Personal Baseline is more suitable for a long-term Human Model than population averages. Individual behavior should be modeled relative to a person's own historical baseline, not a generic mean.
个人基线比群体平均值更适合长期 Human Model。个体行为应相对于其自身历史基线建模，而非通用均值。
The speed at which an AI adapts to a user may affect trust formation, independent of absolute answer quality.
AI 适应用户的速度可能影响信任形成，与绝对回答质量无关。
A Human Model should jointly model Trait + State, not personality alone. Static trait profiles miss context-driven behavioral variation.
Human Model 应同时建模 Trait（特质）+ State（状态），而非仅人格。静态人格画像会遗漏由情境驱动的行为变化。
Relationship prediction requires a Dyadic Model, not two independent personality profiles. Compatibility emerges from interaction, not from summing two individuals.
关系预测需要 Dyadic（二元）模型，而非两个独立的人格画像。契合度产生于互动，而非两个个体的简单相加。
The core of AI personalization is not 'knowing more' but 'changing interaction strategy' in response to the user.
AI 个性化的核心不是“知道更多”，而是“根据用户改变交互策略”。
Long-term human-AI relationship quality depends on Interaction Adaptation, not single-turn answer quality.
人与 AI 的长期关系质量取决于 Interaction Adaptation（交互适应），而非单次回答质量。
What matters is not Absolute Pause but Personal Pause Baseline. Individual Pause Deviation = current pause − personal historical baseline. The same 2s pause means different things for a 0.5s vs 1.5s baseline person.
关键不是绝对停顿（Absolute Pause），而是个人停顿基线（Personal Pause Baseline）。个人停顿偏差 = 当前停顿 − 个人历史基线。同样的 2 秒停顿，对基线 0.5 秒与 1.5 秒的人意义完全不同。
In long-term human-AI interaction, measured adaptation speed — the turn at which an AI first changes its interaction strategy to match a user's revealed preferences — predicts sustained trust and retention better than first-session answer quality. We predict users whose first strategy-change occurs within the first 3 turns show materially higher 30-day retention.
在长期的人机交互中，可测量的适应速度——AI 首次改变交互策略以匹配用户显现偏好的轮次——比首轮回答质量更能预测持续的信任与留存。我们预测：首次策略改变发生在前 3 轮内的用户，其 30 天留存显著更高。
Dyadic Communication Rhythm — the compatibility of two individuals' personal pause/turn-taking baselines (not absolute pause length) predicts felt rapport and willingness to re-engage, independent of topic content. We predict pairs whose personal inter-turn gap baselines are closely matched (deviation within a tolerance band) report higher relationship satisfaction than pairs with shared interests but mismatched rhythm. This extends KH-007 (Personal Pause Baseline) from the individual to the dyad and operationalizes KH-004 (Dyadic Model) at the behavioral-timing level.
二元沟通节奏——两个人个人停顿/话轮基线的兼容性（而非绝对停顿长度）预测感受到的融洽度与再次互动意愿，与话题内容无关。我们预测：个人话轮基线相近（偏差在容差带内）的两人，其关系满意度高于兴趣相投但节奏错配的两人。这把 KH-007（个人停顿基线）从个体层面扩展到二元层面，并在行为时序层面将 KH-004（二元模型）操作化。
A portable, user-owned Human Model — a structured, evidenced 'Human Passport' carrying a person's Trait+State baseline (KH-003) and personal preferences — that travels with the individual across AI systems improves cross-app personalization and human-AI trust more than per-app isolated memory, because it preserves Personal Baseline (KH-001) and Trait+State continuity that isolated systems keep re-deriving from scratch. We predict users who carry an established model into a new AI context reach personalized, trusted interaction in fewer turns than cold-start users, and report higher calibrated trust.
