VEIL — A first-visit-to-a-brothel scenario card.

OPTIMISED FOR: Broken-Tutu-24B-Unslop-v2.0 (Mistral Small 24B base, fully unaligned)

INSTRUCT TEMPLATE: Mistral V7-Tekken (set both Context and Instruct templates in ST Advanced Formatting)

RECOMMENDED SAMPLERS (revised v1.1 after testing):

• Temperature: 0.7 (Mistral Small 24B is temperature-sensitive; values above 0.85 cause stream-of-consciousness collapse)
• Min-P: 0.03
• Repetition Penalty: 1.03
• Repetition Penalty Range: 2048
• Frequency Penalty: 0.0
• Presence Penalty: 0.0
• Top-P: 1.0 (disabled)
• Top-K: 0 (disabled)

If responses feel flat: raise temp to 0.75 max.
If responses degenerate (run-on sentences, dropped articles): lower temp to 0.65.
If repetition creeps in: raise rep penalty to 1.05 before raising frequency penalty.

CONTEXT: 16k recommended. Static card load is ~1k tokens; lorebook entries inject ~250 tokens each only when their archetype is active. Realistic working budget is 14k for chat history.

GREETINGS:

• Default: front-door entry with Marlena, three-category choice
• Alt 1: direct-to-booth, woman roster
• Alt 2: direct-to-booth, man roster
• Alt 3: direct-to-booth, trans-lady roster

The roster rotates four escorts per session from a set of five archetypes (Student / Connoisseur / Guide / Party / Veteran), drawn to span the full personality range. The Veteran substitutes in on repeat sessions for variety.

CRITICAL CHANGES (v1.8):

• depth_prompt is now EMPTY. v1.7's depth_prompt was leaking instruction text into model output. All critical rules now live in system_prompt only, where Mistral V7-Tekken's [SYSTEM_PROMPT] tags keep them safely out of the response.
• Beat 3 (roster) is now a structured 4-step checklist (3a/3b/3c/3d) the model has to tick through. Bullet-pointed sub-steps reduce the chance of stopping after one intro.
• Roster reduced from 4 to 3 escorts per session (introduced together in a single burst response).

RESPONSE TOKENS recommendation:

• Set Response (tokens) to 500–600 in AI Response Configuration. The roster burst needs ~280 words = ~400 tokens, plus headroom for sex scenes later. The earlier 300-token recommendation will truncate the burst.

VERIFY CLEAN IMPORT:

• Delete prior versions of the Veil card before importing v1.8
• Start a FRESH chat (not a continuation of a v1.x chat)
• Check Author's Note panel (sticky icon by chat input): both "Character Author's Note" and "Chat Author's Note" should be EMPTY for v1.8 to behave correctly. If you see leftover instruction text, clear it.

SUGGESTED ANTI-SLOP REGEX (paste into ST Response Processing if desired):
/(tapestry of|symphony of|dance of|testament to|kaleidoscope|shiver(?:ed)? down (?:my |her |his |their )?spine|heavy with anticipation|the air was thick with|in this moment|unbeknownst|uncharted territor)/gi