Buzz: full test sweep (tests T1 to T33)

Rendered from the working notes 05-full-sweep.md on 2026-10-11.

Full sweep: buzz on channels 2 and 20, TRC, 2026-09-25

Written 2026-09-25 late evening from 03-diagnosis.md, 04-new-data-greg-2026-09-25-2300.md, 01-session-reconstruction.md (sections 1 to 5) and 02-system-model.md (sections 0 to 6), plus a direct read of the Radial Catapult manual (/Users/gregspero/trc-studio-manuals/radial-engineering-catapult-rx4__24.pdf, pages 3 to 4).

Revision 2 folds in 06-gear-research.md (Radial pinout, lift behaviour, GX phantom sensing, Trident grounding, Whitlock bridge arithmetic) and 07-history-mining.md (iMessage and lifelog history back to 2024). Seven hypotheses were added (A15 to A19, B12, D6), and every score and the ranking were redone.

Source keys: [01] 01-session-reconstruction.md, [02] 02-system-model.md, [03] 03-diagnosis.md, [04] 04-new-data-greg-2026-09-25-2300.md, [06] 06-gear-research.md, [07] 07-history-mining.md, [CAT] the Catapult manual. Test IDs (P#, T#) are the ones defined in [01] section 1. chat.db row numbers are cited as [07] lists them.

Labels: [F] fact from a source, [R] reported speech, [I] my inference. Confidence: high / moderate / low / unknown.

0. The bottom line

• Best conclusion (moderate confidence): the buzz is a mains-derived voltage difference between the mix-room ground and the mic-closet ground. It rides the ch 2 and ch 20 signal pairs between the two rooms and is converted into audible buzz by something that differs on those two channels. Studio lighting is a proven contributor to that voltage, but not the only one. The history shows the same shape since the first week of the October 2024 install, when channels 9 and 10 buzzed ([07] chat.db 71841).
• The split is built the way Radial says not to split between two consoles. It is a passive Y with no transformers. Radial specifies TX4M/RX4M transformer modules for exactly this job ([06] 1.5 to 1.7).
• The per-channel converter is not located. Two candidates lead. The first is Trident +48 V on for 2 and 20 and off for 1 and 16, which takes 30 seconds to test. The second is a fault in the hand-wired, screw-terminal terminations Thai made in October 2024 ([07] chat.db 1321). A DMM leg-to-pin-1 check on ch 2 and 20 tests that in 5 minutes. The bad channels have moved over time, from 9 and 10 in 2024 to 2, 20 and 22 in 2026. That favours a setting or a hand termination over a fixed fault in a factory-built box.
• A faulty GX4816 is the weakest converter candidate. It needs two independent faults. The ch 20 residue and the QSC test P15 point upstream of the GX, and a fixed box fault does not move between channels.
• The lighting side is worse than earlier files assumed. The stage distro shares the audio panel by the contractor's choice. It runs on two of three phases, and its panel had no neutral until November 2025. A lighting power transformer exists, and nobody has recorded how its secondary is grounded ([07] items 3, 4, 6, 7). The lighting PoE switch has also carried the SQ-7, the Hear Back hub and the NAS ([07] item 8).
• Fix direction regardless of which converter wins: RX4M transformer modules on the mic-closet leg end the loop for every channel. An electrician should also bond the IG system and the lighting transformer's secondary to the service ground.

1. The 12 key observations

Supplemental evidence used as tie-breakers (outside the 12):

2. Scoring method

• Each hypothesis gets F (fits), N (neutral), X (contradicts) against O1 to O12.
• Group A hypotheses are drivers (they make the noise voltage). A driver alone predicts every channel buzzes equally, so O8 is marked P (needs a per-channel partner from groups B or C) and counts zero.
• Score: F = +1, N = 0, P = 0, X = -2. A contradiction weighs double because one solid contradiction kills a hypothesis while one fit only fails to kill it.
• Tie-break adds or subtracts 1 for each supplemental item S1 to S13 that bears on it. A converter that needs two independent hardware faults takes -2. A fault that moves between channels over time (S9) costs a fixed factory-box fault -1 and credits a setting or a hand termination +1.
• Limit of this method [I]: O1 to O7 are facts about the loop's topology. Every driver and every converter that sits on the ch 2 pair fits them. The existing tests therefore cannot separate most converters. The decision tree in section 6 exists to do that.

3. The fit matrix (48 hypotheses)

4. Each hypothesis: mechanism, key fit notes, cheapest test, information needed

A. Ground-potential drivers

A1. Open isolated-ground (IG) system. - Mechanism: the IG greens reach no electrode, so each rack's chassis settles at whatever its leakage paths push it to. The two racks differ by some mains-frequency voltage, and the ch 2 pair bridges them. [F] open ground on every IG outlet tested, [02] §3b. [I] the bridge. - Fit notes: O3 fits because the GX chassis stays tied to the amps and IEM transmitters through its aux outputs even with its own cord out ([02] §3a). O12 fits because floating grounds pick up load-dependent leakage. - Cheapest test: DMM, AC volts from GX chassis to Trident chassis, then the 12 AWG temporary bond (tree steps 3 and 4). - Needs: outlet and circuit feeding each rack; whether each is an orange IG outlet.

A2. Chassis floating through EMI filter Y-capacitors. - Mechanism: on an open-ground outlet, every mains filter's line-to-chassis and neutral-to-chassis capacitors pull the chassis toward roughly half mains voltage through a high impedance. Capacitors pass harmonics more easily than 60 Hz, so the result sounds like buzz rather than hum. [I] - Fit notes: strongest fit to O10. A high-impedance node is moved by a nearby body. Fits O9 because every aux load reconnects the GX chassis to some other device's leakage. The 30 to 60 V reading on an open ground is typical of this mechanism, and it is also typical of a DMM "ghost" reading (see section 5). - Cheapest test: chassis-to-chassis AC volts read twice, once with the DMM alone and once with a 10 kΩ resistor across the leads (or the meter's LoZ mode). A reading that collapses under the resistor is a high-impedance float (A2). A reading that holds is a low-impedance drive (A10, A11, A13). - Needs: the two readings.

A3. ISP entrance ground (the "modem"). - Mechanism: cable TV coax is normally bonded to the service ground at its entry point, and Whitlock documents large noise between the CATV ground and the power safety ground ([06] 5.6, 5.7). If that shield is the only conductor still tied to a real electrode, anything networked to the modem is anchored to it while the mic closet floats. [I] - Fit notes: the internet is Frontier fiber since March 2025, and the router stack is in the office, fed by "existing blue Cat5e from electrical to office" ([07] item 11). Fiber carries no conductive shield. The mechanism survives only if a cable coax service is still connected, or through that copper Cat5e run from the electrical room. Lower prior than revision 1. - Cheapest test: with lights off and the residual audible, disconnect the ISP device's network cable (and any coax) for 60 seconds. - Needs: what "the modem" is (fiber ONT or cable modem) and whether any coax enters the building.

