mechanical-preprocess

Automating manuscript cleanup before AI editing requires deterministic, rule-based text scanning to fix mechanical artifacts like em-dashes and forbidden words chapter-by-chapter. This approach generates diffs for every change, preserving meaning and preventing attention drift in long-form writing pipelines.

Checking adverb density and forbidden words across book chapters is done through a structured mechanical scan that quantifies repetition signals and formulaic constructions. This deterministic scan produces a report with counts and flags for each chapter to highlight issues before higher-level rewriting.

Safely normalizing em-dashes and whitespace in a manuscript is achieved by applying controlled bash rules that only execute 'always safe' replacements. It generates diffs for every mechanical change and routes ambiguous fixes to a diff-only AI approval loop to preserve voice.

Running deterministic text cleanup on a long-form manuscript with bash scripts is fully supported. This bash-first approach targets context overflow and repetitive artifacts by scanning chapter files alongside voice policy and genre context to safely apply mechanical replacements.

AI manuscript edits drift in long chapters due to context overflow and repetitive mechanical artifacts degrading consistency. Handling mechanical cleanup deterministically before the AI rewriting pass prevents this drift by reducing the noise the AI agent processes.

Voice policy and genre context files are required to preprocess a book manuscript. The cleanup process needs chapter files plus a voice DNA file and foundation context to ensure safe, rule-based replacements preserve the author's intended meaning and style.