The Great Flange-to-Sensor Distance Article

Also, after skimming through the second article, I had to wonder: if the first article found dropped cameras and other problems in flange-to-sensor data, wouldn't they also find mount tilt if they looked for it? Given that drops don't usually impact the camera exactly orthogonal to the plane of the flange, it seems like flange-to-sensor distance won't be their only problem. But maybe replacing the flange on these also fixes any tilt. Then again, there may be damage that affects the tilt but not the flange-to-sensor distance.

Someone posted technical illustration of the Denz measuring equipment, and I wonder if they couldn't make a similar apparatus with 3 or 4 pairs of measuring beams (or 2 in a cross pattern), thus it could measure the distance at several points and thus the parallelism of the 2 planes.

For DSLRs the distance through the backmirror to the AF sensor must be exactly identical to the distance of the image sensor, because the AF sensor focuses on the wrong distance otherwise. DSLRs generally have an adjustment for this (often per-lens-ID). (Worth pointing out that DSLRs are calibrated at the factory. I suspect they go for a flange distance that is slightly under the spec, because this direction is easy to remove by calibration. Note how the distributions in the article match up rather well with this expectation).

These issues are one of the main reasons why lenses don't have hard infinity stops any more. The other is that it's expensive to do, and the third is that it is less necessary with autofocus.

If the optical axis is not normal to the sensor(s) however you have all sorts of problems focusing across the frame.

For Phase Detection autofocus (PDAF) this is an instant detection with no hunting using the polarization of the light to indicate distance; more expensive cameras have calibrations per lens available, as do some lenses themselves (Sigma and their USB dock specifically). As another comment notes, if it's a Digital Single Lens Reflex (DSLR) that implies the PDAF sensor is fed by the mirror in front of the sensor, and has multiple points of tolerance/failure. For Mirrorless, or Compact System Cameras (CSCs), if they use PDAF then it's embedded in the sensor. This allows for significant improvements in accuracy, at the cost of reducing the amount of data captured by the sensor for the image (some of that data space is now being used for focus data) and in some cases artifacting due to the pattern of those sites (Olympus E-M1 vs E-M5 for astrophotography, etc.).

Something that's often misunderstood about all of this is that focus isn't an ON/OFF thing; you can think of it as a curve. There is a point of maximum focus, then it drops off in space both in front of and behind that point. How fast it drops off is defined by both Focal Length and Aperture; this means that with something like a phone it's not very hard to take a picture that's "In Focus" because the Depth of Field (DoF) that LOOKS in focus is very deep. This is why PDAF/CDAF work as well/quickly as they do, even with all these stacking tolerances; it's not trying to hit an ideal spot, just close enough to look good. Spend more, and you can hit it more accurately and narrow that band, giving you a different (generally interpreted as more professional) look.