A4. Lorex PoE cameras and NVR. - Mechanism: the NVR's power supply leaks into its chassis, and camera housings touch building metal. Cable shields and PoE switch grounds can tie those into the data network. [I] - Fit notes: a residual candidate for O12. One Lorex camera is on Ethernet in the Mixing Room near the sub, the NVR is in the office, and the camera relays sit in the electrical closet ([07] item 10). The first camera install took a breaker from the HVAC ([07] item 16). That gives a physical path into the mix room, though it bonds to the rack only if the camera shares an outlet, mount or metal with it. Shielded camera cable with both ends grounded forms a loop; unshielded rarely does ([06] 5.8). - Cheapest test: unplug the Mix Room camera's Ethernet, then the NVR and its PoE switch, during the lights-off state. - Needs: whether the Mix Room camera cable is shielded and what it is mounted to.

A5. Network switch tying the SQ-7 network port to modem ground. - Mechanism: SQ-7 Dante port to a switch to the modem, so the SQ-7 chassis follows the modem's ground. [I] - Fit notes: contradicts O4. With the SLink out, the SQ-7 has no connection to the GX at all, and the buzz stayed ([01] T13, T25). It could still shift the mix-room side if the SQ-7 shares that side, which is undocumented. - Cheapest test: pull the SQ-7 network cable (tree step 9). - Needs: whether the SQ-7 Dante port is cabled to the house network.

A6. Lighting LED drivers and dimmers leaking into the ground system. - Mechanism: LED drivers and phase-cut dimmers are switch-mode loads with filter capacitors to ground. On an open-ground system that leakage current has nowhere good to go and raises the floating grounds, with a harmonic-rich waveform. [I] - Fit notes: fits O12 as one contributor. As the only cause it would contradict O12, because lights off should then kill the buzz. RodCo's quoted isolation transformer was never billed ([02] §3b), but a "power transformer for the lights" does exist in the Live Room ([07] item 6); see A19. - Cheapest test: the lights-off test already done, repeated with a number (tree step 0 plus step 3 readings with lights on and off). - Needs: whether stage lighting is LED on DMX or dimmer packs; which panel feeds it.

A7. HVAC. - Mechanism: compressor or blower motor currents on a shared neutral or leaking to ground. [I] - Fit notes: unknown whether HVAC ran during the lights-off test. - Cheapest test: cycle HVAC off at the thermostat with lights off, listen. - Needs: whether HVAC was running at 23:00; its panel.

A8. Subs on separate wall outlets. - Mechanism: GX output 14 feeds the subs, which sit on their own wall outlets ([02] §3a). The XLR shield ties the GX chassis to those outlets' grounds. [I] - Fit notes: fits O3 and O9. T9 unplugged outputs one at a time, never all at once. - Cheapest test: unplug all 16 GX output XLRs at once (tree step 7). - Needs: which outlets and whether they are IG or standard.

A9. Furman-fed Crown amp and PSM900 transmitters. - Mechanism: GX outputs 1 to 8 feed four PSM900s and main L/R feeds a Crown amp, both on the Furman M-8x2 at stage right beside the lighting "power distributor" ([02] §3a, §4). Their grounds tie the GX chassis to the Furman's outlet. [I] - Fit notes: same as A8. - Cheapest test: same as A8. - Needs: the wall outlet feeding the Furman.

A10. Missing-neutral or bootleg-ground wiring from Farzad's rewiring. - Mechanism: a circuit that lacks a neutral is sometimes finished by returning load current on the ground wire. Load current on the grounding system produces a voltage proportional to that load. [F] Farzad rewired circuits "lacking neutral wire" and the lobby speakers buzzed after ([02] §3b). [I] the mechanism. - Fit notes: a load-modulated driver fits O12 well. - Cheapest test: N-G AC volts at the rack outlets with lights on, then off (tree step 3). A load-dependent change points to current on the ground or neutral. - Needs: which circuits Farzad's crew touched.

A11. Neutral-to-ground bond in a downstream subpanel. - Mechanism: if the "Outdoor / Stage Lights" subpanel has its neutral bonded to its ground, stage-lighting return current divides between the neutral and every ground path, including audio shields. [R] a grounding conductor was landed in that subpanel by Farzad's electrician ([02] §3b). [I] that a bond exists there. - Fit notes: best single mechanism for "lights on, more buzz". Lights off leaves the other loads on that panel. - Cheapest test: electrician opens the subpanel and looks for a bonding screw or strap. Studio-side proxy: N-G volts at the rack outlets, lights on vs off. - Needs: a photo inside that subpanel (electrician only; live panel).

A12. 3-phase distro running on two of three phases. - Mechanism: the stage distro "requires 3 phase power" but "Only 2 of the phases were connected in" ([07] item 4, chat.db 82216, 82106). On a 208Y/120 V feed, two phases 120 degrees apart do not cancel in the neutral: the neutral carries roughly a full line current plus every third-harmonic current from the LED fixtures, which add. On a 120/240 V split-phase feed, only the imbalance between the two legs flows, and triplen harmonics cancel when balanced. Either way the neutral voltage rises with lighting load. [I] - Fit notes: fits O12. RodCo described the service as a 240/120 V side with 2-pole breakers and a separate 3-pole side "only 240V. No 120V. Which would be where the transformer comes in" ([07] chat.db 82231). [I] If the 3-pole side is 240 V delta, the lighting transformer (A19) may be what derives 208Y/120 V for the distro. Service configuration remains unresolved ([02] §3b). - Cheapest test: clamp-meter current on the distro neutral with lights at full (electrician, or a clamp meter if the studio has one). Studio-side proxy: N-G at the rack outlets, lights on vs off. - Needs: photo of the main gear and of the breaker feeding the distro (2-pole or 3-pole).

A13. Stage-lighting load raises the shared panel's neutral. - Mechanism: audio and "the big lights" are reported on the same panel ([02] §3b). Lighting current across the feeder neutral's resistance lifts every neutral on that panel relative to ground. [I] - Fit notes: fits O12. Overlaps A11; they differ in whether current flows on the ground conductor. - Cheapest test: same N-G reading as A10. - Needs: panel schedule for the rack circuits.

