Some of the calibration offsets are device specific, so for a full calibration the unit as a hole needs to be calibrated.
Early on they did calibration in store with machines like these[0], but it caused too many issue.
In part because geniuses, like the one in question, like to draw conclusions based on incomplete data and as a consequence get overconfident in their ability repair and asses.
Where this confidence comes from is beyond me, because most of their work has been relegated to a paint by numbers process using tools that abstract as much as possible away, limiting critical thinking. Ironically, the fuckups caused by this confidence, causes more to be abstracted away limiting the experience and knowledge they build up, and the more this is limited the more confident they get and so this Dunning–Kruger cycle goes on and on.
In any case, in part because of too many issues with these machines Apple has worked to move as much as the calibration process away to the factory, limiting the actions that are necessary in the Apple Store to using a special app nowadays[1].
Whatever the case, both the machines used back in the day and the app currently in use combine local data with remote data.
Information about factory calibration is looked up in a cloud database based on screen ID number, that calibration data is downloaded and combined with calibration data that is collected locally by running a special firmware to run some local calibration tests.
Everything is verified before being finalized and the part is married to the device via serial numbers, storing the relevant calibration data in the relevant sides of the equation (i.e. screen controller and device).
If at any point in the future there's a mismatch, the offsets (i.e. calibration data) is ignored entirely and no offsets are applied at all.
This serves a couple of purposes, but the important ones being the prevention of damage and a way to remind the person observing it (i.e. presumed to be the worker repairing it) that calibration is still necessary and/or has not been completed successfully.
0: https://theapplewiki.com/wiki/3D_Touch_Calibration_Procedure 1: https://theapplewiki.com/wiki/RepairCal
That sounds like a process thing that could go either way. I can imagine a system where they wait until the ipad is assembled and then use the hardware on the ipad to run the color calibration against the screen being simpler/more sensible to integrate into production than a system where panels need to be plugged into a separate thing earlier on.
The real test would be grabbing two screens with calibrated controllers and switch the controllers out, then compare accuracy after the swap with accuracy before the swap.
That's what they showed in the video. The replacement screen was from a different iPad; it wasn't an entirely new one.
If the cal data is stored in the controller, it's likely that using the old controller with a different screen means the calibration is off a little, but if that's better than using the new screen and controller (presumably with cal data for the new screen) together, there's something going on beyond that.