Mooer GE150 Rig Sheet

Two things were wrong, and the second one is the whole story.

• The Pro preset schema is structurally different. A genuine Pro preset, pulled off the pedal by dragging slot 33A into COMPUTER, is 2,440 bytes against our 958. Its effectModule keys are the numeric strings "0" through "8", not module names, and its fileInfo reads device: "GE150PRO", app: "GE150PRO Edit", schema: "GE150PRO Preset". Parameter names differ too: F.BACK rather than FEEDBACK, P.DELAY rather than PRE DELAY.
• The editor requires the filename to equal preset_name. Proven with a two-file control: ZZTESTA.mo holding preset_name "ZZTESTA" appeared in the COMPUTER list. zz_test_b_file.mo holding "ZZTESTB", byte-identical otherwise, did not. The same pair of outcomes repeated with CAST MARE.mo against 01_CAST_MARE.mo. This is why every earlier import copied the file into the library and still showed nothing.
• The COMPUTER panel is reached by right-click, not the folder icon. Its menu is Add, Delete, Refresh, Open This Directory. The folder icon only opens the library in Finder, which is where the “already exists” message came from.
• No empty slots exist. All 200 are named, so the twelve went to 48A–50D, which the existing .backup files restore.

On the pedal now: 48A TAPE HUGE, 48B TAPE DUB, 48C TAPE SLAP, 48D BUILT TAPE, 49A MODEST TAPE, 49B PAVE CROOK, 49C PAVE GROUND, 49D PAVE SLANT, 50A CAST SKY, 50B CAST SWELL, 50C CAST LOWOWL, 50D CAST MARE.

Readback check: 48A and 48B were dragged back off the pedal into COMPUTER and compared field by field against the files that were sent. TAPE HUGE returned TIME 720, F.BACK 70, LEVEL 48, DECAY 62, reverb LEVEL 36. TAPE DUB returned TIME 650, F.BACK 64, LEVEL 39, DECAY 46, reverb LEVEL 25. Both matched on every delay and reverb field. Master set: ~/Projects/_outputs/ge150/mo-pro-final/.

Twelve was one pack out of the whole ask. ChatGPT had packaged one file, GE150_Appleseed_Pavement_Tape_Presets.zip, and written everything else out as settings tables it never packaged: six Mayer presets, a Pantera rhythm patch with three album variants, and a block-by-block Appleseed Cast patch with a Low Level Owl variant. Those twelve are now built and on the pedal too.

This count was wrong when it was written. A second zip existed and this page missed it. The correction, and the twenty presets inside it, are in the card above.

Model numbers were read from the editor, not guessed. AMP: 65 US DX 0, 65 US TW 1, PV 5050 DS 36, CAROL CL 40, CAROL OD 41. CAB: US BASS 410 is 2, PV 5050 412 is 18. DS/OD: 808 is 1, PURE BOOST 2, FULL DRV 12, GOLD CLON 19. DELAY: TAPE is 3. REVERB: SPRING is 4.

A defect in the first twelve, now fixed. They inherited REVERB TYPE 3 from the reference preset, which is PLATE. ChatGPT specified SPRING for the El Capistan character. All twelve were rebuilt on TYPE 4 and re-written to the pedal. A readback of TAPE SLAP from slot 48C returns REVERB TYPE 4 and DELAY TYPE 3 with DECAY 22, LEVEL 12, P.DELAY 12, TONE 52 and TIME 118, F.BACK 13, matching what was sent.

Where ChatGPT gave no numbers, delay level and feedback keep the reference values rather than invented ones. The EQ question that held this section open is now settled, below.

Masters: ~/Projects/_outputs/ge150/mo-pro-final/ (the tape pack) and ~/Projects/_outputs/ge150/mo-pro-rest/ (Mayer, Pantera, Appleseed).

ChatGPT scored its own pack 6 out of 10 and asked for gain staging, gates, effect levels and modulation speed to be checked. Three of those are stored numbers, so they were checked against a baseline pulled off the pedal: 15 factory presets dragged out of banks 01 to 15, one per amp model, covering every one of the nine amps the twenty use.

Finding 1, the cabinet was never specified. All twenty blueprints left CAB empty, so all twenty inherited the reference preset's EL34 power tube, SM57 and mic distance 10. The factory presets assign a tube per amp and run the mic at 25 to 100. Seven had the wrong power tube. The two Jazz 120 presets were running an EL34 on an amp that is solid state and whose factory preset sets TUBE to OFF.

