Though potentially they just all got lucky and caught a problem early enough that most units with the problem batteries haven't left Ford control yet.

Instead, our PLCs (programmable logic controllers) report to in-line vehicle sequencing databases. Every step of the process is logged to ILVS, and - unlike our work with furniture or consumer goods manufacturers - that is the only thing they care about. It doesn't matter if you can see that there's a physical assembly with properly torqued bolts that we've delivered to Ford's loading docks, we only get paid if the torques are correctly recorded for that serial number in ILVS. Each line tracks each part with serialized 2D barcodes. Any time a measurement is made or an operation is performed, the PLC will trigger an upload to record that data in ILVS, whether that's a barcode quality grade of the 2D scanner camera, or a force measured to drive a clip in, or a temperature measured while setting a heat stake, or the current drawn by a sensor, or...anything. For jellyroll battery winders I've done, that will include obvious stuff like Hipot test results but also minutiae like the row and column and stack height in the tray that transfers the jellyroll between the winding machine and the pack assembly machine.

This sort of database provides not only recall protections, but also ANOVA gauge reliability and repeatability to retain control over the process, and predictive maintenance to know when a machine or mold is wearing and causing a trend in the measured data (even if it's all within acceptable tolerance).

If I was recording bolt torques that were manually driven by a human (rather than a robot), I'd probably use a wireless electronic torque wrench with a UWB RFID tag on the socket (or some discrete tags on the jig) to verify that you'd reached the appropriate torque at the appropriate location. You'd be able to automatically query the database for the barcode of the assembly with any data that came with its forming/casting/other production data, the lot ID of the bolt, the calibration source of the wrench, the employee ID, the time and date that it was torqued, how much time it took the operator to torque each bolt, and on and on.

We do still have video cameras pointing at our machine recording to a ~1 week circular buffer, but that's more so that we can reconstruct the sequence of events that led to a machine fault.

The data-to-noise ratio of video is pretty bad (most of it going to be wasted on data for... the factory floor or whatever), compared to something like logging the torque of every bolt on the line (which, they already do in some situations).

Of course there's usually always a tradeoff with other properties, like weight or cost.

[1]: For example https://en.wikipedia.org/wiki/Lithium_iron_phosphate_battery...

Why are you embarrassed by reality?