A14. Trident Star Ground Link. - Mechanism: the link joins or separates the console chassis and audio ground; Trident advises against lifting it except to fight loops ([06] 4.3). Trident's own scheme connects mic screens at both ends ([06] 4.4, 4.5), which is exactly the Catapult pin-1 path to the GX. Either link state sets where the mix-room end floats. [F] its state at TRC is unrecorded ([02] §3a). - Fit notes: a setter of the reference, not a cause of the lights dependence. - Cheapest test: photo of the PSU rear panel link. - Needs: that photo.

A15. SQ-7 network port on the lighting / camera PoE switch. - Mechanism: in May 2025 the SQ-7 at FOH was on the same PoE switch as the PTZ camera, lighting control, Hear Back and the NAS ([07] item 8, chat.db 810505, 17776). A shielded network cable bonds the SQ-7 chassis to that switch and everything else on it. [I] - Fit notes: contradicted by O4 as the path to the GX. With the SLink out, the SQ-7 has no connection to the GX, and the buzz stayed. It remains a likely source of other buzz in the SQ-7 system (the October 2025 livestream buzz Alex blamed on the lights, [02] §5). - Cheapest test: pull the SQ-7 network cable (tree step 9). - Needs: whether the SQ-7 is still on that switch.

A16. GX or other mic-closet gear on the lighting / camera PoE switch. - Mechanism: a network drop into the mic closet was proposed in January 2026 ([07] chat.db 874882). If a GX expansion port, a Hear Back unit or any networked device in the mic-closet rack reaches that switch, the mic-closet ground follows the lighting system's ground regardless of the SLink. [I] - Fit notes: fits every observation, including O4 and O12. No record says the drop exists. - Cheapest test: look for any Ethernet cable into the mic-closet rack other than the SLink and the Catapult link; unplug it. - Needs: a photo of every network cable in the mic-closet rack.

A17. Hear Back hub and NAS on the lighting PoE switch. - Mechanism: Eric Lloyd got "the lights, hear system and networked storage working on the PoE switcher" ([07] chat.db 17776). The Hear Back hub takes audio from the mix-room rig, so a shielded Cat run can bond the mix-room ground to the lighting network's ground, which the lighting load moves. [I] - Fit notes: fits O12 directly and survives O4, because it acts on the mix-room end, which the SLink tests never touched. - Cheapest test: during the buzz, unplug the Hear Back hub's and the NAS's network cables for 60 seconds each. - Needs: whether the switch is still shared; whether its cables are shielded.

A18. Console-to-PTZ-camera audio feed. - Mechanism: Vince asked for a "stereo (LR) board feed from the audio mixing board (1/4 inch) to the back of the ptz camera (1/8 inch)" ([07] chat.db 810510). An unbalanced cable from the console into a PoE camera on the lighting switch is a direct ground bridge. [I] - Fit notes: if the feed comes from the SQ-7, it is contradicted by O4 like A15. If it comes from the Trident or the mix-room rig, it acts on the mix-room end and would fit fully. Whether it exists is unknown. - Cheapest test: look at the back of the PTZ camera for an audio cable; unplug it. - Needs: a photo of the PTZ camera's rear panel and where any audio cable goes.

A19. Lighting power transformer secondary reference. - Mechanism: Eric Lloyd, October 2025: "There's the power transformer for the lights and that big rack mount case" ([07] item 6). If it is an isolation or delta-to-wye transformer, it creates a new neutral that must be bonded to ground at the transformer and tied to the building electrode. Unbonded, the lighting neutral floats and every fixture's leakage finds its own path to ground. Bonded to the wrong ground (for example the IG bus or the stage-lights panel's grounding conductor), lighting return and leakage currents run on the audio grounds. [I] - Fit notes: fits O12. It contradicts the no-transformer inference in [02] §3b. Type, primary panel and secondary bonding are all unrecorded ([07] open thread 1). - Cheapest test: photo of the transformer nameplate. The electrician checks for a bonding jumper at the secondary. - Needs: that photo, and the electrician's check.

B. Per-channel conversion points

B1. GX input 2 / 20 passive network asymmetry. - Mechanism: each GX input leg reaches the chassis through a phantom resistor, an RF capacitor and clamp diodes, even when the unit is off. A drifted or damaged part on one leg converts common-mode noise into differential buzz ([03] §B). - Fit notes: fits all 12. Weighed down by S1: the QSC in place of the GX also buzzed (if that was ch 2). Weighed down by S3: the ch 20 residue with its XLR pulled at either end shows noise reaching ch 20 without GX input 20. [I] inputs 2 and 20 are likely on different input boards, so this needs two independent faults. The bad channels have moved over time (9 and 10 in 2024, [07] item 1), which a fixed GX fault cannot do. Whitlock's arithmetic confirms that a 1% leg mismatch is enough to be audible ([06] 5.3). - Direct check added from [06]: with the GX off and nothing plugged in, compare DMM ohms pin 2 to pin 1 against pin 3 to pin 1 on GX inputs 1, 2, 16 and 20. Legs that differ by more than about 1% point to the part. - Cheapest test: the swap, Catapult out 2 into GX input 1 (tree step 6). - Needs: the swap result.

B2. Mic-closet Catapult position 2 / 20. - Mechanism: a cold solder joint, worn XLR contact, or corrosion on one leg of that position gives one leg more resistance than the other. [I] - Fit notes: fits all 12 and S1. S5 (March: bypassing the Catapult cleared ch 2) fits. [I] with 12-channel units, 2 and 20 are in different units, so this also needs two faults unless the cause is shared, such as a hot-patch surge. - Cheapest test: move the mic-closet Cat6 to a different bank jack and patch the shifted XLR into GX input 2 (tree step 6c). - Needs: the unit model, and which unit and bank carry 2 and 20.

B3. Mix-room Catapult position 2 / 20. - Mechanism: same as B2, in the split or receive units behind the Trident. "Clean alone" does not clear it, because an imbalance only shows once a second ground is attached ([03] §B). - Fit notes: fits all 12. - Cheapest test: if tree step 6c leaves the buzz on, the mix-room side holds the fault. - Needs: same as B2.

B4. Trident input 2 / 20 imbalance. - Mechanism: the Trident's own input network on strips 2 and 20 (phantom resistors, input coupling) is mismatched, so common-mode current on the pair converts at the Trident end. [I] - Fit notes: fits all 12. S1 fits (the Trident was connected when the mic-closet QSC buzzed). S2 would favour it if the Cat6 and GX were connected during P14, which is not recorded. - Cheapest test: substitute a QSC for the Trident on ch 2 in the mix room, with the Cat6 and GX input 2 connected (tree step 6a). - Needs: the P14 conditions, or the redo.