Finding 2, one master level for all twenty, now closed. Every blueprint wrote AMP master 65, identically. The factory presets run master 52 to 100 (median 98.5, mean 87.6, n=14) and Brandon’s other 35 presets run 78 to 100 (median 100, mean 93.2), so 65 left the new twenty quieter than everything around them on the same pedal. All twenty were re-written at master 93, the mean of the presets he already keeps, which is a loudness match against his own library rather than an invented number. Per-preset level matching still needs a guitar and an ear, so 93 is a floor to play from, not a final mix.

Finding 3, there is no guitar plugged in. Recording the pedal's own USB output with a gain-100, gate-off factory preset loaded returns -91 dBFS, which is digital silence. Nothing can be said about how the noise gates behave on a real signal until there is a real signal. The pedal does present itself as a full duplex Core Audio device at 44.1 kHz, which looked like a re-amp route. The manual closes that door. Page 18 describes the USB-C port as an OTG playback path whose level is set by the phone’s own volume, page 19 gives its output a separate OTG LEVEL control, and the CAB SIM THRU section lists “OTG audio output” beside the left, right and headphone jacks. USB audio is mixed in at the output stage the same way the AUX IN practice jack is, so it never reaches the amp and effects chain. Re-amping is impossible on this unit, and the gate check needs a guitar in the instrument jack and someone playing it.

What was corrected and re-written. All twenty were rebuilt with the factory tube, mic, centre and distance for their amp, then rebuilt again at master 93, and all twenty written to slots 40A to 44D. Readbacks of 40A VOLTA TETRA and 42B RUSH LIMELIGHT come back byte-identical to the v3 masters, both carrying MST 93.

A slot defect the re-write caught. 40A was holding a second copy of ZEP ACHILLES, duplicating 40B, and VOLTA TETRA was on no slot at all. Nineteen of the twenty had been written; the twentieth had been silently lost to a mis-targeted drag in the first pass. Reading the slot names off the device list before each write is what found it.

Masters: ~/Projects/_outputs/ge150/mo-pro-new20-v3/. Baseline: ~/Projects/_outputs/ge150/factory-baseline/ (15 factory presets).

The master-level pass is finished. Slots 40A to 44D each hold the preset they are supposed to hold, each at AMP master 93. The device list was read back after every batch, and 40A VOLTA TETRA and 42B RUSH LIMELIGHT were pulled off the pedal and compared byte for byte against the v3 masters. Both match.

Two failure modes worth writing down

The screensaver eats drags. Fourteen consecutive writes reported no import dialog and nothing changed on the pedal. The cause was the 8-minute idle timer firing mid-batch: macOS composites a full-screen secure window over everything, so each drag landed on that instead of the editor, and the Mac then sat at the password prompt. A batch of GUI writes has to hold the display awake for its whole run.

caffeinate -d -i -w $$ &   # for the lifetime of the script

Signalling the screensaver process is the wrong recovery. It leaves an orphaned secure surface that WindowServer will not give up, which black-screened this Mac once before. Input dismisses a saver; a signal does not.

Estimated row positions overwrite the wrong slots. Two writes aimed at 43D and 44B landed on 41C and 42B, quietly replacing TOE GOODBYE and RUSH LIMELIGHT. Reading row centres off a screenshot by eye was accurate to about 11 points against a 27-point row pitch, close enough to hit most of the time and close enough to miss by a whole row after a scroll. The fix is to measure rather than estimate: the selected row draws a solid rgb(152,198,24) band, so scanning one pixel column for that colour gives an exact anchor, and every other row follows from the 26.9-point pitch.

ys = [y for y in range(280,1250) if px[300,y] == (152,198,24)]
centre = (ys[0] + ys[-1]) / 4          # screenshot is 2x retina
row(n) = centre + (n - anchor_index) * 26.9

Both damaged slots were restored from the measured grid, and the device list was re-read to confirm all twenty names.

The twenty that were missing were in the most recent ChatGPT conversation, packaged as GE150PRO_20_New_Tone_Designs_2026-11-03.zip and never converted. ChatGPT had shipped them as JSON tone recipes rather than .mo files, on purpose, so as not to repeat the invalid-format problem. With the parameter dictionary extracted, converting them is mechanical.

Every model name resolved against the extracted table, with one spelling correction: ChatGPT's “65 US DLX” is the editor's 65 US DX. The rest matched on normalised names.

