Reliability/consistency of "chia plots check"

[gromran]

First of all: I have not yet understood exactly what the check does. How can more than 30 proofs be found with an n-value of, for example, 30?

And I once took the trouble to check a few arbitrarily chosen plots with different n-values.
As long as n is the same, the same proof number always comes out for a plot, so that seems to be stable. But different n-values sometimes give very different proofs found:

This is still quite consistent with individual plots. For others, the ratio across the various n-values is very different. How does that come about?

[loppefaaret]

First off, the challenge from chia plots check is static. Meaning if you run -n 30 10 times against the same file, it should give the same result everytime.
Imagine that the check issues a challenge with a diff matching the k size of the plot, so it on average should return -n amount of proofs.
Adding in the cryptographic variances, this makes the values swing as you can see - a plot might be lucky against the first 10 or even 50 challenges, but once you get up to 1000, the variance starts zeroing out.

chia plots check is meant as a tool for verifying a plot. Not for measuring how "lucky" it is - as you are only running it against one single challence - it defaults to -n 30 - and if you see a plot run less than 0.7, you target that plot with -g and run a more thorough check.

[qpirsel]

chia plots check is meant as a tool for verifying a plot. Not for measuring how "lucky" it is - as you are only running it against one single challence - it defaults to -n 30 - and if you see a plot run less than 0.7, you target that plot with -g and run a more thorough check.

Then the given result as a quality number is really missleading. And even more because one would except that the more n-values are used the results should converge without those "huge" ups and downs seen in many of his results for a given file.