B5. XLR patch tail at the GX: pin-1-to-shell short or pin swap. - Mechanism: a shell short ties chassis to pin 1 at an odd point; a pin swap puts pin 1 on a signal leg. [I] - Fit notes: contradicted by O5; a different mic cable still buzzed ([01] T8). Ch 20's patch was not swapped. - Cheapest test: continuity-check the ch 20 patch pin by pin and pin to shell with the DMM. - Needs: nothing beyond the reading.

B6. DPA multipin fault on 2 / 20. - Mechanism: a pin fault on the Trident-side multipin produces the same one-leg imbalance. [I] - Fit notes: two DPA snakes tested clean ([01] P3), but only in the mix room alone. That test cannot show an imbalance that needs a second ground. - Cheapest test: tree step 6a clears or implicates everything on the Trident side at once. - Needs: DPA connector type.

B7. Wall-panel tie-line leg leaking to the panel box or conduit. - Mechanism: damaged insulation lets one leg of tie line 2 touch the metal panel box or conduit, a third ground reference on one leg only. [I] - Fit notes: fits all 12. Low prior because the buzz is identical with no mic in the panel ([01] P8), though the tie line itself stayed connected. - Cheapest test: unplug tie line 2 from the mix-room Catapult input (tree step 5). Buzz unchanged clears the panel. - Needs: that result.

B8. Cat6 pair assignment (looser-twist pair). - Mechanism: pairs in one Cat6 cable have different twist rates, and channels on the loosest pair pick up more magnetic field. A new cable of the same type keeps the same assignment, so O6 fits. [I] - Fit notes: contradicted by O8. Radial's pinout puts bank position 2 on the blue pair 4-5 and position 4 on pins 1-2 ([06] 1.1, 1.2). With contiguous numbering, ch 20 and clean ch 16 are both position 4, so they share a pair and differ in behaviour ([06] summary). No standard fixes twist rate by colour, so "blue and brown are loosest" is false as a general premise ([06] 2.1, 2.2). Twist also explains magnetic pickup, not a buzz that needs a second grounded device ([06] 2.4). One residue: ch 2 and ch 22 are both position 2 (blue), worth one cable-orientation check. - Cheapest test: moving the bank (tree step 6c) also moves the pair. - Needs: the channel-to-bank map to confirm contiguous numbering.

B9. RJ45 jack contact damage inside a Catapult. - Mechanism: a bent or oxidised jack contact passes a cable tester (continuity only) while adding resistance on one leg. [I] The second tester's "all green" ([01] T18) checks continuity, not resistance. - Fit notes: part of B2/B3 physically; listed separately because it survives both a new XLR and a new Cat6. - Cheapest test: tree step 6c. - Needs: nothing extra.

B10. Split pair in field termination. - Mechanism: a floor box, keystone or hand crimp puts the two legs of a channel on two different pairs, so the pair stops rejecting pickup. [I] - Fit notes: contradicted by O6 if the silver cable ran Catapult to Catapult with factory ends, which the transcript implies ([01] T29). - Cheapest test: confirm whether the silver cable was hand-crimped. - Needs: that answer.

B11. Pin-map mismatch or internal miswire between Catapult units. - Mechanism: two different Catapult models, or one miswired unit, route a channel's legs across two pairs. Survives any cable change. [I] - Fit notes: fits the observations, but Radial states every Catapult module uses one pin map and interoperates ([06] 1.2). A miswire would be a unit defect, and it would make signal leak between channel numbers. The March description of channels "crossing" ([07] MAR) is the only hint of that. - Cheapest test: talk or tone into panel input 2 and listen at every mic-closet output of that bank for bleed. - Needs: model labels on every unit at both ends.

B12. Hand-wired screw-terminal terminations on 2 / 20. - Mechanism: Thai hand-wired the October 2024 install: "Snip, strip to bare wire, insert into connector and screw down" ([07] chat.db 1321). The split was re-wired again weeks later to bypass the MEU system ([07] chat.db 70384). A stray shield strand touching one leg, a loose screw, or insulation nicked under a terminal gives one leg a leakage path to pin 1 or a higher series resistance. That is a per-channel imbalance, in whichever channels were terminated badly. [I] - Fit notes: fits all 12. It explains channels that differ by accident rather than by bank or pair position. It explains faults that move and appear over time (9 and 10 in 2024; 2, then 20 and 22, in 2026; S9, S10), since screw terminals loosen and strands migrate. It fits S1 and S3 if the terminations sit on the Catapult or Trident side. Where exactly the screw terminals are is not recorded: the Catapult Rack itself has XLRs and RJ45s, so they are most likely on the Trident-side multipin, the tie lines, or the monitor split. - Cheapest test: with ch 2 unplugged at both ends of each hand-wired segment, DMM ohms from pin 2 to pin 1 and pin 3 to pin 1 (should read open), and pin 2 to pin 2 against pin 3 to pin 3 end to end (should match within a fraction of an ohm). Then open the terminal block for 2 and 20 and look. - Needs: where Thai's screw-terminal connectors are; a photo of them.

C. DC and phantom paths

C1. Trident phantom on for 2 and 20, off for 1 and 16. - Mechanism: with phantom on, each leg of the Trident input connects to the phantom supply through 6.8 kΩ. The common-mode impedance at the Trident end drops from high to about 3.4 kΩ against a stiff rail. Far more ground-difference current then flows through the pair into the GX input network, and ordinary resistor mismatch converts it into buzz. Phantom current also has to return. [F] the Catapult passes phantom only over the cable shield, and the ground lift "will disconnect phantom power" ([CAT] pp. 3-4). [I] if the shield is absent or lifted, the only return for phantom current into the GX input is through the building grounds. - Fit notes: explains two bad channels (three in March) with one setting instead of two separate faults. Explains the August "loud buzz when phantom power is engaged" (S4). Explains why the GX being off changes nothing, because the Trident supplies the phantom. [F] the GX +48V / PP indicator "detects voltage at the socket whether supplied by the GX4816 or received from an external source", triggered at 24 V ([06] 3.1). Some input LEDs in rows 1 and 2 looked lit ([03] §C1). Because the Catapult is a Y, whichever console has phantom on powers both consoles' input pins ([06] 3.3). Magnitude check: a 1% phantom-resistor mismatch with 1 V between grounds gives about -83 dBu at the input, 50 dB above the preamp noise floor after 50 dB of gain ([06] 5.3). With a floating Y-cap chassis instead, about 3.4 mV differential before loading ([06] 6.5). Both are audible. Staff already link the buzz to phantom: Elijah is buying Triton FetHeads for the Coles ribbons ([07] chat.db 942186), which would not break a split ground loop. A setting also explains why the bad channels moved between 2024 and 2026 (S9). - Cheapest test: switch Trident phantom off on 2 and 20, wait 30 seconds, listen (tree step 1). - Needs: photo of the Trident phantom switches on 1, 2, 16 and 20; the T26 result.

