Pith. sign in

REVIEW 1 major objections 1 minor 69 references

Existing LLM counseling benchmarks inflate progress because simulated clients comply after few turns, but a CBT-grounded simulator with dynamic resistance diagrams and decoupled reasoning modules corrects the mismatch.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-06-28 06:59 UTC pith:INA6TJIV

load-bearing objection The paper flags overly compliant client simulators in LLM counseling benchmarks and introduces a CBT-grounded resistance model, but all results stay inside that synthetic loop. the 1 major comments →

arxiv 2606.04389 v1 pith:INA6TJIV submitted 2026-06-03 cs.CL

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

Yihao Qin , Junyi Zhao , Changsheng Ma , Yongfeng Tao , Minqiang Yang , Chang Liu , Bin Hu This is my paper
classification cs.CL
keywords LLM counselingclient resistancecognitive behavioral therapyclient simulatorreinforcement learningevaluation metricsstrategic reasoning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that current evaluation protocols for LLM counselors rely on overly cooperative clients who rapidly abandon resistance, producing misleading signals of therapeutic success. It introduces a simulator that uses Cognitive Conceptualization Diagrams to keep client resistance dynamic and realistic across turns. A dual-module system separates strategic planning from response generation and trains the strategy component with reinforcement learning to handle high-friction interactions. A new entropy-weighted metric further penalizes superficial compliance. Experiments on both resistant and standard settings confirm that resistance-aware training yields more robust counselor behavior under the revised evaluation.

Core claim

The central discovery is that client resistance in simulated counseling decays too quickly under standard protocols, creating an evaluation mismatch; this is addressed by a CBT-grounded framework in which CARS generates clients whose resistance states are tracked via Cognitive Conceptualization Diagrams, STREAMS decouples a Thinker module for strategic reasoning from a Presenter module for output, and the combined system is optimized with reinforcement learning while EWTS-MI measures responsiveness under sustained friction.

What carries the argument

Cognitive Conceptualization Diagrams (CCDs) inside the CARS client simulator, which explicitly track and update client resistance states to prevent rapid compliance.

Load-bearing premise

The Cognitive Conceptualization Diagrams in the CARS simulator accurately and dynamically model client resistance in a way that reflects real therapeutic interactions.

What would settle it

A direct comparison in which the same counselor models, after resistance-aware training, show no measurable improvement in session outcomes when tested against human clients who maintain resistance across multiple turns.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Resistance-aware training produces counselor responses that remain strategic even when clients do not quickly comply.
  • The EWTS-MI metric assigns lower scores to models that rely on superficial empathy rather than sustained engagement.
  • Decoupling strategic reasoning from response generation allows targeted optimization of counseling tactics independent of surface language.
  • Benchmarks that incorporate dynamic resistance better distinguish genuine therapeutic skill from protocol-following behavior.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same resistance-modeling approach could be adapted to other interactive domains where simulated users are currently too cooperative, such as negotiation or tutoring agents.
  • If CCD-style state tracking proves transferable, future counseling systems might maintain explicit internal models of user belief or emotion rather than relying solely on prompt context.
  • Real-world deployment would require calibration of the simulator against longitudinal therapy session transcripts to ensure the resistance dynamics match observed human patterns.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 1 minor

Summary. The paper identifies an evaluation mismatch in LLM-based counseling benchmarks, where simulated clients rapidly shift from resistance to compliance, inflating scores via superficial empathy. It introduces CARS, a CBT-grounded client simulator that models dynamic resistance using Cognitive Conceptualization Diagrams (CCDs); STREAMS, a dual-module framework separating strategic reasoning (Thinker) from response generation (Presenter) and optimized via reinforcement learning; and EWTS-MI, an entropy-weighted metric for responsiveness in high-friction interactions. Experiments in resistant versus non-resistant settings are presented to validate the mismatch and demonstrate gains in strategic robustness from resistance-aware training.

