So I tested it with your fitted TLE of 65488, for LW2DYB’s reception time (4 Nov, 21:42:17 UTC):
65488 was over northern Argentina at that moment (subsatellite point 21.5° S, 65.9° W), so in view for him.
The Doppler shift there: depending on where in Argentina he is, +7.5 to +8.4 kHz. Around Buenos Aires, Córdoba or Mar del Plata, the frequency he’d see is 401.5575–401.5584 MHz.
What he reported: 401.558 MHz, with only 279 Hz of correction (CFO) in his decoder.
That matches within a few hundred hertz. His 401.558 is simply 65488’s 401.550 MHz seen while it was still approaching.
Your new zoom (relative to 401.555 MHz) supports it too:
The fitted 065488 curve now runs right over the middle trace.
The two sidebands sit at about ±9 kHz, more than my rough ±7 from the first screenshot. That fits 19k2: an alternating bit pattern (preamble or idle) at 19,200 baud produces sidebands at ±9.6 kHz. At 9k6 they’d be at ±4.8 kHz, as you said.
Around 03:23–03:25 there are wide, faint bursts on the middle trace: packets.