C2. SQ-7 / GX phantom on for 2 and 20. - Mechanism: the SQ-7 turns on GX phantom for those inputs, pushing 48 V back into the Trident input through the split. [I] - Fit notes: contradicted by O3. With the GX off there is no GX phantom, and the buzz stayed. It can still add to the buzz when the GX is on. - Cheapest test: SQ-7 phantom page, all off. - Needs: SQ-7 phantom state for 2 and 20.

C3. Trident phantom back-feeding an unpowered GX input through its clamp diodes. - Mechanism: with the GX off, its supply rails sit near 0 V. Trident phantom arriving on the input can forward-bias the input protection diodes, which conduct unevenly and rectify. [I] - Fit notes: specifically explains why powering the GX off never helped. Does not explain the same buzz with the GX on unless C1 also operates. - Cheapest test: included in tree step 1 (phantom off). - Needs: same as C1.

C4. Trident phantom-rail ripple. - Mechanism: the phantom rail's ripple appears on both legs through the 6.8 kΩ resistors. It cancels into a high-impedance load, but the GX input network draws current and mismatch lets it through. [I] - Fit notes: would not depend on lights, so it cannot explain O12 alone. - Cheapest test: tree step 1. If the buzz goes with phantom off, repeat on a different Trident strip with phantom on to separate C1 (ground current) from C4 (rail ripple). - Needs: nothing extra.

D. Electromagnetic pickup

D1. Magnetic pickup on the Cat6 route. - Mechanism: lighting feeder current near the run induces voltage into the pair. The far-end load closes the loop, so pulling the XLR stops the current. [I] - Fit notes: contradicted by O10; a body near the rack barely changes a magnetic field. The new cable probably took a similar route. - Cheapest test: the battery recorder test (tree step 2). Magnetic pickup survives a floating far end; a ground loop does not. - Needs: the Cat6 route.

D2. Electric-field pickup from lighting cabling. - Mechanism: capacitive coupling from lighting wiring into an imbalanced pair. [I] - Fit notes: contradicted by O2. Electric-field pickup gets worse when the far end is unplugged and floating, and here unplugging kills the buzz. - Cheapest test: none needed beyond step 2. - Needs: nothing.

D3. Cat6 unshielded, or shield not bonded at both ends. - Mechanism: without a shield there is no pin-1 bond between the rooms. The whole ground difference then sits across the pair's common mode, and the ground-lift switch has nothing to lift. [F] the Catapult carries pin 1 only on the shield ([CAT] p. 3). [R] Greg: "these are shielded", unverified ([01] §5). - Fit notes: an enabler, not a cause. [F] Radial: the lift "disconnects the shield at the RJ45 connector" and never touches pins 2 and 3 ([06] 1.4), so O11 is explained with or without a shield. Radial also warns that "it is common for cable infrastructure to be unshielded, so it may not pass phantom power" ([06] 1.9). - Cheapest test: photo of the plug ends (metal shell or plastic) and the jacket print on both cables. DMM continuity from mix-room Catapult chassis to mic-closet Catapult chassis with only the Cat6 connecting them. - Needs: the photos and the reading.

D4. A high-impedance floating node on the ch 2 path. - Mechanism: some point on the loop (most likely the GX chassis) is held only by capacitance, so a nearby body changes its voltage. [I] This is the node-level view of A2. - Fit notes: strongest fit to O10 and O9. - Cheapest test: the 12 AWG bond (tree step 4). A bonded node stops responding to body position. - Needs: body-position check repeated after bonding.

D5. RF demodulation. - Mechanism: PSM900 transmitters, Wi-Fi or a phone put RF on the pair, and an imbalanced input rectifies it into audio. Cellular bursts produce buzz. [I] - Fit notes: fits O9 (PSM900 feeds were among the outputs unplugged) and O10. Does not explain the lights dependence. - Cheapest test: switch off the PSM900 transmitters and move phones away. - Needs: nothing extra.

D6. Shield-current-induced noise. - Mechanism: mains current flowing on a foil-and-drain shield converts to differential noise through small imbalances in each pair's coupling to the shield ([06] 5.5). [I] Applies only if the Cat6 is shielded. - Fit notes: contradicted by O11 and O2. Lifting the shield at the RJ45 should stop shield current, and it changed nothing. Pulling only the ch 2 XLR leaves the shield bonded through the other channels' pin 1, so shield current would continue. - Cheapest test: none beyond tree step 4 (a bond diverts shield current). - Needs: nothing.

E. Device faults

E1. GX4816 active input-board fault. Contradicted by O3 (the board is unpowered) and O7 (a direct source is clean). Killed, high confidence. The passive-network version is B1.

E2. SQ-7 fault. Contradicted by O3, O4 and O7. Killed, high confidence.

E3. Whole Catapult unit fault. A bank-wide defect would hit ch 1 with ch 2 if 1 and 2 share a bank ([01] P12). Contradicted by O8. Low.

F. Procedural and labelling

F1. Numbering and labels. Mic-closet XLR tails restart at 1 to 8 per row, and GX input numbers were inferred from the tails ([02] §2). The channel used in P13, P15 and T15 was never recorded ([01] §5). Any of O2, O7 or O8 could be about a different physical path than assumed. Test: label check with a tone on panel input 2, traced to every destination. Needs Elijah's lost test sheet.

F2. Bank-level cause. Anything shared by a whole bank (the cable, its shield, its lift switch, the RJ45 as a whole) cannot make ch 2 buzz while ch 1 in the same bank stays clean. Contradicted by O6, O8 and O11. Killed if 1 and 2 share a bank, moderate confidence.

F3. "16 and 20 on the same route." [I] On 12-channel racks, 16 and 20 sit in the same unit (13 to 24) but different banks (13-16, 17-20). "Route" most likely means "same unit", not "same RJ45". Under that reading, ch 1 and ch 2 share an RJ45 and ch 1 is clean, which kills every bank-level cause (F2). With Radial's pinout, ch 1 is on brown 7-8 and ch 2 on blue 4-5 in the same cable; ch 16 and ch 20 are both on pins 1-2 of their banks ([06] 1.1). Needs the channel-to-bank map.

F4. Listening method. Every result tonight was judged by ear through iPhone Live Listen on soloed monitors ([01] §1b). "Decreased" in the lights test is unquantified, and a louder hiss or a gain change could mask the difference. Test: record 20 seconds of Trident ch 2 into Pro Tools for every state and compare levels.

5. Top 10, the case against, and the best conclusion

Ranked top 10

Rank uses the matrix total, then prior likelihood, then how cheaply it can be tested.

