Nightly loss of registration since plant maintenance — 12 hours of gateway telemetry, all customer equipment eliminated

I'm hoping to reach someone who can route this to maintenance. I've been through front-line support and the troubleshooting steps offered don't address what the data shows. I work in infrastructure professionally and I've already eliminated everything on my side of the demarc.

**Symptom:** Loss of external connectivity nightly, roughly 10 PM to 10 AM, beginning 7/23. Onset coincided with construction at the pedestal serving my address. Gateway is an XB8 (CGM4981COM) in bridge mode.

**Customer side fully eliminated.** I removed the in-line amplifier from the path entirely, removed all splitters, disconnected every other coax device, and factory reset the modem. The modem is now the only device on the coaxial plant, on a direct run from the ground block. Every measurement below was taken in that configuration.

**Telemetry — one-minute polling, 2:00 AM to 2:00 PM on 7/28:**

*Seven complete losses of registration.* Downstream bonded channel count fell from 33 to near zero and rebuilt at 02:16 (4.5 min), 02:22 (12.5 min), 02:47 (5.5 min), 02:57 (5.5 min), 05:17 (20.5 min), 07:26 (7.0 min), and 09:22 (6.5 min). Separately, my downstream channel plan changed over the course of the day — it ran 525–711 MHz in the morning and 453–639 MHz by evening. Same 33 channels, shifted down 72 MHz.

*Upstream bonding will not hold.* Four upstream carriers are provisioned to me at 16, 22, 29, and 35 MHz. In the overwhelming majority of samples only one is bonded — a single carrier at 35 MHz. All four appear together only occasionally and briefly. My polling interval is roughly 69 seconds, so I can't characterize how long the four-channel state persists, but across twelve hours my modem was running on one of four provisioned carriers nearly all of the time. Upstream transmit renegotiates repeatedly between approximately 31 and 43 dBmV.

*Downstream levels out of spec.* Power ranged from 16.9 dBmV per channel at 2 AM down to 13.5 dBmV at 2 PM, above the +15 dBmV DOCSIS per-channel maximum for most of the night. Across 32 bonded carriers that's roughly +31 dBmV composite at the receiver. Thirteen of 32 channels measured below 33 dB SNR, worst case 26.1 dB.

All seven registration losses occurred while downstream levels were above 15.8 dBmV. None occurred after levels fell below that.

**Other data points:**

- Xfinity's own diagnostic reports DOCSIS Health "red."
- A representative confirmed maintenance on my segment under tickets [Edited: "Personal Information"], [Edited: "Personal Information"], and [Edited: "Personal Information"], and confirmed the maintenance timeline coincides with the onset of my problem. Those tickets are closed; the fault continues.
- Multiple neighboring households on the same segment report concurrent outages in the same nightly window.
- I logged outdoor air temperature at one-minute intervals for five days. Temperature alone doesn't explain it — service failed at 80.7F and recovered at 68.0F, and one outage persisted through an 85.5F afternoon. The registration losses track downstream level, not weather.

**What I'm asking for:**

1. Post-maintenance signal level verification at my drop and tap. Levels are roughly 15 dB hot with nothing of mine in the path.
2. My node ID, and confirmation whether a maintenance ticket is currently open on that segment.
3. Investigation of return-path impairment on my node, given the sustained upstream bonding failure above.
4. My work order for Sunday 8/2, 8-10 AM was created as a "Professional Install" with a $100 charge. I'm an existing customer reporting a service fault, not requesting an installation — requesting it be reclassified as a service repair with the charge removed.

An FCC informal complaint is open on this matter. I have the full telemetry export and charts available, and I'm happy to provide my name, service address, account number, and work order number by direct message.

EDIT: Correcting two things in my own post before anyone has to. (1) My polling interval is ~69 seconds with occasional longer gaps, so I originally over-characterized the upstream bonding behavior. I can support that only one of four provisioned carriers is bonded in the large majority of samples — not how long each state lasts, and not a count of bonding attempts. (2) I originally reported a downstream channel plan change based on a sensor whose index is a table position rather than a fixed frequency; the frequency set was unchanged in that instance and the conclusion was mine in error. The registration losses, the out-of-spec downstream levels, and the single-bonded-upstream-carrier finding are unaffected.