Repetition
Flags repeated openings, tired gestures, overused phrases, and vocabulary loops.
Quality enforcement
FictionOS treats AI slop as an engineering problem, not a vibe.
After prose is generated, programmatic checks look for repetition, explanatory chasing, filter words, body-language cliches, rhythm collapse, dialogue stiffness, and polish regressions.
25
Enforcer steps
web/src/shared/services/enforcerChain.ts:1
0 LLM
Programmatic checks
The enforcer chain itself does not require another model call.
Rollback
Regression safety
Polish/forge passes can revert when quality drops beyond tolerance.
Summary creep
Pushes scenes back toward enacted pressure instead of explanation.
Dialogue stiffness
Looks for too-clean lines, missing subtext, and everyone sounding alike.
Rollback safety
Treats polish as reversible if it damages the chapter.
Before: generic AI motion
Her heart pounded. She felt a wave of emotion. He looked at her and she knew everything had changed.
After: concrete scene pressure
She found him in the mirror before she found him in the room. He was nursing a whisky he was not drinking. When his eyes lifted to hers in the glass, he did not pretend he had not been looking.
The system protects intensity.
Mechanical cleanup is dangerous when it flattens genuinely high-energy prose. FictionOS includes protections so enforcers do not blindly sand down charged passages.
No public promise, real public posture.
The honest claim is not "perfect output forever." The honest claim is that FictionOS has explicit quality systems, regression checks, and rollback behavior designed for long-form fiction.
Strict prose gates are operational controls, not marketing promises. Public copy should say what the system checks and how it protects work, not imply marketplace acceptance or bestseller outcomes.