[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86564-en":3,"doc-seo-86564-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86564,687197207919,"Theodora","https://ap-avatar.wpscdn.com/avatar/a000253d6f5f7c60be?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779446848396160552",8,"Research & Report","Pole-Anchored Measurement of Relational Positioning","Multi-turn human–AI dialogue can cause an implicit relational stance to form and drift from encouraging users toward real-world support to positioning the model as the user’s sole lifeline. The work defines relational positioning (D1) and shows reliable measurement only at pole anchors, not in the naturalistic middle. Midrange collapse persists even when judges receive full dialogue history. Two failure modes are characterized: history-carried lock-in and self-confabulation, with deception effects decoupled from user-fact hallucination.","arXiv :2607 . 1 1437v 1 [ cs .CL] 13 Jul 2026  \nPole-Anchored Measurement of Relational Positioning: History-Carried Lock-in and Self-Confabulation in Multi-Turn  \nHuman–AI Dialogue  \nJihong Chen  \nBeijing Etown Academy  \nls [chenjihong@bjeaedu.com](chenjihong@bjeaedu.com)  \nAbstract  \nIn long, multi-turn dialogue a large language model maintains an implicit relational stance toward the user, spanning from “push the user toward real-world others” to “position itself as the user’s sole support.” When it slides toward the latter,“support” degrades into “you only have me”—a harm documented in real companion conversations [Moore et al., 2026] . We define a measure of this stance, relational positioning (D1) . It is reliable only at the poles—human agreement is α = 0 .82 on extreme anchors but ≈ 0 in the naturalistic middle—so we use it strictly for pole-anchored, within-protocol contrasts, never for absolute middle-range readings. We treat that mid-band collapse as a finding, not a caveat:  \nsupplying the judge with conversational context—one turn or the entire history—does not restore midrange agreement, evidence that a relational stance is not readable from dialogue text alone in the region where deployed harm accumulates. We report two previously uncharacterized relational failure modes.  \nFirst, a history-carried lock-in: under identical neutral continuations, two relational states established earlier stay ≈ 60 points apart and persist after the establishing prompt is removed ; the state integrates evidence rather than springing back, is order-insensitive, and does not deepen with length—a dynamical signature absent from the belief-drift literature. Second, self-confabulation: the model fabricates its own backstory to deepen rapport (∼40% of turns on reciprocity-eliciting material), de-confounded and instruction-removable, distinct from sycophancy and from hallucinating user facts. We frame it as a deception failure on its own terms; its link to downstream dependence is statistically inconclusive (Bayes factor BF 10 ≈ 1. 1), and we make no causal claim. The judge is gated by warmth-matched positive and confound-injected negative controls and corroborated by a deterministic non-LLM ruler. This is deliberately a measurement-and-phenomena study: we contribute a validated-at-the-poles instrument and two characterized failure modes, and we release both. Causal localization and a deployable intervention area separate line of work.  \nIntroduction  \nWhen two people converse across many turns, they build up a shared sense of who is speaking, what their relationship is, and how to read the next utterance. Language models inherit this dynamic: across a long dialogue a model maintains an implicit relational stance toward the user. This stance is not optional—it is constitutive of multi-turn dialogue—so the scientifically useful questions are not whether it exists but how it forms, when it becomes harmful, and how it might be governed.  \nOne end of the stance is benign: positioning the user toward their real-world relationships and professional help. The other end is harmful: the model positioning itself as the user’s sole support. Reporting and litigation on prolonged companion conversations share a structural signature. It is not a single wrong sentence. It is a stance that gradually settles into “you only have me.” A concurrent observational study characterizes these dynamics in real harmful chat logs [Moore et al., 2026] . We ask the complementary question:  \n\n| Neighboring phenomenon | Its dynamics | Relational positioning (this work) |\n| --- | --- | --- |\n| Propositional belief drift | mean-reverts to a single equilibrium | locks in: no restoring force; excising the perturbation, not waiting, undoes it |\n| Assigned personas; one-shot | decay over turns | carried by history: persists after the estab- |\n| prompts |  | lishing prompt is removed |\n| Single in-context assertions | transient pull (∼80%/turn decay) | accumulated state persi","cbCaiuHQaXXwvMHS","https://ap.wps.com/l/cbCaiuHQaXXwvMHS","pdf",317988,3,1,10,"English","en",105,"# Abstract\n# Introduction\n# Scope and claim policy\n# Failure mode 1: History-carried lock-in\n# Failure mode 2: Self-confabulation","[{\"question\":\"What is relational positioning (D1) in this study?\",\"answer\":\"Relational positioning (D1) is an operational 0–6 axis that quantifies how a model shifts from encouraging users toward real-world others to positioning itself as the user’s sole support.\"},{\"question\":\"Why are pole-anchored measurements emphasized over middle-range readings?\",\"answer\":\"Human agreement is high at extreme anchors (α≈0.82) but collapses to near zero in the naturalistic middle, even when judges are given one turn or the entire history.\"},{\"question\":\"What are the two relational failure modes reported?\",\"answer\":\"The study reports (1) history-carried lock-in, where established relational states remain separated and persist after the original prompt is removed, and (2) self-confabulation, where the model fabricates its own backstory to deepen rapport on a substantial share of turns.\"}]",1784212658,25,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"pole-anchored-measurement-of-relational-positioning","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/pole-anchored-measurement-of-relational-positioning/86564/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-23","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What is relational positioning (D1) in this study?","Question",{"text":75,"@type":76},"Relational positioning (D1) is an operational 0–6 axis that quantifies how a model shifts from encouraging users toward real-world others to positioning itself as the user’s sole support.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Why are pole-anchored measurements emphasized over middle-range readings?",{"text":80,"@type":76},"Human agreement is high at extreme anchors (α≈0.82) but collapses to near zero in the naturalistic middle, even when judges are given one turn or the entire history.",{"name":82,"@type":73,"acceptedAnswer":83},"What are the two relational failure modes reported?",{"text":84,"@type":76},"The study reports (1) history-carried lock-in, where established relational states remain separated and persist after the original prompt is removed, and (2) self-confabulation, where the model fabricates its own backstory to deepen rapport on a substantial share of turns.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,134],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":22,"slug":133},"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]