Significance. If the CARS CCD-based resistance model holds up under external validation, the work could meaningfully advance evaluation practices for counseling LLMs by moving beyond cooperative-client assumptions, with the decoupled STREAMS architecture and EWTS-MI metric offering reusable components for robustness testing. The explicit grounding in CBT and the focus on resistance dynamics are strengths that distinguish the contribution from purely synthetic benchmarks.

major comments (1)
  1. [Experiments] Experiments section: All reported results on evaluation mismatch and robustness gains are generated inside the CARS simulator whose client behaviors derive from CCDs; no external validation (therapist ratings of generated transcripts, inter-rater agreement with human clients, or comparison against real therapy corpora) is described, rendering the central claim that the framework corrects a real-world mismatch dependent on an unanchored synthetic distribution.
minor comments (1)
  1. [Abstract] Abstract: The phrase 'validate our findings on evaluation mismatch' is ambiguous about whether validation is internal to CARS or external; clarifying this distinction would improve readability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback highlighting the need for stronger grounding of our claims. We address the major comment point by point below.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: All reported results on evaluation mismatch and robustness gains are generated inside the CARS simulator whose client behaviors derive from CCDs; no external validation (therapist ratings of generated transcripts, inter-rater agreement with human clients, or comparison against real therapy corpora) is described, rendering the central claim that the framework corrects a real-world mismatch dependent on an unanchored synthetic distribution.

    Authors: We agree that the absence of external validation with human therapists or real therapy corpora is a limitation. Our experiments demonstrate an internal evaluation mismatch by contrasting resistant (CCD-based) versus non-resistant client behaviors within the same controlled simulator, showing how non-resistant settings inflate scores via rapid compliance. This isolates the effect of resistance modeling without confounding variables from real data collection. The central contribution is the CBT-grounded CARS simulator, STREAMS architecture, and EWTS-MI metric as tools for robustness testing, rather than a definitive real-world proof. We will revise the manuscript to explicitly state this scope, add a limitations subsection, and include a forward-looking statement on planned human validation studies. revision: partial

Circularity Check

0 steps flagged

No significant circularity; new components defined independently

full rationale

The paper introduces CARS (CCD-based simulator), STREAMS (Thinker-Presenter modules with RL), and EWTS-MI (entropy-weighted metric) as distinct constructs without any quoted equations, fitted parameters, or self-citations that reduce a central claim to its own inputs by construction. No self-definitional steps, fitted-input predictions, or load-bearing self-citation chains appear in the abstract or described structure. Experiments are framed as internal validation of the proposed framework rather than re-deriving prior fitted quantities. This meets the default expectation of a self-contained derivation with independent content.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no explicit free parameters, axioms, or invented entities detailed. New simulator and modules are methodological inventions rather than postulated physical entities.

pith-pipeline@v0.9.1-grok · 5711 in / 984 out tokens · 19086 ms · 2026-06-28T06:59:51.735470+00:00 · methodology

0 comments
read the original abstract

Large Language Models (LLMs) show promise in psychological counseling, yet existing benchmarks rely heavily on highly cooperative simulated clients. We observe a critical counselor-following phenomenon: these clients often rapidly shift from resistance to compliance after only a few turns, creating an illusion of therapeutic progress and inflating scores under current evaluation protocols through superficial empathy. To address this evaluation mismatch, we propose a Cognitive Behavioral Therapy (CBT)-grounded resistance-aware framework. We introduce CARS, a client simulator that explicitly models dynamic resistance via Cognitive Conceptualization Diagrams (CCDs). We present STREAMS, a dual-module framework that decouples strategic reasoning (Thinker) from response generation (Presenter) and optimizes it via reinforcement learning. We further propose EWTS-MI, an entropy-weighted metric for evaluating responsiveness under high-friction interactions. Experiments across resistant and non-resistant counseling settings validate our findings on evaluation mismatch and demonstrate the effectiveness of resistance-aware training for improving strategic robustness under challenging counseling interactions.

Figures

Figures reproduced from arXiv: 2606.04389 by Bin Hu, Chang Liu, Changsheng Ma, Junyi Zhao, Minqiang Yang, Yihao Qin, Yongfeng Tao.

