ballad at the same integrated LUFS as the banger will feel louder. Match the experience of level across transitions.
• Tonal consistency across the record. Play the last 20 seconds of one track into the first 20 of the next. If the low end jumps or the top end dulls, that's a mastering EQ move, not a mix recall.
• Gaps and crossfades chosen per transition. Two seconds is a default, not a rule. Some songs want zero.
• Full album listened front to back, once, on one system, without a meter open. The only question is whether it holds together.

Delivery

The part that never feels like mastering and causes the most emails. (What happens to these files next is its own list — the album release checklist.)

• Streaming master: 24-bit WAV at the session's native sample rate. 44.1 or 48 kHz, or 96 kHz if that's where you worked and the distributor accepts it. Don't upsample.
• CD master, if there's a CD: 16-bit / 44.1 kHz with dither applied exactly once. Plus a DDP image if the plant wants one, with ISRCs embedded.
• Lossy check. Encode to AAC 256 and MP3 320 and listen to the intro and the loudest chorus. This is where inter-sample peaks turn into crackle.
• Metadata complete: ISRC per track, UPC for the release, title and artist spelled exactly as they'll appear, featured artists credited in the fields the distributor requires, not just in the title.
• Alternate versions mastered to match the main. Instrumentals and clean edits go through the same chain, not a copy-paste of the limiter.
• Final files named with version and format, and the mastering session archived. Keep the exact plugin chain, so a change request in three weeks is a ten-minute job.
• One last listen to the actual delivered files. Not the session — the files — on the second system. This catches the wrong bounce, the muted alt version, and the fade you forgot to re-render.

Where Masters Die

The same handful of things, over and over: comparing to references at different loudness and concluding the mix is thin; chasing a LUFS number instead of a sound and squashing the chorus to get there; widening the low end because it sounded huge in headphones; sample-peak metering that reads −0.1 and a stream that crackles; "final" bounced from a session with a muted vocal double nobody noticed; and — genuinely the most common — never listening to the actual delivered file, start to finish, on anything other than the mastering monitors. Every one of those has an item above. That's the whole reason the list exists.

Copy This

The short version, formatted to paste into a notes app, a task manager, or — since this is my page and I'm allowed one plug — the checklist on a track card in SessionFlow's Master stage, where it'll sit next to the song's waveform and the timestamped notes you took on it.

MASTERING CHECKLIST
Prep
[ ] Mix peaks -6 to -3 dBFS, no bus limiter
[ ] 24-bit, native sample rate, no dither
[ ] Clean head/tail, alt versions bounced
[ ] 2-3 references, level-matched
[ ] Mix notes with timestamps
Master
[ ] Tonal balance vs references (matched loudness)
[ ] Mono check, low end mono below ~100 Hz
[ ] Sibilance/harshness at loud volume
[ ] Still reads at quiet volume
[ ] Integrated LUFS on a real meter
[ ] True peak <= -1 dBTP (-2 if hot / Amazon)
[ ] Limiter not pumping (A/B matched)
[ ] Fades intentional, silence at start/end
[ ] Full listen on second system
Album
[ ] Order decided
[ ] Track-to-track level by ear
[ ] Transitions checked, gaps chosen
[ ] Front-to-back listen, no meters
Delivery
[ ] 24-bit WAV streaming master
[ ] 16/44.1 dithered CD master + DDP if needed
[ ] AAC/MP3 encode checked
[ ] ISRC, UPC, credits complete
[ ] Alts mastered to match
[ ] Session archived, files versioned
[ ] Delivered files listened to, start to finish

Questions

Sources

Loudness figures cross-checked August 2026 against Spotify for Artists' loudness normalization documentation, Apple's Sound Check specification, and current 2026 platform-target references from working mastering engineers (Mat Leffler-Schulman Mastering, Peak Studios, Hanna Eng). True-peak and inter-sample behaviour per ITU-R BS.1770. If a number here disagrees with a platform's own page, the platform is right and I'll fix this one — support@session-flow.app.