20 dB downstream tilt — OFDM ch 34 at -13.3 dBmV / 31 dB SNR, 29.5% uncorrectable codewords

My downstream has excessive tilt and the top of the band is failing. All 32SC-QAM channels (375-561 MHz) are clean with zero codeword errors. Both OFDMchannels are degraded, and the one at 957 MHz is effectively unusable.Gateway: Comcast CBR2, business line.CHANNEL DATA (representative; all 32 QAM channels look like the first four)Ch   Freq       Power        SNR       Modulation2    375 MHz    +7.0 dBmV    45.6 dB   256 QAM5    399 MHz    +6.3 dBmV    45.3 dB   256 QAM29   543 MHz    +1.8 dBmV    41.1 dB   256 QAM32   561 MHz    +1.4 dBmV    41.3 dB   256 QAM33   722 MHz    -1.4 dBmV    40.0 dB   OFDM34   957 MHz   -13.3 dBmV    31.0 dB   OFDMCODEWORD COUNTERSCh 34 (957 MHz, OFDM)Unerrored       80,183,143Correctable     32,095,887Uncorrectable   47,018,299     = 29.5% of all codewordsCh 33 (722 MHz, OFDM)Unerrored       62,245,416Correctable     25,340,030     = 28.9% of all codewordsUncorrectable        2,569All 32 SC-QAM channels: 0 correctable, 0 uncorrectable.ANALYSISDownstream tilt is 20.3 dB, from +7.0 dBmV at 375 MHz to -13.3 dBmV at 957 MHz.The roll-off rate is not uniform:375 -> 722 MHz    2.4 dB per 100 MHz722 -> 957 MHz    5.1 dB per 100 MHzThe rate more than doubles above 722 MHz. Normal RG-6 attenuation across thisspan accounts for roughly 2-3 dB, not 20 dB, so cable length does not explain it.31.0 dB MER is well below what OFDM subcarrier modulation requires, which isconsistent with 29.5% uncorrectable on that channel. Ch 33 at 40.0 dB MER ismarginal: FEC is holding, but 28.9% of codewords need correction to get there.The error rates on both OFDM channels track their MER exactly as you wouldpredict. This is a signal problem, not an unused-profile counting artifact.WHAT I HAVE RULED OUT- Not inside the building. Nothing on my side is frequency-selective at 900+ MHz,and my equipment sits downstream of the gateway.- Not broadband corrosion or water ingress. Those would degrade the upper QAMchannels as well, and 375-561 MHz is perfectly clean.- The knee above 722 MHz is characteristic of a passive with a frequency cutoff.I would suspect a splitter or passive in the drop rated to 1 GHz or below;957 MHz sits above the cutoff of a legacy 5-860 MHz component.I am reporting this on the measurements rather than a specific outage. I wouldrather have it corrected before it becomes one.WHAT I AM ASKING FOR1. Flap list for this modem.2. T3 and T4 timeout counts.3. Which OFDM profile I am running on ch 34, and whether I have been downshifted.4. RxMER per subcarrier on ch 33 and ch 34.5. Tech dispatch to check the drop for a sub-1 GHz passive, plus connectors andthe ground block.6. Interim: can ch 34 be excluded from the bonding group, or the profiledownshifted, until this is corrected? I have roughly 1.2 Gbps of clean SC-QAMcapacity, so I lose almost nothing by dropping it, and it stops theretransmits in the meantime.There is an existing ticket on this. Happy to provide the number and accountdetails by private message.