> The improvement is apparently due to the changes made in response to Challenger and Columbia

A purely technical discussion on the problems that could down a Shuttle ignores the meta problem of poor risk analysis by the Shuttle managers. Basically they normalised deviance[0], in effect deciding that a technical problem wasn't an issue because they got away with it previously. This intrinsic weakness did not go away after the Challenger loss (hence Columbia [1]). Arguably, the only reason there wasn't another fatal loss was that the programme was grounded before after another accident could occur.

[0] https://en.wikipedia.org/wiki/Normalization_of_deviance

[1] https://en.wikipedia.org/wiki/Space_Shuttle_Columbia_disaste...

The analyses leading to the loss estimates I mentioned were done in 2010, shortly before the Shuttle's retirement, due to recognition of the previously poor risk analyses. As is clear from the issues seen with Starliner, this normalization of deviance has been significantly reduced, in that it doesn't seem to affect the newer style of contracting. Both Dragon and Starliner have had to scrub launches and make fixes for things that would be considered to not be a big deal if they weren't carrying humans, and are expected to review and justify any deviation from expectations even if it was fine previously.

We're seeing plenty of normalization of deviance with SLS and Orion though. There were the issues with various backup systems and IIRC no life support in Artemis 1, with issues with the heat shield only recently being made public, yet Artemis 2 is intended to carry a crew. So I would not be surprised if Congress only finally gives up on that waste when they cause another crew to be killed and can blame NASA instead of themselves.

I would guess that some of the improvement was from operational changes following from more realistic risk analysis, and the gradual acceptance that the space shuttles would never live up to the expectations behind that name.