Offset, polarity and drift, settled before Demucs and DrumSep run
~/Music/Ableton Projects/4-Mic Alignment.als
7-Stem Mix Template.als
The mix session, every lane's chain preloaded at the plan's numbers
~/Music/Ableton Projects/7-Stem Mix Template.als
Alignment set. Four mic lanes plus four spares. Every mic lane carries a Utility
for the polarity decision, and its track delay is switched to samples rather than
milliseconds, so the physical offset can be nudged one sample at a time. Master is empty on
purpose. One offset and one polarity setting per microphone carries through both separation stages.
Mix set topology. Bass and Guitar are outside a DRUM BUS
group holding Kick, Snare, Toms, HiHat and Cymbals. A ninth lane runs the parallel bass amp. One
plate return, one spare return, and the master chain on the master track.
2. Rig audit: what this Mac owns
Checked against Live's own scanned plugin table, Live-files-53.db, 1737 rows.
A folder listing was not treated as evidence.
Plan calls for
Status
What goes in that slot
FabFilter Pro-Q 4
MISSING
Waves F6 for the dynamic bands, Live EQ Eight for the high-passes
FabFilter Pro-L2
MISSING
Waves L2
Helix Native SVT / B-15
MISSING
Waves Voltage Amps Bass
UAD API 2500
NOT OWNED
Waves API-2500, same unit, same Thrust control
Waves Trans-X Wide
INSTALLED
as written
Waves CLA-2A
INSTALLED
as written
Waves MaxxBass
INSTALLED
as written
Waves J37
INSTALLED
as written
Waves Reel ADT
INSTALLED
as written
Abbey Road Plates
INSTALLED
as written
UADx SSL G Bus Compressor
INSTALLED
as written
UAD Pultec EQP-1A
INSTALLED
Waves PuigTec EQP1A is loaded instead, see the note below
The one deliberate swap. UADx Pultec EQP-1A is installed and works. Its only
saved instance on this Mac is in Fiona.als, which Ableton Live 9.7.5
wrote, and Live 9 device blocks omit element ids that Live 11 requires. That format mismatch is
what made three build attempts load-fail. The master carries Waves PuigTec EQP1A,
which is the same Pultec circuit. Dragging the UADx one over it takes a second in the browser.
3. Per-lane chains and settings
Drum lanes, inside the DRUM BUS group
Lane
Chain as built
Settings from the plan
Kick
EQ Eight → F6 Mono → TransX Wide Mono
High-pass 28 Hz, 12 dB/oct. Trans-X range about +3 dB, sensitivity set on real hits.
Snare
EQ Eight → F6 Mono → TransX Wide Mono
High-pass 45 Hz. Same Trans-X treatment.
Toms
EQ Eight → F6 Mono
High-pass 45 Hz.
HiHat
EQ Eight → F6 Mono
High-pass 180 Hz. Definition from 2.5 to 4 kHz.
Cymbals
EQ Eight → F6 Mono
High-pass 180 Hz. Definition from 2.5 to 4 kHz, because 8 to 12 kHz mostly raises separation residue.
DRUM BUS
Waves API-2500 Stereo
3:1, attack 10 ms, release 100 ms, Thrust Medium, about 2 dB gain reduction.
High-pass 28 Hz. F6 rides the 79 to 83 Hz bloom down by 2 to 3 dB while note fundamentals stay. CLA-2A about 3 dB gain reduction. MaxxBass around 95 to 110 Hz, restrained.
Bass Par SVT
EQ Eight → Voltage Amps Bass Mono
Parallel midrange lane. High-pass 120 Hz, low-pass 2.5 kHz on EQ Eight band 8. Blend 10 to 15 percent.
Guitar
EQ Eight → F6 Mono → J37 Mono → Reel ADT
High-pass 90 Hz. F6 controls 220 to 350 Hz congestion dynamically. J37 at 15 ips, saturation 2 to 3, minimal wow and flutter. Reel ADT on a 10 percent send for restrained width.
Return and master
Slot
Chain as built
Settings from the plan
Return A
EQ Eight → Abbey Road Plates Stereo
Decay 0.9 to 1.2 s. High-pass the return at 250 Hz, low-pass at 4.5 kHz. One shared return for the whole kit.
Master
UADx SSL G Bus Compressor → PuigTec EQP1A → L2 Stereo
SSL at 2:1, attack 30 ms, release auto, 1 to 2 dB gain reduction. Target -14 to -15 LUFS, output ceiling -1 dBTP, under 3 dB of limiting.
Kick and bass stay centred. Every artificial stereo effect gets checked in mono,
Reel ADT included. The Renaissance family is left out on purpose.
4. Verification
Both sets open in Live 11.3.43. The mix template instantiated
21 plugin devices, every one the intended plugin, with no fallback to the default
document and no missing-plugin line. Counted from Live's own log at 16:31 on 2026-09-02:
F6 Mono ×7, TransX Wide Mono ×2, then one each of CLA-2A Mono, MaxxBass Stereo,
Voltage Amps Bass Mono, J37 Mono, Reel ADT Mono/Stereo, API-2500 Stereo, Abbey Road Plates Stereo,
UADx SSL G Bus Compressor, PuigTec EQP1A Stereo and L2 Stereo, plus the two devices left on the
spare return.
Four defects the load test caught
Duplicate device ids. Two devices in one chain shared Id="4", and Live called the whole document corrupt. Members of a device list are now numbered by position.
Live 9 device blocks. The UAD Pultec block came from a set Live 9.7.5 wrote, missing ids Live 11 requires. That source is out of the harvest entirely.
Dangling automation. Renumbering the pointee space left an envelope pointing at an id that no longer existed, logged as Invalid Pointee ID 10. A fresh template carries no automation, so the envelopes come out.
A silent identity swap. Each device block holds two class-id blocks. Replacing only the first let five lanes load Smack Attack while the file read correctly on inspection. Both blocks are now written, and the log shows the right plugin on every lane.
An offline validator now reproduces Live's rejection without a restart, learning from four
reference sets which parent-child tag pairs Live always writes with an id, then flagging any
member that lacks one. It reports clean on both sets and on Brandon's own working session.
5. What happens next
Four mic files. Drop them into the alignment set. Their path was not given, so no audio is placed yet.
Separate per microphone. Run each aligned mic through Demucs and DrumSep independently, then pick the clearest capture per lane by listening for bleed, attack, body and separator artifacts. A separator pass over the finished composite would bake every microphone compromise into every lane.
Seven stems in. Same length, same start, no normalisation, no processing.
Clip gain before compression. The large source changes are at 1:44, 2:46, 3:02, 4:18, 5:20, 6:22, 8:02 and 9:26. Handle genuine capture changes there and leave musical dynamics alone.
The source composite, measured. 10:35, mono, 24-bit, 44.1 kHz, about
-22.5 LUFS with a -2.7 dBTP peak. Roughly 72 percent
of its energy is below 315 Hz, concentrated near 79 Hz, and the spectrum falls
sharply above 4 kHz, which matches the 24 kbps AAC source ceiling. That ceiling is the reason
cymbal definition is taken from 2.5 to 4 kHz.