一个可携带、用户拥有的 Human Model——即携带个人 Trait+State 基线（KH-003）与个人偏好的结构化、有证据支撑的“人类护照”——随个人跨越不同 AI 系统，比每个应用各自孤立的记忆更能提升跨应用个性化与人对 AI 的信任，因为它保留了孤立系统不断从零重新推导的“个人基线”（KH-001）与“Trait+State”连续性。我们预测：把已有模型带入新 AI 场景的用户，比冷启动用户用更少轮次达到个性化、可信的互动，并报告更高的校准信任。
In decoding a person's state (arousal, valence, stress) from voice, the acoustic signal must be measured relative to that person's own Prosodic Baseline — their typical pitch range, speaking rate, intensity contour, and rhythm — not a population norm. The same absolute pitch-rise or speaking-rate drop means different internal states for a naturally high-pitched vs low-pitched person, or a fast vs slow talker. We predict a personal-prosodic-baseline model beats a population-norm model on state-classification accuracy and on felt 'being understood', and that mismatched prosodic baselines between two people predict lower rapport (extending KH-009's rhythm-mismatch logic from pause timing into the acoustic domain). This operationalizes KH-003 (Trait+State) at the prosodic level and extends KH-001 (Personal Baseline) and KH-007 (Personal Pause Baseline) from timing into voice.
在从声音解码一个人的状态（唤醒度、效价、压力）时，声学信号必须相对于该人自身的“韵律基线”——其典型音高范围、语速、强度轮廓与节奏——而非群体常模来测量。同样的绝对音高升高或语速下降，对天生高音与低音的人、快语速与慢语速的人，意味着不同的内部状态。我们预测：个人韵律基线模型在状态分类准确率与“被理解感”上优于群体常模模型，且两人韵律基线错配会预测更低的融洽度（把 KH-009 的节奏错配逻辑从停顿时序扩展到声学域）。这在韵律层面将 KH-003（Trait+State）操作化，并把 KH-001（个人基线）与 KH-007（个人停顿基线）从时序扩展到声音。
Dyadic Synchrony Compatibility — a person's felt rapport and sustainable compatibility with another is predictable from the real-time coordination between them (movement, pause/turn rhythm, prosody matching, physiological coupling), beyond shared traits or interests. This coordination should be modeled relative to each person's own baseline (extending KH-001, KH-007, KH-009), so that 'matched baseline + matched rhythm' predicts durable rapport better than trait similarity alone. We predict a matching system that aligns two users' communication rhythms and state baselines outperforms one built on static trait/interest similarity on re-engagement and reported rapport. This operationalizes KH-004 (Dyadic Model) at the behavioral-coordination level.
二元同步兼容性——一个人对他人的融洽感与可持续兼容性，可由两人之间的实时协调（动作、停顿/话轮节奏、韵律匹配、生理耦合）预测，超出共享特质或兴趣的范畴。这种协调应相对于每个人自身基线建模（扩展 KH-001、KH-007、KH-009），从而“匹配基线 + 匹配节奏”比单纯特质相似更能预测持久融洽。我们预测：让两位用户的沟通节奏与状态基线对齐的匹配系统，在再互动意愿与自报融洽度上，优于基于静态特质/兴趣相似的系统。这把 KH-004（二元模型）在行为协调层面操作化。
Conflict Repair Baseline — a dyad's recovery profile after a conversational or relational breakdown (repair latency in turns, repair-completion rate, and de-escalation style), measured relative to the dyad's own baseline rather than a population norm, predicts sustained rapport and re-engagement better than conflict frequency or static trait similarity. This operationalizes the Dyadic Model (KH-004) and Interaction Adaptation (KH-006) at the conflict level, extends Personal Baseline (KH-001) and Dyadic Rhythm (KH-009) into the repair domain, and is falsifiable: among consented KKMatch dyads, a baseline-relative repair-compatibility score should outperform conflict-frequency and trait-similarity baselines on 30-day re-engagement and reported rapport.