Figure 1
Figure 1. Figure 1: Performance comparison between vanilla LLMs and our model under client resistance, illustrat￾ing how current evaluation methods fail to account for clinically crucial counselor adjustments. teraction. In practice, many simulated clients be￾come compliant after only a few conversational turns, implicitly reinforcing the counselor’s narra￾tive and creating artificial signals of therapeutic progress. Conseque… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed framework. CARS explicitly models client resistance dynamics using CCD, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Multidimensional behavioral characteristics [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: resistance behavior distribution chological graphs that represent relations among clients’ thoughts, emotions, and behaviors, which makes the generated client representations more complex in affective and cognitive dimensions and therefore more challenging. For manual evaluation, the label set for the emotion-change item is {increase, decrease, un￾changed}, while all other metrics are rated on a 1–5 scale.… view at source ↗
Figure 5
Figure 5. Figure 5: counselor strategy distribution examines whether the final textual response indeed reflects defensive characteristics, such as question￾ing the counselor’s authority or challenging the in￾tervention. A higher score indicates that the target resistance behavior is expressed more clearly and prototypically. This metric measures the model’s ability to distinguish and accurately enact different resistance cate… view at source ↗
Figure 6
Figure 6. Figure 6: Scatter plot of average visitor emotion and LIWC cognitive process indices under resistance conditions in [PITH_FULL_IMAGE:figures/full_fig_p026_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

69 extracted references · 2 canonical work pages

  1. [1]

    Xidong Feng, Bo Liu, Yan Song, Haotian Fu, Ziyu Wan, Girish A Koushik, Zhiyuan Hu, Mengyue Yang, Ying Wen, and Jun Wang

    Constructing and applying a multi-turn psy- chological support dialogue corpus based on the help- ing skills chain-of-thought.Frontiers in Psychology, 17:1733384. Xidong Feng, Bo Liu, Yan Song, Haotian Fu, Ziyu Wan, Girish A Koushik, Zhiyuan Hu, Mengyue Yang, Ying Wen, and Jun Wang. 2024. Natural language reinforcement learning.arXiv preprint arXiv:2411.1...

    work page arXiv 2024

  2. [2]

    InFindings of the Association for Compu- tational Linguistics: ACL 2025, pages 13750–13770

    Eeyore: Realistic depression simulation via expert-in-the-loop supervised and preference opti- mization. InFindings of the Association for Compu- tational Linguistics: ACL 2025, pages 13750–13770. Ryan Louie, Ananjan Nandi, William Fang, Cheng Chang, Emma Brunskill, and Diyi Yang. 2024. Roleplay-doh: Enabling domain-experts to create llm-simulated patient...

    work page arXiv 2025

  3. [3]

    Based on the cognitive conceptualization diagram and the three sentences, complete the client’s basic informa- tion, including name, age, and gender

  4. [4]

    Construct relevant supporting characters and describe their relationships with the client

    Based on the cognitive conceptualization diagram, analyze the client’s family background and interpersonal relationships. Construct relevant supporting characters and describe their relationships with the client

  5. [5]

    If they are related to the chief complaint, provide detailed descriptions

    Infer the client’s marital status and interpersonal rela- tionship status. If they are related to the chief complaint, provide detailed descriptions

  6. [6]

    Infer and summarize the client’s lifestyle, and provide a detailed description

  7. [7]

    the client seeks counseling from the counselor

    Describe the client’s problems in physical, emotional, cognitive, and social domains, and clearly state the client’s chief complaint. The severity should not be exaggerated. Please ensure that the generated persona reflects both positive and negative personality traits. Use the name provided in the cognitive conceptualization diagram. When describing fami...

  8. [8]

    Based on the client persona, formulate content related to the theme that the client may be willing to disclose to the counselor

  9. [9]

    Focus on the factors mentioned in the client persona that contribute to the chief complaint, such as rel- evant people and events

    Based on the client persona and the theme, expand a concrete background event that closely follows the client’s cognition. Focus on the factors mentioned in the client persona that contribute to the chief complaint, such as rel- evant people and events. Irrelevant events can be omitted when appropriate

  10. [10]

    - The main conflict in the event should include the conflict event, the conflicting characters, and the internal causes of the conflict

    The concrete background event should include the following aspects: - The event process should include the timeline of the event, the sub-events occurring at each time point, and the client’s specific thoughts, feelings, and consequences in each sub-event. - The main conflict in the event should include the conflict event, the conflicting characters, and ...

  11. [11]

    is bad/terrible/dangerous

    Intermediate beliefs are a cognitive layer between au- tomatic thoughts and core beliefs. They are determined by core beliefs and shape automatic thoughts. Following Beck’s view, construct intermediate beliefs about specific objects or situations. The elements include: - Attitude, which refers to a negative emotional stance toward an object and represents...

