Your new single Dark frame is a very good match for the Master Dark you have been using. That means that the problem is with the session of Light frames from several nights ago.
I cannot think of anything that would cause that one set of Light frames to be darker than those taken the next night. The usual suspect would be that the Offset had changed. However, the FITs Header shows that both Gain and Offset match between the sessions and yet one session's subs came out dark versus the next night's session where the levels are where they would be expected.
The only differences between the two nights of shooting appears to be ambient temperature and (as a result) focus position. Neither of those parameters should cause a difference in average brightness. You might think that a more transparent night could cause the overall image to be darker but that cannot be the cause here since the Lights dropped in brightness enough to be darker than a matching Dark frame. A dark sky background cannot be less bright than a Dark where no light hits the sensor at all.
Is there a possibility that you used a different connection method to the camera on those two nights? It is a wild outlandish thought but I am wondering if you used the ASCOM driver on the earlier session and then went back to the native ZWO (SDK) driver for the other nights? For that to cause this, it would also require that a subtle bug (or bugs) be present in the version of NINA you are using. ASCOM does not report the Offset for the camera to the capture software. If the ASCOM driver were set for a lower Offset, --AND-- your version of NINA defaulted to the last used / set native driver Offset for recording the FITs header data, then the darker subs might result. However, NINA would normally have reported the camera as "ASI Camera (1)" or similar when the ASCOM connection is used. This line of thought doesn't sound feasible since it would require multiple bugs in just the right places. Such a set of coincidences is very highly unlikely -- akin to stacking a bunch of BBs one atop the other.
How this could have happened is a mystery for me. Does anyone else have a way to explain how one set of subs could have a 600 ADU darker median value even though the FITs Header shows the key parameters as matching? It has me stumped.
I did notice one more thing you should modify going forward. You have been using a USB setting of 40 for the Light frames but appear to be using a USB setting of 80 for the Dark frames. All frames should ideally use the same setting. This should not account for the Light frame differences since they appear to be the same in both Light sessions. However, for the best calibration, you should standardize on one setting (I use 40) for everything -- Light, Dark, Flat, Flat-Dark (or Bias). It has been reported that is can sometimes make a slight difference in the amount of Dark Current in the frames.