Just below the cut: A4 Lorex camera cable in the Mix Room (low-moderate), D3 unshielded Cat6 (enabler, moderate-low), A16 mic-closet network drop (low), A3 ISP entrance (low; the service is fiber), A7 HVAC (unknown). A15 and A18 (SQ-7 on the PoE switch, console feed into the PTZ camera) are contradicted as the path into the GX by O4. They remain plausible causes of buzz in the SQ-7 system itself and should be removed anyway.

Moved since revision 1: B8 pair assignment is now contradicted by the Radial pinout (ch 16 and ch 20 share a pair position). B1 dropped from 6th to 9th because the bad channels have moved over time. B12 entered at 3rd from the install history.

What "lights off reduced it but did not remove it" implies

• At least two contributions exist, or one driver is modulated by load. Lights alone cannot be the whole driver, or the buzz would have stopped ([04]).
• The lighting side has four documented ways to push current onto grounds. It shares the audio panel by the contractor's choice ([07] item 3). The distro runs on two of three phases, so its neutral carries uncancelled current ([07] item 4). Its panel had no neutral at all until November 2025, so its 120 V loads returned somewhere else before that ([07] item 7). A lighting power transformer exists with an unknown secondary bond ([07] item 6). Any of these makes the ground voltage rise and fall with lighting load, which is what Greg heard.
• The timbre tells the two readings apart. If the lights-off residual sounded like the same buzz, only quieter, one driver is being modulated by load current (A10 to A13, A19). If it turned into a smoother, lower hum, the lights were adding their own harmonics (A6, A12) on top of a separate 60 Hz loop. Buzz rather than hum is the signature of Y-cap leakage, which weights the upper harmonics ([06] 6.2, 6.4). Greg has not described the timbre; a recording answers it.
• "Lights off" may not have turned the lighting network off. The lighting PoE switch carries lighting control, Hear Back and the NAS ([07] item 8). If it stayed powered, its ground bridge into the mix room (A17) stayed in place during the test.
• The residual needs no outside source. [I] Every powered device in both racks leaks through its own filter capacitors into a floating ground, at up to about 100 µA per device ([06] 6.4). That leakage stays on with the lights off, so the residual is expected under A2 alone.
• The "modem" (A3) is weaker than it looked. Internet is Frontier fiber, and fiber carries no shield ([07] item 11). It matters only if a cable coax service is still connected, or through the copper Cat5e from the electrical room to the office.
• Lorex (A4) is stronger than it looked. A Lorex camera cable terminates in the Mix Room near the sub, with the NVR in the office and relays in the electrical closet ([07] item 10). That is a real path into the mix-room end, relevant if the cable is shielded or the camera shares an outlet or metal with the rack.

The strongest case against the leading model

The leading model is common-mode ground noise converted by a per-channel imbalance ([03] §B).

1. It has never been measured. Nobody has read a voltage between the two chassis. The 30 to 60 V N-G reading on an open ground is also what a high-impedance DMM shows on any floating conductor through capacitance, a "ghost" voltage. Whitlock's arithmetic shows 1 V between grounds is enough and that 1 V is common ([06] 5.3), which makes the model plausible but does not measure TRC.
2. The converter is unnamed. "One leg differs from the other" fits every test because no location is committed to. A model that cannot fail on tonight's data has not been confirmed by it.
3. It needs coincident faults on two channels (three in March, two others in 2024) unless a shared cause exists. Revision 2 names two shared causes: a phantom setting (C1) and a single careless hand-wiring job (B12). The bad channels moving from 9 and 10 to 2, 20 and 22 fits either and argues against fixed factory faults.
4. Pickup explains the same pattern. A pair picking up a magnetic field (D1), or a split pair (B10, B11), also needs the far-end load to close the loop (O2), survives GX power off (O3) and is clean when the Cat6 is out of the path (O7). O10, the new-cable result and the single Radial pin map lean against it.
5. P14 is unexplained. A QSC on the mix-room Catapult output was clean. If the Cat6 and GX were connected at the time, a grounded device in the mix room did not see the buzz, which points at the Trident end (B4, C1, B12 if the terminals are on the Trident side) rather than the ground difference generally.

Resolved since revision 1: the ground-lift "no change" is no longer an objection. Radial states the lift opens only the shield at the RJ45 and never touches pins 2 and 3 ([06] 1.4), which is the path the model uses.

Verdict on the case against: points 4 and 5 are the only ones that could replace the model. The battery recorder test (tree step 2) and the Trident substitution (step 6a) settle both in under 20 minutes. Point 3 is now answered by named shared causes. The model stands at moderate confidence, with phantom and hand terminations as its two leading converters.

Best conclusion

The buzz is most likely a mains-frequency voltage difference between the mix-room ground and the mic-closet ground. It is pushed through the ch 2 and ch 20 signal pairs because the Catapult split is a passive Y between two consoles on grounds that reach no electrode. Radial's own guidance is to use transformer modules for exactly this split ([06] 1.7). Lighting load raises the voltage. The lighting feed has four documented ground problems: a shared panel, a two-phase distro, a panel that lacked a neutral until November 2025, and a transformer with an unknown bond. The rest comes from the gear's own leakage and possibly the Lorex camera path. The two-channel shape has existed since the first week of the October 2024 install, on different channels then.

Something on channels 2 and 20 converts that voltage into buzz, and it is still unlocated. The most economical candidate is Trident phantom switched on for those channels, a 30-second test. Next is a bad hand termination from the 2024 screw-terminal wiring, a 5-minute DMM test. Then come the Catapult positions and the Trident inputs, with GX inputs 2 and 20 last. An A&H replacement request is premature.

The fix that works for every branch is RX4M transformer modules on the mic-closet leg, which breaks the loop for every channel. An electrician should bond the IG bus and the lighting transformer's secondary to the service ground to lower the voltage building-wide ([02] §1a, [03] §D7, [06] 1.7).

6. Decision tree of tests (each under 10 minutes)

Tools: DMM, spare XLRs, a 12 AWG wire with clips long enough to reach between racks, a 10 kΩ resistor, a battery-powered recorder (Zoom or similar), a transformer isolator or passive DI, a QSC powered monitor, the SQ-7 phantom page, the Trident phantom switches.

Safety rules for every step: mute or turn down the monitors before switching phantom or plugging XLRs. Never lift a mains safety ground (no cheater plugs). If any chassis-to-chassis or ground-to-earth reading holds above 30 V AC with the 10 kΩ resistor across the leads, stop and call an electrician; that is fault current, not noise.

Step 0. Baseline (3 min). Buzz audible, lights on. Record 20 seconds of Trident ch 2 and ch 1 into Pro Tools, fader and gain untouched for the whole session. Note the peak level. Every later "better" or "worse" is read off this.