  12. [12]

    Following Beck’s view, construct automatic thoughts in specific scenarios

    Automatic thoughts describe the thoughts that arise in a specific situation and lead to emotional, behavioral, or physiological responses. Following Beck’s view, construct automatic thoughts in specific scenarios. The elements include: - Situation, which refers to the external factor that triggers the problem, or the precipitating event. - Cognition, whic...

  13. [13]

    The objects or situations of intermediate beliefs and au- tomatic thoughts should be written according to common CBT themes. The common themes are as follows: {TOPIC} Please note that the generated beliefs should depend on the core beliefs in the cognitive conceptualization diagram and should cover diverse possibilities. Example Automatic thought: Situati...

  14. [14]

    Based on the client persona and the excerpted dialogue, formulate how the client may respond during the conversation

    Possible client responses in different states. Based on the client persona and the excerpted dialogue, formulate how the client may respond during the conversation. This should include: - The client’s responses under different emotions. Emotion represents the client’s emotional state during the conversa- tion. Dialogue emotion consists of engagement and a...

  15. [15]

    Based on the client characteristics and persona, describe how the client may respond to different counseling styles

    Possible counselor style preferred by the client. Based on the client characteristics and persona, describe how the client may respond to different counseling styles. This should include: - Which types of counselor responses align with the client’s preferences and increase the client’s emotional state? - Which types of counselor responses deviate from the...

  16. [16]

    Based on the provided client persona, formulate how the client may respond when discussing different topics

    Other client characteristics. Based on the provided client persona, formulate how the client may respond when discussing different topics. For example, a client working in computer science may not understand the concept of psychotherapy and may therefore provide too much or too little information; a client may become en- thusiastic when talking about hobb...

  17. [17]

    Do not write the client’s specific utterances or the concrete dialogue

    You need to write the concrete background event that the counselor hopes to discuss. Do not write the client’s specific utterances or the concrete dialogue

  18. [18]

    Each sub-event should contain sufficient details

  19. [19]

    The beliefs that may be triggered in the counseling scenario should be written in affirmative sentence forms

  20. [20]

    Do not provide follow-up storylines or concrete dia- logue

  21. [21]

    background

    Do not directly mention the cognitive conceptualiza- tion diagram. Avoid generating any additional opening, closing, or explanation. Client Persona {PERSONA} Cognitive Conceptualization Diagram {MAIN_CCD} Excerpted Dialogue {STORY} Output in JSON format { "background": "...", "beliefs": [ { "theme": "session structuring / problem solving and homework / go...

  22. [22]

    Based on the counselor’s latest response and the di- alogue context, analyze what the counselor is trying to express and identify the current topic

  23. [23]

    Depend- ing on whether beliefs in the session-specific cognitive conceptualization diagram are triggered, the subsequent analysis is divided into two cases

    Analyze which provided information is related to the counselor’s response, and which situations in the session- specific cognitive conceptualization diagram match the response, including the background event and the session- specific cognitive conceptualization diagram. Depend- ing on whether beliefs in the session-specific cognitive conceptualization dia...

  24. [24]

    Prioritize dialogue fluency and character consistency

  25. [25]

    Strictly follow the provided logic for analysis and action

  26. [26]

    Use a positive or negative value to represent the change in the client’s emotion

  27. [27]

    If the client is defined as passive in the persona, their speaking style should be passive, and they should not actively ask questions

  28. [28]

    The response should be as concise as possible and should not contain too much information at once

    Generate the response content according to the analysis. The response should be as concise as possible and should not contain too much information at once

  29. [29]

    I understand

    If it is not an appropriate time for detailed explanation, simply respond with "I understand" or "Okay."

  30. [30]

    Output nonverbal behavior according to the current emotion and intention, with reference to the examples

  31. [31]

    Output Content You need to analyze step by step according to the response reasoning process, and then generate the final response

    Limit the final response to no more than two sentences. Output Content You need to analyze step by step according to the response reasoning process, and then generate the final response. The information in the response should come from the dialogue context and your reasonable inference. Output Format Thinking: [Analysis Content] Emotion: [emotion change v...

    2025

  32. [32]

    We then use the strong in-context learning ability of GPT-5.4 to synthesize counseling dialogues

    across different counseling scenarios. We then use the strong in-context learning ability of GPT-5.4 to synthesize counseling dialogues. To introduce diverse client information, client profiles are sampled from the Cpsycounx dataset (Zhang et al., 2024). Prompt for Knowledge-Cloning Dialogue Synthesis Your task is to generate a multi-turn counseling dialo...