• paraPos is the inheritance key. A model with no parameters of its own uses the parameters of the model that defines that paraPos. TRI-CHORUS carries paraPos 12, so it takes ANA-CHORUS's RATE, MIX, TONE and DEPTH; RING MOD carries paraPos 6, so it takes TREMOLO's RATE, MIX and TONE. Neither was guessed.
• Rate was given in hertz, the pedal wants a 0 to 100 knob. Every design asked for 0.25 to 1.2 Hz, so they are mapped at ten times the hertz figure, landing between 3 and 12. Stated here so the figure stays traceable, and the hertz-to-knob curve is the one thing in this batch that is derived rather than read.
• Three parameters ChatGPT specified do not exist on the model it chose: PHASER has no MIX, RING MOD has no DEPTH, FLANGER has no DEPTH. Those are left at the model's own default rather than forced onto a neighbouring knob.
• Byte format matters. The pedal's own files are four-space indented with sorted keys and a trailing newline. The reference preset round-trips byte-identical under those settings, and the generated files now match at 2,381 to 2,420 bytes against the pedal's 2,440.

The EQ scale is resolved. The definition reads -16 to +16 with default 0. The pedal's own reference preset stores 16 in every band. Stored value is decibels plus 16, so flat is 16. Written into APC 3 LIBRAS as 14, 15, 16, 16, 16 and read back in the editor as -2, -1, 0, 0, 0. EQ is now on in all twenty.

Read back off the hardware. RUSH YYZ was dragged back off slot 42C and APC 3 LIBRAS off 44D. Both files the pedal returned are byte-identical to the files that were sent, so every parameter survived the round trip.

Masters: ~/Projects/_outputs/ge150/mo-pro-new20/ (the twenty) and ~/Projects/_outputs/ge150/mo-pro-cap/ (the eleven El Capistan modes, now at 37B–39D).

The editor carries its own parameter dictionary. The model names are not plain strings anywhere in the 32 MB binary, in any encoding, because they sit in a zlib-compressed Qt resource. Decompressing the stream at offset 13,722,184 yields 21,425 bytes of XML that defines every chain, every model, and every parameter's range.

Validation first. Nine model indices had already been read off the editor by eye. The extracted table agrees with all nine: 65 US TW 1, PV 5050 DS 36, CAROL CL 40, 808 1, GOLD CLON 19, TAPE 3, SPRING 4, PV 5050 412 18, ANA-CHORUS 11.

• 210 models across 9 chains, with 139 parameter definitions: 75 amps, 46 cabs, 28 drives, 25 mods, 12 comps, 9 delays, 8 reverbs, 4 EQs, 3 gates.
• The SUB-D enum, which had been a guess: OFF, 1/4, 1/4T, 1/4D, 1/8, 1/8T, 1/8D, 1/16, 1/16T, 1/16D, indexed 0 through 9.
• Ranges: delay LEVEL and F.BACK are 0 to 100, TIME is 40 to 2500 ms, reverb P.DELAY is 0 to 500, EQ bands run -16 to +16 dB with step 1 and default 0.
• DUAL DELAY does not have a TIME field. It uses TIME A, SUB A, TIME B and SUB B.

It immediately caught six wrong files. Three El Capistan Fixed presets had no subdivision set, and the three Multi presets were written with a TIME field that Dual Delay does not accept. All six were rebuilt against the extracted definitions and re-verified in the COMPUTER list.

Reference: ~/Projects/_outputs/ge150/GE150PRO-model-definitions.xml and GE150PRO-model-index.json. This is the app's own declarative description of its interface, which is a different thing from reconstructing an audio algorithm. For the USB wire protocol or the pedal firmware, static reading of this kind gets you nothing and a real disassembler would be the tool.

A separate session ran a ten-minute nudge loop against the ChatGPT conversation that authored the tone blueprints. Its final pass reported four model labels as wrong against Mooer's printing: 65 US DX should read “65 US DLX”, US BASS 410 should read “US Bass 410”, FULL DRV should read “Full Drv”, and “Yellow Comp” should read “Yellow comp”.

All four proposed corrections are wrong, measured against the editor's own model definitions. Counted in the decompressed Qt resource, each form the editor prints appears exactly once and each form the audit proposed appears zero times.

The audit's two columns were reversed. The first three labels already matched the editor, so acting on it would have replaced three correct labels with the blueprint's own spellings, which is where they came from: the 20-preset blueprint on disk contains “65 US DLX” four times, “Full Drv” four times and “Yellow comp” twenty times, and YELLOW COMP never.

The fourth finding was real, and its proposed fix was wrong too. This page had written “Yellow Comp” in two preset rows, matching neither the editor's casing nor the audit's. Both rows now read YELLOW COMP.

