The code:

• cpython/Lib/csv.py

  Line 473 in 719e5c3

  if checked > 20:

This is an off-by-one.

The docs:

• https://docs.python.org/3/library/csv.html#csv.Sniffer.has_header

• cpython/Doc/library/csv.rst

  Line 297 in 379161d

  Twenty rows after the first row are sampled; if more than half of columns +

Twenty rows after the first row are sampled

I err on correcting the docs, instead of the code, given that users may rely on the current behavior.

Linked PRs

• gh-137952: update csv.Sniffer().has_header() docs to describe the actual off-by-onish behavior #137953