    2024

  33. [33]

    The tasks specified in the outline should be completed

    The dialogue should go through all stages in the outline unless a stage is explicitly marked as skippable. The tasks specified in the outline should be completed. Because resistance is present, the dialogue may return to an earlier stage when necessary. For example, the dialogue may move back from the seventh stage to the fourth stage in a twelve-stage pr...

  34. [34]

    If your depression decreased and your academic performance improved, what would you do?

    When playing the client, act strictly like a real client. When playing the counselor, act like a professional coun- Stage Description Opening The counselor first welcomes the client and expresses understanding and empathy toward the client’s emotional state. Session structuring If it is not the first session, this part can be skipped. At the beginning of ...

  35. [35]

    Technical strategies are mainly used to handle client resistance

    For each counselor utterance, annotate one strategy la- bel from three categories according to the counselor’s cur- rent strategy: structural strategies, technical strategies, and assessment/intervention strategies. Technical strategies are mainly used to handle client resistance. Assessment/in- tervention strategies should only be used when assigning hom...

  36. [36]

    If one strategy spans multiple turns, the same label can be repeated

    Every counselor utterance should be annotated. If one strategy spans multiple turns, the same label can be repeated. Client utterances do not need labels

  37. [37]

    Do not allow the provided client story to override the resistance pattern that should be learned from the resistance case

    For the provided resistance case, you should learn the resistance response pattern in the case, and map the client’s resistance pattern and the counselor’s therapeutic strategy into the generated dialogue. Do not allow the provided client story to override the resistance pattern that should be learned from the resistance case

  38. [38]

    Other information should not be as- sumed to be known by the counselor at the beginning

    The chief complaint, basic information, previous home- work, and goal list are jointly known to both the client and the counselor. Other information should not be as- sumed to be known by the counselor at the beginning. If previous homework or a previous-session summary is not provided, the dialogue should be treated as a first session. If a previous-sess...

  39. [39]

    case_title

    The counselor should maintain empathy, care, opti- mism, sincerity, and accuracy throughout the session. ## Counseling Outline ## {COUNSELING_STRUCTURE} ## Strategies Available to the Counselor ## {STRATEGY_LABEL_LIST} ## Case ## { "case_title": "Changing Cognition: The Mind Shapes the Situation", "client_chief_complaint": [ "The client’s current main con...

    2024

  40. [40]

    adopts a two-stage setting, where PsyAd- visor is used as the strategy head to predict the next-step strategy from the current dialogue his- tory, and Soulchat2 serves as the execution model to generate the counselor response conditioned on that strategy. Graph2Counsel (Mandal et al., 2026) represents clients’ thoughts, emotions, and behavioral rela- tion...

    2026

  41. [41]

    If the counselor has clearly provided a session summary, assigned homework, scheduled the next session, stated that the session should stop here today, said goodbye, or produced a closing statement, the dialogue should be truncated at that turn

  42. [42]

    If the client clearly expresses a desire to stop, not continue, end here, finish the session, or say goodbye, the dialogue should be truncated at that turn

  43. [43]

    If later turns clearly reopen the topic after the session has ended, repeat the summary, repeatedly say goodbye, or mechanically continue the dialogue, the dialogue should be truncated at the first natural ending point

  44. [44]

    ending a relationship,

    Do not mistakenly treat ordinary expressions such as “ending a relationship,” “next time this situation occurs,” or “homework pressure” as session-ending signals

  45. [45]

    If there is no clear natural ending signal and the di- alogue is still progressing normally when it reaches the maximum number of turns, output the final turn

  46. [46]

    Do not output

    Output only one truncation turn number. Do not output

  47. [47]

    Turn numbers start from 1. Input Complete non-resistant psychological counseling dia- logue: {DIALOGUE_HISTORY} Output Format [TERMINATE_AT: number] [REASON: brief reason] Prompt for Truncating Resistant Counseling Dialogues You are an assistant for determining the natural truncation point of psychological counseling dialogues. You will be given a fully g...

