Some episodes are failing because the extraction is truncated

[xpilasneo4j]

Lost of episodes fail on Theme or Episode because the _extraction is truncated and finished by CLEAN CLEAN CLEAN instead of the expected output "Answer:" or "- ..."

Output of _extraction
"And that's what I like to do in that could not have been scripted better. You actually got back catch, well done. Loyalty. That is all we have time for today on access all areas. We'll be back again next CLEAN CLEAN CLEAN"

[xpilasneo4j]

Seems like a size issue: the prompt+transcript is too big: it I truncate the transcript to 21k characters, it fails less

[xpilasneo4j]

I added a loop to reduce by 500 characters each try and then it can load ALL the files in few minutes. Let me know if you want the fix.
I also created a script to run the commands for creating py38, instead of allowing people to run it manually, that helps to reduce the risk of issues when the setup is happening