Step 1. Phantom audit (5 min). Photograph the phantom state of 1, 2, 16 and 20 on the Trident, on the SQ-7 phantom page, and the GX phantom LEDs with the GX powered. Then, with SQ-7 phantom off for 2 and 20 and the GX powered, look at the GX +48V / PP LEDs on inputs 2 and 20. The LED senses any voltage above 24 V at the socket, so a lit LED with SQ-7 phantom off proves Trident phantom is arriving through the split ([06] 3.1, 3.3). Then switch every phantom off on those four channels at both desks, wait 30 seconds, listen and record. - Buzz gone: branch C. Turn Trident phantom back on for ch 2 only. Buzz returns means C1, C3 or C4. Turn it on for ch 1 instead: buzz now on ch 1 means C1 (ground current through phantom, any channel will do it); ch 1 stays clean means ch 2's Trident phantom resistors are mismatched (B4 combined with C1). Go to step 4 for the permanent fix. - Buzz unchanged: C1 to C4 are dead. Leave phantom off for the remaining steps and continue.

Step 2. Battery recorder at the far end (5 min). At the mic closet, move the Catapult out 2 XLR from GX input 2 into a battery-powered recorder, phantom off, recorder on batteries, nothing else plugged into it, hands off. Listen on Trident ch 2. - Buzz gone: the buzz needs a second ground. Ground loop confirmed; D1, D2, B8, B10 and B11 are dead. Continue at step 3. - Buzz stays: the pair itself is picking up noise. Branch D. Run a spare Cat6 on the floor along a different route, away from lighting and the LED wall, and repeat. Buzz follows the route: D1. Buzz stays on any route: B11 (check model labels and pin map with a tone).

Step 3. Meter the grounds, lights on and lights off (10 min). - AC volts, GX chassis (a rack screw) to Trident chassis, using the long 12 AWG wire as one lead. Read it with the meter alone, then with 10 kΩ across the leads (or LoZ mode). - AC volts from the ground pin of the outlet feeding each rack to a known earth: a copper cold-water pipe or the service panel enclosure in the electrical room. - AC volts N-G at the outlet feeding each rack (the GX outlet and the Trident PSU outlet). - Repeat all readings with the stage lights off. - Reading: chassis-to-chassis under 50 mV means the ground difference is too small to matter; go back to branch D. A reading that collapses under 10 kΩ is a high-impedance float (A2, D4). A reading that holds and rises with lights is load current on the grounding system (A10, A11, A13). An outlet ground more than 1 V from the water pipe confirms the open IG ground (A1).

Step 4. Temporary 12 AWG bond (3 min). Clip the 12 AWG wire from GX chassis to Trident chassis, listen, record. - Buzz drops sharply: driver confirmed. The bond is a safe interim measure while the electrician bonds the IG bus to the service ground. - Buzz unchanged: the difference is not between these two chassis. Move the far clip to the mix-room Catapult rack, then to the Furman rack, and repeat. - Also repeat Greg's body-position check with the bond on. No body effect now confirms D4.

Step 5. Clear the wall panel (2 min). Unplug tie line 2 from the mix-room Catapult input. Buzz unchanged: B7 dead. Buzz gone: tie line 2 or the panel box is involved.

Step 5b. Leg-balance ohms checks (10 min, power off, everything on the segment unplugged). - Hand-wired segments (B12): for ch 2, 20 and control ch 1, DMM ohms pin 2 to pin 1 and pin 3 to pin 1 at one end with the other end open. Anything other than open (OL) is a leak. Then end-to-end pin 2 to pin 2 against pin 3 to pin 3; a difference above a fraction of an ohm is a bad joint. Open the screw terminals for 2 and 20 and look for stray strands. - GX inputs (B1): GX off and unplugged, nothing in the inputs. DMM ohms pin 2 to pin 1 and pin 3 to pin 1 on inputs 1, 2, 16 and 20. Each input's two legs should match within about 1% ([06] 5.3). A leg that differs from its partner is the part. - The same pair of readings at the Trident's ch 2 and 20 inputs (Trident off) covers B4.

Step 6. Locate the converter (three sub-tests, each under 10 min). - 6a. Trident substitution. In the mix room, unplug the ch 2 feed to the Trident (at the DPA fan or the receive Catapult output) and plug a QSC powered from a mix-room outlet in its place. Cat6 and GX input 2 stay connected. Buzz in the QSC: the Trident is not required; the converter is in the Catapults, pair or GX. Clean QSC: the Trident input 2 is the converter (B4, or B6 for the DPA). - 6b. GX swap. Catapult out 2 into GX input 1, Catapult out 1 into GX input 2. Listen on the Trident. Buzz stays on Trident ch 2: upstream of the GX (B2, B3, B4, B9). Buzz moves to Trident ch 1: GX input 2 is asymmetric (B1). Repeat with 20 and 16. - 6c. Mic-closet bank shift. At the mic closet only, move the Cat6 carrying bank 1-4 into the bank 5-8 jack. Signal 2 now leaves on mic-closet XLR 6. Patch XLR 6 into GX input 2. Buzz gone: mic-closet position 2 hardware (B2, B9). Buzz stays: mix-room Catapult side (B3). The earlier port move ([01] T15) was ambiguous because no one tracked where channel 2 went.

Step 7. Strip the GX grounds (5 min). Mute the PA. Unplug all 16 GX output XLRs at once, then also the GX IEC cord. Buzz gone with outputs out: the aux loads (A8, A9) are the GX end of the loop. Buzz stays with everything out: the GX is still referenced through the other 47 inputs' pin 1 and the Cat6 shields, so the reference comes from the mix room; go back to step 4 results.

Step 8. Hunt the residual, lights off (10 min). With stage lights off and the residual audible, one at a time, 60 seconds each, recording each: - the network cables of the Hear Back hub and the NAS where they meet the lighting PoE switch (A17); - the Mix Room Lorex camera's Ethernet, then the NVR and its PoE switch (A4); - any Ethernet into the mic-closet rack other than the SLink and the Catapult link (A16); - any audio cable into the back of the PTZ camera (A18); - the ISP device's network cable, and any coax (A3); - HVAC off at the thermostat (A7); - the PSM900 transmitters off (D5). The one that drops the residual is the second driver. Also note whether the lighting PoE switch itself stays powered when the lighting breaker is off.

Step 9. SQ-7 network (1 min). Pull the SQ-7 network cable, which was on the lighting PoE switch in 2025 ([07] item 8). Expected no change on ch 2 given O4; any change points to A5, A15 or A18. Leave it out either way until the SQ-7 has its own non-lighting switch, as Vince recommended ([07] chat.db 810507).

