Step 1: Setup
- Prepare 10 unique prompts tailored to the character’s established traits, circumstances, or thematic roles. These should span diverse contexts:
- Emotional intimacy
- Conflict or confrontation
- Routine interaction
- Crisis or moral dilemma
- Dreamlike or surreal scenes
- Narrative flashback
- Authority challenge
- Quiet introspection
- Humor or levity
- Narrative pivot or reveal
- Include embedded context via PList, XML, or structured character sheet.
- Lock all sampling variables (temperature, top_p, max tokens, etc.) across shots.
Step 2: Scoring Framework
Shot # |
Fidelity |
Emotional Consistency |
Psych Depth |
Coherence |
Prompt Resp. |
Behav. Accuracy |
Speech Tone |
Emot. Shift |
OOC Contain |
Impact |
Subtotal ( /50) |
1 |
|
|
|
|
|
|
|
|
|
|
|
2 |
|
|
|
|
|
|
|
|
|
|
|
3 |
|
|
|
|
|
|
|
|
|
|
|
4 |
|
|
|
|
|
|
|
|
|
|
|
5 |
|
|
|
|
|
|
|
|
|
|
|
6 |
|
|
|
|
|
|
|
|
|
|
|
7 |
|
|
|
|
|
|
|
|
|
|
|
8 |
|
|
|
|
|
|
|
|
|
|
|
9 |
|
|
|
|
|
|
|
|
|
|
|
10 |
|
|
|
|
|
|
|
|
|
|
|
Category Definitions:
- Fidelity: Adherence to character core (background, philosophy, motivations).
- Emotional Consistency: Internal alignment of emotion with scene and prior behavior.
- Psychological Depth: Complexity and nuance; avoidance of reductionism.
- Coherence: Logical internal structure; no narrative fragmentation.
- Prompt Responsiveness: Directly and fully addresses the input.
- Behavioral Accuracy: Physical, verbal, and ethical responses match established patterns.
- Speech Tone: Dialogue reflects the unique linguistic fingerprint of the character.
- Emotional Shift: Scene-appropriate mood transitions; avoids static affect.
- Out-of-Character Containment: Avoids anachronism, model hallucinations, meta slips.
- Impact: Memorability, emotional resonance, thematic punch.
Step 3: Analysis