# Held-out split (issue #139 / H-1).
#
# These task names form the subset a harness PR must *never* iterate
# against. Authors tune against the train split (`run_bench.py
# --no-heldout`) and see the held-out result reported separately by the
# nightly — so a change cannot be overfit to the evaluation set.
#
# Lines starting with '#' and blank lines are ignored. One task name per
# line. The runner skips these under --no-heldout; the nightly reports
# their pass rate as its own line.

# A deliberate, diverse trio (one per capability axis) so the held-out
# subset is representative, not the easiest or hardest tasks.
count-defs-in-tree
parse-noisy-config
rename-symbol
