coloristlisted

<!-- SPDX-License-Identifier: MIT --> # colorist An AI colorist that measures before it grades. Core discipline: **measure, decide, apply, verify.** Numbers from decoded pixels drive every decision; numeric gates decide whether output ships. You (the agent) choose intent and targets; deterministic solvers compute values; ffmpeg applies them; gates verify. You never judge color from a rendered image by eye, and never hand-tune a number a solver can compute. ## Setup preflight Run once per session: `ffmpeg` present with the `scale`, `lut3d`, and `scdet` filters (`ffmpeg -filters`) and the `ffv1` encoder (`ffmpeg -encoders`, since ffv1 is an encoder not a filter), Python 3.11 to 3.13 with the package installed (`python -m colorist --help`). Missing capability = say exactly what is missing and stop; no degraded guessing. ffmpeg and ffprobe are resolved from PATH, or from COLORIST_FFMPEG / COLORIST_FFPROBE if set. ## The commands (this is the whole CLI) `--preset` is required and has no default: the shipped presets are provisional (see the honesty note at the end). For `qc` and `consistency`, `--out` is a run DIRECTORY for the reports, masks, and any review PNGs, and existing reports are not overwritten. For `propose-cuts` and `confirm-cuts`, `--out` is a CSV FILE path. Two preset facts decide whether a run can PASS, so state them when a user writes their own. Hard gates are always required. A soft gate is required only when its `coverage` name appears in the preset's `requ