You’re off by an order of magnitude. Loots like ~% 1 of ethereum mainnet blocks missing on average.
During the Eth Tokyo hackathon my team and I made an attempt at an implementation of the censorship oracle and applying it to an optimistic bridge on Gnosis, a chain with the same consensus mechanism as Ethereum, except the slot time is 5 seconds, and the epochs are a different length (a couple more differences, but the point is the censorship test applies to Gnosis equally well).
Here’s a dashboard of weekly averages of missing blocks on Ethereum and Gnosis.
Notice that the shapella hardfork caused >10% of missing blocks in the first 24 hours (the daily resolution is not shown here, but if you check it’s ~10%). That’s the first “bump”. The second “bump” is the recent inactivity leak issue.
We also made this censorship test parameter calculator in google colab. The calculator uses the mpmath library for arbitrary precision. The sample calculation matches the examples you provided.