I read the entire thread and couldn't find a single reference to what these settings would be, just a lot of hand-waving and "I haven't isolated them". Seems like an extreme jump to conclusions.

But, let's imagine there is such code - what kind of optimization could actually be done, that would meaningfully improve the tests? The car is on a sled, at a predefined speed. All that can be changed is maybe the timing of airbags or the collision sensor sensitivity, and if there was an ideal set of parameters that improves safety results, I can't imagine that is not the one used in the real world. Am I lacking imagination?

I could imagine there is tension between airbag deployment times and wanting to reduce false positives.

If you knew the airbag needed to deploy in a test, perhaps you could score better with faster deployment, but that wouldn’t be something you want to send to your entire fleet for fear of false positives.

But just a guess. I don’t work in cars.

ANCAP do appear to test Automatic Breaking and Lane Assistance.

Why would you need a testing specific software build exist at all?

Or put differently what legitimate reason would you have for this if you’re trying to test existing safety feature?

Noting that no wrong doing has been found (yet) that seems to require a pretty immediate and credible answer

Maybe extra diagnostics for Tesla engineers to learn from.