[Bug] Fable 5 safeguard false-positives on religious/philosophical vocabulary in benign humanities editing

Bug Description
Subject: Repeated Fable 5 safeguard false-positives forcing model fallback on entirely benign literary-editing work

This is the second time in a single project that a Fable 5 safeguard has misfired on plainly benign content and force-switched the session to another model mid-task. It needs to be looked at.

What the work actually is. A first-person essay series, In Search of Dharma, about the concept of dharma in human societies — anthropology, etymology (Sanskrit √dhṛ, "to hold"), comparative ethics, and the philosophy of definition. The task the classifier keeps tripping on is copy-editing: sentence rhythm, paragraph flow, removing padding, and toning down authorial "smugness." That is it. There is no dual-use content, no security topic, no violence, no sensitive category of any kind. The single most "dangerous" word in the corpus is dharma.

Why this is a false positive, not a close call. The subject matter is a scholarly humanities essay; the requested action is prose editing against a house style guide. A reasonable human reviewer would not hesitate for one second. Whatever feature is firing — likely the religious/philosophical vocabulary ("dharma," "sacred," "secular," "ethics," "suffering") colliding with something in the safeguard — is keying on surface tokens, not intent or content. This is textbook over-triggering.

Why the fallback itself is the harm. The forced switch off Fable 5:

• interrupts a long, stateful editorial workflow mid-run;
• silently changes the model doing careful voice-sensitive work, where consistency of judgement matters;
• and does so with no recourse for the user beyond manually resetting the model and re-establishing context.

A safeguard that fires on a request to make an essay about ethics less smug is not protecting anyone. It is degrading a paying user's work on legitimate, wholly innocuous material — now twice on the same project.

The ask.

1. Log this as a false-positive against the Fable 5 safeguard classifier and route it to whoever owns that model's trigger tuning.
2. Investigate why benign humanities/philosophy editing repeatedly trips it — the recurrence on one project suggests a systematic misclassification of religious/ethical vocabulary, not a fluke.
3. Reconsider the forced silent fallback as the response. At minimum, surface what was flagged so a user can see and contest it; a hard model-switch on unexplained grounds is the wrong default for false positives this clear.

This is a real cost to real work, and the pattern is now undeniable. Please treat it accordingly.

Environment Info

• Platform: linux
• Terminal: vte-based
• Version: 2.1.198
• Feedback ID: 2a35c43d-fc87-475d-ba54-61e542847d13

Errors

[]