CARS

When Clients Stop Following: A Cognitive Conceptualization Diagram-driven Framework for Strategic Counseling

Venue: arXiv preprint (2026)
Paper: arXiv

Overview

CARS (Cognitive Alignment & Resistance Simulator) simulates a CBT client whose resistance emerges dynamically from Cognitive Conceptualization Diagrams (CCDs) rather than being scripted. It models when and why a client disengages from counselor guidance, providing a challenging test-bed for a therapist's exploration ability.

Key Features

• Dual CCD structure: A stable main CCD (relevant history, core/intermediate beliefs, coping strategies) plus a session-specific CCD (automatic thought, intermediate belief, resistance triggers).
• Dynamic resistance reasoning: Each turn checks whether the topic triggers a resistance schema and whether the counselor's strategy confirms or conflicts with the client's beliefs.
• Emotion state (0-100): Updated by a signed delta each turn; extremely low emotion ends the session.
• Style-modulated cooperation: A preferred counselor style and individual traits shape whether the client cooperates or resists.

How It Works

1. Profile Loading: Loads a client profile — persona, background, main + session CCDs, preferences, and characteristics.
2. Structured Reasoning: Each turn the model reasons over topic → schema trigger → strategy-consistency check → updated cognition, emotion, and interaction goal.
3. Response Generation: Produces a structured turn (step-by-step reasoning, signed emotion change, intention, nonverbal behavior, and the final utterance).
4. State Update: Clamps emotion to 0-100, carries the interaction goal into the next turn, and ends the session if emotion falls to/below the termination threshold.

Usage

CLI

patienthub simulate client=cars

Python

from patienthub.clients import get_client

client = get_client(agent_name="cars", lang="en")

response = client.generate_response("Let's set some treatment goals for today. How does that sound?")
print(response)

Configuration

Option	Description	Default
prompt_path	Path to prompt file	data/prompts/client/cars.yaml
data_path	Path to character file	data/characters/cars.json
data_idx	Character index	0
initial_emotion_state	Starting emotion (0-100)	50
termination_threshold	Emotion at/below which the session ends	10

Character Data Format

{
  "name": "Abe",
  "persona": {
    "name": "Abe",
    "age": "42 years old",
    "gender": "Male",
    "occupation": "Recently unemployed (former manager).",
    "family_background": "Father left when Abe was 11; his overburdened mother was critical when he could not meet her unrealistic expectations.",
    "interpersonal_relationships": "Recently divorced; reluctant to lean on his adult son or ask anyone for help.",
    "physical_condition": "No significant medical problems; reports low energy and disrupted sleep under stress.",
    "lifestyle": "Unemployed with little daily structure; avoids tasks that feel like tests of competence.",
    "chief_complaint": "Feels incompetent and like a failure after losing his job and his marriage; avoids challenges and asking for help."
  },
  "background": "The current counseling topic is treatment goal setting ...",
  "main_ccd": {
    "name": "Abe",
    "relevant_history": "Abe's father left when he was 11; his critical mother had unrealistic expectations. He recently lost his job and went through a divorce.",
    "core_beliefs": ["I am incompetent.", "I am a failure."],
    "intermediate_beliefs": [
      "If I try hard things I'll fail, so it's safer to avoid challenges.",
      "If I ask for help, people will see how incompetent I am."
    ],
    "coping_strategies": ["Avoids asking for help.", "Avoids challenges and difficult tasks."]
  },
  "session_ccd": {
    "theme": "goal setting",
    "structured_cbt_strategy": "Goal list or agenda setting",
    "automatic_thought": {
      "situation": "...",
      "cognition": "...",
      "reaction": "..."
    },
    "intermediate_belief": {
      "attitude": "...",
      "rule": "...",
      "assumption": "..."
    },
    "resistance_triggers": ["Authoritative or commanding goal-setting"]
  },
  "preferences": { "positive": ["..."], "negative": ["..."] },
  "client_characteristics": {
    "possible_responses_under_different_emotions": ["..."],
    "other_client_characteristics": ["..."]
  }
}

CARS characters can be authored by hand or produced by the CARS generator.

Emotion-Utterance Corpus

The paper's response step (§3.1.2) selects sentence patterns from a "pre-defined emotion-utterance corpus." The authors do not publish that corpus, so PatientHub does not ship one and it is not wired into the CARS prompt — the client generates its utterance directly from its reasoning and current emotion. This is a deliberate omission to avoid fabricating content the paper does not provide.