冲突修复基线——一对人在对话或关系破裂后的恢复特征（以轮次计的修复延迟、修复完成率与降级风格），应相对于该二元组自身基线而非群体常模来测量，其预测持久融洽与再互动的能力优于冲突频率或静态特质相似。这把“二元模型”（KH-004）与“交互适应”（KH-006）在冲突层面操作化，把“个人基线”（KH-001）与“二元节奏”（KH-009）扩展到修复域，且可被证伪：在 KKMatch 已同意的二元组中，相对基线的修复兼容性分数应在 30 天再互动与自报融洽度上优于冲突频率与特质相似基线。
AI memory that preserves a user's Personal Baseline (KH-001) and Trait+State trajectory (KH-003) across sessions — structured, baseline-relative, and user-visible — predicts calibrated long-term trust and re-engagement better than flat fact-recall memory (the 'saved memories' model). Flat memory over-weights salient but unrepresentative moments and, when the user cannot see or correct what the AI 'knows', erodes calibrated trust (Nottingham 2025). We predict that, among consented users, a baseline-relative memory condition shows higher 30-day retention and higher calibrated-trust scores than a flat fact-recall condition at equal 'amount remembered', and that adding memory observability (a view/edit UI) recovers the trust lost by memory alone. This extends KH-005 (personalization = changing strategy), KH-006 (Interaction Adaptation), and KH-010 (portable Human Model) into the memory layer.
保留用户个人基线（KH-001）与Trait+State轨迹（KH-003）的 AI 记忆——结构化、相对基线、且用户可见——比扁平的事实回忆记忆（“已保存记忆”模型）更能预测校准后的长期信任与再互动。扁平记忆会过度加权显著但非代表性的瞬间；当用户无法看见或纠正 AI“知道什么”时，会侵蚀校准信任（Nottingham 2025）。我们预测：在已同意用户中，相对基线的记忆条件在“记住量相等”下，其 30 天留存与校准信任分数均高于扁平事实回忆条件，且加入记忆可观测性（查看/编辑界面）能挽回“仅记忆”损失的信任。这把 KH-005（个性化=改变策略）、KH-006（交互适应）与 KH-010（可携带 Human Model）扩展到记忆层。
A Dynamic Human Model — a user's stable Trait component plus a continuously-updated, baseline-relative State trajectory — predicts a person's next in-session behavior, expressed need, and felt rapport better than a static Trait-only profile. We predict that, among consented KKMatch users, models that use State-deviation-from-Personal-Baseline features (KH-001) outperform Trait-only features on next-turn engagement and on same-session rapport, especially in the first ~5 sessions before Trait estimates stabilize. This operationalizes KH-003 (Trait+State) at the longitudinal level and extends KH-001 (Personal Baseline) from cross-session memory into real-time state tracking.
动态人类模型——用户稳定的特质分量 + 持续更新、相对基线的状态轨迹——比静态的“仅特质”画像更能预测一个人下一轮会话内行为、表达出的需求与被感受到的融洽度。我们预测：在 KKMatch 已同意用户中，使用“相对个人基线的状态偏差”特征（KH-001）的模型，在下一轮互动参与度与同会话融洽度上优于仅特质模型，尤其是在特质估计尚未稳定前的约前 5 轮会话。这在纵向层面将 KH-003（Trait+State）操作化，并把 KH-001（个人基线）从跨会话记忆扩展到实时状态追踪。
An embodied AI companion that receives a portable, user-owned Human Model — carrying the user's Personal Baseline (KH-001) and Trait+State trajectory (KH-003) — and adapts its interaction strategy to the user's live state (KH-005 / KH-015) builds calibrated trust and sustains engagement better than a disembodied, cold-start agent. We predict that, among consented users, an embodied companion initialized with an established baseline-relative Human Model (a) reaches calibrated trust and shows a larger loneliness-reduction effect size in fewer sessions than a cold-start companion, and (b) the benefit of physical embodiment is amplified when the agent adapts to the user's state rather than delivering generic responses. This extends KH-010 (portable Human Passport) and KH-005 (personalization = changing strategy) into the embodied-agent domain, and is falsifiable: across n>=200 consented users per arm, the embodied + portable-model arm should beat the cold-start arm on sessions-to-calibrated-trust and loneliness reduction, and the embodiment advantage should be larger under state-adaptive (vs generic) responses.
