All kind of VNA measurements require proper calibration.
Quote: "Calibration must be performed whenever the frequency range to be measured is changed." If you do TDR with a VNA, then S11 and/or S21 are still measured in the frequency domain, and the corresponding time domain data are calculated via IFFT then. The calibration data are stored per frequency point (for those frequencies which are in effect when the calibration is done). So when the frequency points change, the calibration data for the new frequencies can only be interpolated from the old calibration, or even just extrapolated, so that accuracy suffers. So better always re-calibrate when you change start/stop/center/span/#points, or when you change the set (which includes all involved cables and/or adapters up to the calibration planes). You can also restore a saved calibration, which restores start/stop/#points in together with the fitting calibration data. Still it is up to you to ensure that the test set is the same as when the saved calibration was established. Since (at least my) NanaoVNA does drift, I also do not rely on stored calibrations for a longer duration.
EDIT: I see you added images. Much better now. It seems, though, there is an impedance mismatch already at the source, which would imply that the cable impedance were not 50 Ohm. If you switch the trace type to
IMPEDANCE RESISTANCE then you can see the impedance profile over time (distance).
BTW: The start is outside the left border of the screen. Try to shift the traces a small amount to the right by applying say -20 ns Electrical Delay. I guess the horizontal part of the trace covering the cable length will go even a little bit more negative then.
How does it look with the "good" cable, for comparison?