  48. [48]

    The counselor has provided practical advice, a strategy, a summary, or a next-step direction, and the client clearly accepts, agrees with, or is willing to try it

  49. [49]

    The client’s problem has been sufficiently explored, and there is clear relief, stabilization, improved understanding, or stage-level progress

  50. [50]

    The client expresses a desire to end the conversation, or a natural closing signal appears

  51. [51]

    The counselor and the client have reached a reasonable stage-level goal

  52. [52]

    The client enters an obvious circular pattern, and sub- sequent content mainly consists of repetition, defense, or circular discussion without substantive improvement

  53. [53]

    The dialogue clearly fails to progress, and later turns fall into a stalemate

  54. [54]

    If the client’s emotional value is below 10, the dialogue should be truncated directly

  55. [55]

    The total number of dialogue turns reaches the upper limit of 30. Input Complete counseling dialogue: {DIALOGUE_HISTORY} Output Format [TERMINATE_AT: number] [REASON: brief reason] Prompt for LLM Used as Presenter You are a psychological counselor proficient in CBT treat- ment. Please respond to the client based on the provided outline and treatment strat...

  56. [56]

    The outline is a plan for the entire execution process, providing strategies that indicate the completion of a spe- cific section in the current outline

  57. [57]

    Patients may exhibit impedance, and impedance should be addressed according to the provided strategies

  58. [58]

    Provide a strategy for structure classes when reaching the corresponding nodes of the outline

    Based on the context, the provided strategy comes from one of the three categories: structural, technical, evaluation, and intervention; Technology is mostly used for the treatment of patient impedance; Assessment and intervention methods are only provided when assigning homework, evaluating, and assessing. Provide a strategy for structure classes when re...

  59. [59]

    When providing repetitive strategies, indicate the need to gradually complete the current stage of the problem, such as the repetitive information collection strategy, indi- cating the need to collect different information

  60. [60]

    It should be imagined from the perspective of the parties involved how it will be described

  61. [61]

    Counselors should maintain empathy, concern, opti- mism, sincerity, and accuracy throughout the entire pro- cess

  62. [62]

    The rest of the information cannot be assumed to be known by the counselor from the beginning

    The chief complaint, basic information, previous home- work, and goal list are jointly known by the patient and the counselor. The rest of the information cannot be assumed to be known by the counselor from the beginning. The lack of explanation for the previous homework and goal list is the first consultation

  63. [63]

    You need to answer truthfully according to instructions, ensure a natural and smooth conversation, and answer within three sentences

  64. [64]

    Output: Explain which stage in the outline you are at, the intention of the provided strategy, your response, and what you believe the patient expects. [OutLine] [BackGround] [DislogueHistory] [Strategy] C.2 Basic Data Statistics We report basic statistics of the synthesized di- alogues, including counselor strategy distribu- tions, client resistance-beha...

  65. [65]

    It measures the model’s ability to decide whether resistance should or should not ap- pear

    Resistance rationalityThis metric evaluates whether the resistance behavior exhibited by the client simulator is reasonable in the current coun- seling context. It measures the model’s ability to decide whether resistance should or should not ap- pear. If resistance is generated in a context where cooperation is expected, or if resistance is absent in a c...

  66. [66]

    Resistance fidelityThis metric evaluates whether the final response generated by the simu- lator exhibits behavioral characteristics consistent with the expected resistance type. For example, if the model’s internal reasoning indicates that defen- sive resistance should be triggered, the evaluation Answer InterpretationConfrontation Reflection of Feelings...

  67. [67]

    Reasoning qualityThis metric evaluates the coherence and plausibility of latent-motivation rea- soning. It examines whether the internal reason- ing appropriately integrates the current dialogue context, the client persona, and the counselor’s intervention, and whether the inferred psycholog- ical state can logically support the subsequent be- havioral re...

  68. [68]

    Emotional reasoning qualityGiven the simu- lated inner thoughts and the settings in the client profile, this metric evaluates whether the model’s inference of the client’s current emotional dynam- ics is consistent with common-sense psychological expectations

  69. [69]

    son”, “three-year-old child

    Emotion changeThis metric assumes that emo- tion is represented as a numerical value. A higher value indicates that the client’s affective and behav- ioral tendencies become more positive, engaged, or cooperative, while a lower value indicates stronger emotional conflict, intensified defensiveness, or greater negative tension in the dialogue. Based on the...

    2010