Unencoded chunks number is incorrect

[nazar-pc]

I just checked one of my 1.9T plots and have not found a single record that doesn't have all of its chunks encoded.
Yet https://subspace.github.io/protocol-specs/docs/consensus/proof_of_archival_storage#plotting-to-disk in item 9 note says ~0.8% of chunks were unencoded.

Yes, it says that calculation was done for K=17 and we ended up using K=20, but still. I think it should be corrected.

I see that 56..69% of source chunks are encoded (62.87% mean value and 63% median), but when taking erasure coded chunks, it increases to 100% from my testing.

The code is ready for some chunks to not be encoded, but probability of this must be far lower than 0.8% with mainnet parameters.

[teor2345]

Do we have tests or benchmarks for un-encoded chunks? Or the un-encoded chunk rate?

Seems like something worth having to make sure any parameter changes result in a reasonable rate.

[nazar-pc]

I mean we have various tests that may indirectly check this, but ultimately these parameters are more like constants and will not change without a major effort to ensure compatibility.