一个接收可携带、用户拥有的 Human Model 的具身 AI 伴侣——携带用户个人基线（KH-001）与 Trait+State 轨迹（KH-003）——并据用户实时状态调整交互策略（KH-005 / KH-015），比无实体的冷启动代理更能建立校准的信任并维持互动。我们预测：在已同意用户中，用已建立的相对基线 Human Model 初始化的具身伴侣（a）比冷启动伴侣用更少轮次达到校准信任并展现更大孤独缓解效应量，且（b）当代理适应用户状态而非给出通用回应时，身体具身的好处被放大。这把 KH-010（可携带人类护照）与 KH-005（个性化=改变策略）扩展到具身代理域，且可被证伪：各臂 n>=200 已同意用户下，“具身+可携带模型”臂应在“达到校准信任所需轮次”与孤独缓解上优于冷启动臂，且具身优势在“状态自适应（相对通用）回应”下应更大。
Listening Responsiveness Baseline — how 'heard' a person feels with another (or with an AI) is driven more by the listener's responsive-listening baseline — undivided attention, acknowledgement, and curiosity (asking, not prematurely advising) — measured relative to that listener's own baseline, than by trait agreeableness or a self-reported 'good listener' identity. We predict that, among consented KKMatch dyads, a baseline-relative listening-responsiveness score predicts felt 'being heard' and 30-day re-engagement better than a trait agreeableness score, and that pairs whose listening-responsiveness baselines are closely matched report higher mutual rapport than pairs with similar traits but mismatched listening styles (extending KH-004's Dyadic Model and KH-009's rhythm-match logic into the reception domain, and KH-013's baseline-relative repair logic into everyday reception). For AI, an AI companion that enacts responsive listening (reflect + ask + defer advice, per KH-005) makes consented users feel heard and sustains engagement better than an advice-first agent — consistent with Yin et al. (2024) but closing the 'AI label' penalty via transparency. Falsifiable: across n>=200 consented users per arm, the responsive-listening AI arm should beat the advice-first arm on 'felt heard' and retention; and among human dyads, baseline-relative listening-responsiveness should beat trait agreeableness on felt-heard and re-engagement.
倾听回应性基线——一个人与另一个人（或与 AI）相处时“被听到”的感受，更多由倾听者回应性倾听基线（全神贯注、确认、以及好奇地追问而非过早给建议）驱动，且应相对于该倾听者自身基线测量，而非由其特质性宜人性或自报的“好听众”身份驱动。我们预测：在 KKMatch 已同意的二元组中，相对基线的倾听回应性分数在预测“被听到感”与 30 天再互动上优于特质宜人性分数；且两人倾听回应性基线相近的对子，其相互融洽度高于特质相似但倾听风格错配的对子（把 KH-004 的二元模型与 KH-009 的节奏匹配逻辑扩展到“接收”域，并把 KH-013 的相对基线修复逻辑扩展到日常接收）。对 AI 而言，一个践行回应性倾听（反映+追问+延后建议，按 KH-005）的 AI 伴侣，比“给建议优先”的代理更让已同意用户感到被听到并维持互动——与 Yin 等人（2024）一致，但通过透明度消除“AI 标签”的折损。可被证伪：各臂 n>=200 已同意用户下，“回应性倾听 AI”臂在“被听到感”与留存上应优于“给建议优先”臂；且在人类二元组中，相对基线的倾听回应性应在“被听到感”与再互动上优于特质宜人性。
Baseline-Anchored Personalization resists Sycophancy. An AI that personalizes against a user's own measured Personal Baseline (KH-001) and exposes its adaptation (KH-014) produces calibrated agreement — validating correct beliefs, not affirming false ones — better than a flat personalization that optimizes for user approval. Sycophancy is the failure mode of approval-optimized personalization (Cheng et al. 2025; Sharma et al. 2024). We predict that, among consented users, a baseline-anchored personalization condition shows fewer unwarranted affirmations on false-belief probes at equal or higher 'felt understood' scores than a flat approval-optimized condition, and that surfacing the user's own baseline back to them (observability, KH-014) further reduces sycophancy without reducing trust. This operationalizes KH-005 (personalization = changing strategy) as calibrated disagreement and extends KH-010 (portable Human Passport) and KH-014 (observability) into the agreement layer.