Every model number on this page was re-checked in the same pass. Thirteen claims, each matched to the extracted table by converting its one-based display number to the zero-based index the .mo files store: 65 US DX 0, 65 US TW 1, PV 5050 DS 36, CAROL CL 40, CAROL OD 41, US BASS 410 2, PV 5050 412 18, 808 1, PURE BOOST 2, FULL DRV 12, GOLD CLON 19, TAPE 3, SPRING 4. Thirteen of thirteen agree.

The blueprint pack probably stopped changing at its ninth pass, and that is a self-report, not a measurement. The claim that all 55 designs went byte-for-byte unchanged at V9 and stayed identical through V14 comes from the generator asserting it about its own output. Nothing was diffed. The packs exist only as attachments in the conversation, and a search of ~/Downloads and ~/Projects/_outputs/ge150 for any V9 to V14 artifact, changelog, slot audit or parameter dump returns nothing, so there is no local file to check it against.

What holds without trusting the generator is weaker and still useful: from the eleventh pass on, the defects it reported stopped naming the pack and started naming its own earlier paperwork, each version bump leaving the previous handoff stale. That is a structural tell rather than a claim about bytes. Treat V9 as the last version where a tone plausibly changed, with “plausibly” carrying real weight. Settling it needs V9 and V14 downloaded and diffed by someone with the thread open. Loop data: ~/Projects/_outputs/chatgpt-nudge-loop/LOOP-DATA-2026-11-04.md.

Reading all twenty back rather than a sample turned up one disagreement. ZEP KASHMIR came off the pedal differing from its master in exactly one field: the MOD block's PITCH, device 0 against master 0.1. Everything else in the file was identical.

The master was the wrong one. The converter contained a special case that inserted PITCH = 0.1 whenever a DETUNE block arrived without a pitch, carrying the comment “smallest audible detune; unit unverified”. The blueprint for ZEP KASHMIR specifies rate_hz 0, depth 9, mix 16 and no pitch at all, so 0.1 was invented rather than derived.

It never reached the pedal either. Every other value in a .mo is an integer, the editor coerced the float to 0, and the hardware has held 0 all along. The extracted table defines DETUNE's pitch as -20_20_0.1_1_, a −20 to +20 range in steps of 0.1, which the file almost certainly stores as an integer count of those steps.

Fixed at the source. The special case is gone from ~/Projects/_outputs/ge150/convert20.py, which now leaves the parameter at its range default and reports it as unspecified instead of filling it. The master was set to 0 to agree with the hardware, so no further write to the pedal was needed, and all twenty now match.

What this left open. ZEP KASHMIR's Detune block was switched on with a pitch of 0, and the blueprint never said how much detune it wanted. Its depth 9 has no home in a DETUNE block, whose only parameters are PITCH, MIX and TONE, and reading depth as the detune amount would be a guess.

Resolved by switching the block off, 2026-11-04. Enabled at zero is not the neutral state it looks like: the block still runs at MIX 16, and a time-domain pitch shifter passing a 16 percent wet copy through its buffer comb-filters against the dry signal whether or not the pitch offset is zero. That is the expected behaviour of the algorithm class rather than something measured on this unit. Switching MOD off removes the colouring without substituting another invented number.

One field changed, effectModule 6 SWITCH from 1 to 0. Every other block is untouched, written to 40C and read back matching at bcdae2e84fe8 with a fresh mtime. A SWITCH check across all twenty presets now reports 179 of 180 blocks agreeing with their blueprints and exactly one divergence, this one, so the deliberate change is visible to a test rather than hidden inside a passing total.

Where the recommendation came from, and what it rests on. The call to disable rather than guess came through Brandon from a ChatGPT pass that also argued the part should lean on the Danelectro in DADGAD and the amp rather than modulation, and noted that the phasing often cited on that recording sits on the drums. Those are its historical claims and they were not checked here. The reason the block is off is the one above: no source gives a detune amount, and zero with the block live is not reliably neutral.

Saying twenty of twenty match is a report. These are the numbers behind it, so anyone holding the masters can check it instead of taking the report's word. Each row is one slot: the sha256 of the master in ~/Projects/_outputs/ge150/mo-pro-new20-v4/, and whether the copy read back off the pedal has the same digest.

Reproduce with shasum -a 256 ~/Projects/_outputs/ge150/mo-pro-new20-v4/*.mo.

What these twenty rows do not cover. They are the twenty presets in 40A to 44D. The other 35 on the pedal were written in earlier passes and have never been read back, so there is no hash evidence for them and none is claimed. And a matching digest says the pedal holds the intended data; it says nothing about whether the data sounds right, which needs a guitar in the instrument jack.