Step 10. Transformer isolation on ch 2 (5 min). Insert a 1:1 mic-level isolation transformer between Catapult out 2 and GX input 2. A passive DI run backwards (XLR in, 1/4 inch out to a line input) with its ground lift engaged is a crude substitute. Buzz gone: confirms the common-mode mechanism and proves the per-bank fix (RX4M modules in the mic closet).

Order logic: step 1 can end the search in 30 seconds. Step 2 decides loop against pickup and removes half the table. Steps 3 and 4 size and confirm the driver. Steps 5b and 6 name the converter. Steps 7 to 9 name the second driver. Step 10 proves the fix before money is spent. Electrician-only items run in parallel and are not studio tests: the lighting transformer's secondary bond, the Stage Lights subpanel, and neutral current on the two-phase distro.

7. What Greg should send (12 items)

1. Photo: Trident phantom switches on channels 1, 2, 16 and 20, plus the SQ-7 phantom page for the same channels.
2. Yes / no: with all phantom off on 2 (both desks, 30 seconds), is the buzz gone?
3. Photo: the model label on every Catapult unit at both ends, and its ground-lift switch positions.
4. Photo: both ends of the original Cat6 and the silver cable, showing the plug shell (metal or plastic) and the jacket print.
5. Reading: AC volts GX chassis to Trident chassis, meter alone and with a 10 kΩ resistor across the leads, lights on and lights off.
6. Reading: at the outlet feeding the GX and at the outlet feeding the Trident PSUs, AC volts N-G and AC volts ground pin to a cold-water pipe or the service panel, lights on and lights off.
7. Photo: everything plugged into the lighting PoE switch (look for the SQ-7, the Hear Back hub, the NAS), the back of the PTZ camera (any audio cable), and any network cable into the mic-closet rack.
8. Yes / no: with Catapult out 2 feeding a battery recorder instead of the GX, is the buzz still on Trident ch 2?
9. Yes / no: with a 12 AWG wire clipped from GX chassis to Trident chassis, does the buzz drop?
10. Which channel: with Catapult out 2 in GX input 1 and out 1 in GX input 2, is the buzz on Trident ch 2 or ch 1?
11. Recording: 20 seconds of Trident ch 2 bounced from Pro Tools with lights on, then with lights off, same gain. Frequency analysis will show 60, 120 or 180 Hz.
12. Photo: the lighting power transformer's nameplate and the breaker feeding the stage distro. An electrician adds whether the transformer secondary is bonded to ground and what is inside the Outdoor / Stage Lights subpanel.

Items 2, 8 and 10 alone would cut the 48 hypotheses to about five. The ohms readings in tree step 5b come next.

8. Addendum from the gear research and the history mining

Sources: [06] 06-gear-research.md and [07] 07-history-mining.md, both read in full. Their facts are already built into sections 0 to 7: the matrix rows, tie-breaks S9 to S13, the detail entries, the top 10, the tree and the checklist. This section collects what changed and answers the five questions put to the sweep.

8.1 The five items scored against O1 to O12

The lighting items (A11, A12, A13, A19) can each explain the lights-dependent part of the buzz. None can explain why only ch 2 and ch 20 buzz: they raise the driving voltage on every channel equally. They sit in the top 10 as one grouped row at rank 7.

8.2 A two-channel fault that changes channel numbers over time

The record: "channels 9 and 10" buzzing in the first week after Thai's October 2024 screw-terminal install ([07] chat.db 71841, 1321). Ch 2 bad from about January 2026, then 20 and 22 "since the new year" ([07] MAR). Ch 2 and 20 on 9/25 ([01]). There are three ways to read the move.

The combined reading (inference, moderate). A setting can cause the physical fault. With phantom on at one console, hot-patching a line through a passive Y dumps the phantom capacitors through the other console's input. That can damage a clamp diode or resistor on one leg. The damaged leg then stays imbalanced after phantom is switched off. Step 1 separates the two: buzz that stops with phantom off is a setting (C1); buzz that stays is physical, and step 5b finds it.

A weak pattern in the pinout (low confidence). With Radial's pinout and straight-across numbering ([06] 1.1), channels 2, 10 and 22 all sit at bank position 2, the blue pair on pins 4-5. That is three of the five channels ever reported bad. By chance alone a spread like this happens about one time in ten. Pins 4-5 sit between the legs of the 3-6 pair at every RJ45 termination, so they are disturbed first by a poor termination ([06] 2.5). Against it: ch 20 is at position 4, clean ch 16 is also at position 4, and the 2024 system may not have used Catapult banks at all. It is worth one check: if tree step 6c shows the buzz following bank position 2 hardware, look at every RJ45 termination on the run for untwisted blue pairs.

8.3 Magnitude check

Whitlock's bridge arithmetic ([06] 5.3): a 1% mismatch in one phantom-feed resistor with 1 V between grounds gives about 2.5 mV differential, unloaded. That is about -83 dBu at the input once a 150 Ω source loads it. After 50 dB of preamp gain it is about -33 dBu, 50 dB above the preamp's noise floor. With a Y-cap-floated chassis in place of a 1 V ground difference, about 3.4 mV before loading ([06] 6.5). A 0.1% mismatch still gives about -103 dBu at the input. Every converter candidate in groups B and C is therefore physically large enough to be heard. This supports the model's plausibility. It does not replace measuring the voltage between chassis (tree step 3).

8.4 What moved in the ranking

• In: B12, the hand-wired screw terminals, at rank 3. A17, the Hear Back hub and NAS on the lighting PoE switch, at rank 8.
• Up: the lighting-side mechanisms, now grouped with the transformer (A19) and the two-phase distro (A12), at rank 7.
• Down: B1 (GX input) from 6 to 9, because a fixed GX fault cannot account for bad channels moving since 2024. A3 (modem) fell below the cut, because the service is Frontier fiber.
• Out: A6 (lighting leakage) and D3 (unshielded Cat6). Radial's statement that the lift opens only the shield removed D3's special role.
• Contradicted: B8 (pair twist), because ch 16 and 20 share a pair position, and D6 (shield current).
• Unchanged: A1 + A2 at rank 1 and C1 at rank 2. C1's total rose to 17 on the GX phantom-sensing fact and the Whitlock number.

8.5 Checklist changes (still 12 items)

• The two outlet readings, old items 6 and 7, merged into item 6.
• New item 7: a photo of everything on the lighting PoE switch, the back of the PTZ camera, and any network cable into the mic-closet rack. It answers A15 to A18 and whether the SQ-7 is still on that switch.
• Item 12 replaced: the main-breaker and subpanel photo became the lighting transformer nameplate and the breaker feeding the distro, with the electrician's check of the transformer's secondary bond.