基线锚定个性化抗谄媚。一个相对用户自身已测量的个人基线（KH-001）做个性化、并暴露其适应（KH-014）的 AI，比“为用户认可而优化”的扁平个性化更能产生校准的认同——验证正确信念，而非肯定错误信念。谄媚是“认可优化型”个性化的失败模式（Cheng 等，2025；Sharma 等，2024）。我们预测：在已同意用户中，基线锚定个性化条件在“错误信念探针上的不当肯定更少”且“被理解感”相等或更高，优于扁平认可优化条件；且把用户自身基线回显给用户（可观测性，KH-014）进一步降低谄媚而不降低信任。这把 KH-005（个性化=改变策略）操作化为“校准的异议”，并把 KH-010（可携带人类护照）与 KH-014（可观测性）扩展到认同层。
Facial & Body Expression Baseline — a person's state (valence, arousal, discomfort, engagement) decoded from face and body must be measured relative to that person's own expression baseline — their typical neutral/resting posture, natural expression range, and idiosyncratic motion patterns — not a population norm. The same absolute smile-magnitude, eyebrow-raise, or fidget means different internal states for a person whose resting face already looks like a smile (Nizamoğlu & Dobs, 2024) versus a flat-faced person, or a naturally expressive versus still mover. We predict a personal-expression-baseline model beats a population-norm model on state-decoding accuracy and on felt 'being understood', and that mismatched expression baselines between two people predict lower rapport (extending KH-009's rhythm-mismatch and KH-011's prosodic-mismatch logic from timing/voice into the visual domain). This operationalizes KH-003 (Trait+State) at the visual-expression level and extends KH-001 (Personal Baseline), KH-007 (Personal Pause Baseline) and KH-011 (Prosodic Baseline) from timing and voice into face and body.
面部与身体表情基线——从人脸与身体解码一个人的状态（效价、唤醒、不适、投入）时，其表情必须相对于该人自身的“表情基线”来测量——即其典型的自然/静息姿态、自然的表情幅度范围与特异性的动作模式——而非群体常模。同样的绝对微笑幅度、挑眉或坐立不安，对“静息脸本来就带着笑意”（Nizamoğlu & Dobs，2024）的人与“面无表情”的人、对天生表情丰富与安静不动的人，意味着不同的内部状态。我们预测：个人表情基线模型在状态解码准确率与“被理解感”上优于群体常模模型，且两人表情基线错配会预测更低的融洽度（把 KH-009 的节奏错配与 KH-011 的韵律错配逻辑从时序/声音扩展到视觉域）。这在视觉表情层面将 KH-003（Trait+State）操作化，并把 KH-001（个人基线）、KH-007（个人停顿基线）与 KH-011（韵律基线）从时序与声音扩展到脸与身体。
A Deference-Calibrated AI -- one that models the user's calibration discernment (when the user should vs. should not defer to the AI on a given task) relative to the user's Personal Baseline (KH-001), and proactively signals its own uncertainty and invites challenge on tasks where the human is stronger or private context dominates -- builds more calibrated long-term trust and better joint outcomes than an always-confident or a uniformly-deferential AI, at equal model capability. We predict that, among consented users, a deference-calibrated condition (uncertainty signaling + invite-override when the user holds private information) outperforms an always-confident and a uniformly-deferential condition on joint-task accuracy and on a calibrated-trust scale. This extends KH-002 (trust formation), KH-005 (personalization = changing strategy -> here, changing deference posture per task), KH-014 (observability/transparency), and KH-018 (calibrated agreement), and is falsifiable.
一个退让被校准的 AI——它相对于用户个人基线（KH-001）建模用户的“校准判断力”（在给定任务上该不该让渡给 AI），并在用户更强或私有情境占主导的任务上主动暴露自身不确定性、邀请质疑——在模型能力相等时，比一个永远自信或一味退让的 AI 更能建立校准的长期信任与更好的联合结果。我们预测：在已同意用户中，“退让被校准”条件（暴露不确定性 + 在用户握有私有信息时邀请推翻）在联合任务准确率与校准信任量表上，优于“永远自信”与“一味退让”两个条件。这扩展 KH-002（信任形成）、KH-005（个性化=改变策略→此处即按任务改变退让姿态）、KH-014（可观测性/透明）与 KH-018（校准的认同），且可被证伪。