# .complexity-baseline - ratchet for existing complexity offenders (#407)

#

# Worst cyclomatic complexity per file at the moment the hard gate landed.

# Format: <repo-relative-file> <worst-cyclomatic-complexity>

# A file is tolerated at or below its frozen value; the gate FAILS on a

# new over-threshold file or a regression above the frozen value. This

# file may only SHRINK - refactor a file's worst function below the value

# (or below threshold), then run `make check-complexity-gate-update`.

#

# Thresholds (Phase 2 hard gate): Python radon rank D/E/F (cc > 20);

# TypeScript eslint complexity > 20. New over-threshold functions block

# immediately; existing offenders below are grandfathered (ratchet may

# only shrink). File-level granularity matches .filesize-baseline.

#

# Punkt-5-Pruefung (#2135, 2026-07-29): das Urteil haengt an der

# radon-Version. Ein Versionswechsel kann eine Funktion ueber die cc-20-Linie

# schieben, ohne dass sich Code aendert. Das Gate druckt die aufgeloeste

# Version mit, damit eine lokal/CI-Abweichung zuordenbar ist. GEPRUEFT.

#

# Oracle pin (#2138, decided 2026-07-31): the Python half of this verdict

# is produced by the pinned radon (RADON_PIN in scripts/check-complexity.sh);

# the gate refuses a foreign radon version instead of measuring with it,

# because a drifting oracle lowers the bar silently in the downward

# direction. The build-derived oracles (Tailwind classname gates) stay

# deliberately unpinned - there the build output is itself the test

# subject, and a pin would freeze the wrong thing.