ask models to estimate work in lines of code. tell the model a single well instructed subagent can easily write 4k clean loc in about 30 minutes. this will help your model estimate actual effort.
fable wants me to report a skill issue.
"Title:
Ran `git checkout ` to undo own small edit, discarding a large uncommitted change in the same file"
" - Cause: I treated git checkout on the file as undoing my edit without checking that the file had other uncommitted changes"
the model yearns for training











