{"id":"46c376b0-21e6-4fbb-b91e-dcb8fbe4c1f2","arxiv_id":"2507.11479","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"PAiR couples PAi's Chronicle identity models with an XR scene generator in a closed loop, illustrated by two proof-of-concept scenarios in the OpenDome engine.","lead":"This paper introduces PAiR, a framework that combines the authors' Perspective-Aware AI with extended reality to personalize immersive environments from user-specific identity graphs called Chronicles. It describes the architecture and two illustrative OpenDome scenarios, a financial helper and a desk environment, but provides no user study or quantitative evaluation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Closed-loop adaptation is asserted, not demonstrated: the paper never specifies or evaluates how user feedback updates the Chronicle, so the central claim lacks support.","rationale":"The reader correctly identified the Chronicle-quality assumption as load-bearing and the absence of evaluation as a red flag. My pass reaches a slightly different focal point: the paper's own text describes the feedback update as optional and provides no mechanism or threshold for applying it, so the closed-loop component of the central claim is not merely unvalidated, it is unspecified. Even if the Chronicle construction from reference [1] were fully trusted, the claimed adaptive loop would still need a concrete update pathway and evidence that it changes behavior. This reinforces the CONDITIONAL verdict rather than moving it: the framework is a plausible architecture, but the core asserted capability is not demonstrated, and the recommended path is to require an end-to-end test of the update loop before accepting claims of adaptive personalization.","tokens_in":148,"tokens_out":2122,"duration_ms":35960,"concrete_test":"Instrument the OpenDome PAiR integration to log the full Chronicle graph before and after a scripted feedback event (for example, the user gazes at the pie chart and detectors emit <user, has_emotion, curious>). Execute the Reasoner's update path, then compute a graph diff listing nodes and edges added, modified, or deleted. Repeat over at least 20 scripted events with varying detector outputs. If the Chronicle state is unchanged or only trivially appended in all cases, the closed-loop claim is unsupported; if changes do occur, verify that they follow from the feedback and that a subsequent scene-generation query returns different results because of them.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is a closed loop where immersive experiences are driven by Chronicles that adapt to evolving user behavior. The only mechanism described is one sentence in Section 3.3: monitoring data 'can also serve as an additional source of information to update the Chronicle ... when appropriate thresholds and validated theories are met.' No thresholds, validated theories, update queries, or implementation details are supplied. In the two proof-of-concept scenarios, feedback is explicitly optional: Section 4.2 Step 6 says captured cues 'can optionally be used to update their Chronicle,' and Section 4.3 repeats the same wording. The scenarios are hand-illustrated traces rather than measured runs, with no logged state before and after feedback. If no actual Chronicle mutation is implemented, PAiR reduces to a one-shot query-and-render pipeline, and the adaptive personalization promised in the abstract is not realized. This is more load-bearing than the Chronicle-accuracy concern: even a perfectly accurate Chronicle would not yield the claimed closed loop without a functioning, evaluated update path. The paper therefore overclaims its demonstrated utility, and the conditional verdict should hinge on evidence of this loop, not just on the quality of the underlying user model.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes PAiR (Perspective-Aware AI in Extended Reality), a framework that integrates the authors' earlier Perspective-Aware AI (PAi) concept, based on Chronicles (dynamic knowledge graphs representing user identity), with XR environments. The manuscript describes two components (XR and PAi) with their modules (Spatial Monitor, User State Monitor, LLM Scribe, Reasoner, Object Synthesizer, Chronicle Pool), a system flow, and two scenario traces implemented in the Unity-based OpenDome engine: a financial visualization scenario and an emotion-driven desk scenario. The central claim is that PAiR creates a closed loop in which immersive experiences are driven by user-specific Chronicles that adapt based on evolving user behavior.","tokens_in":8219,"tokens_out":7487,"duration_ms":78581,"significance":"The paper addresses a genuine gap: current XR personalization is reactive and short-term, while identity-level, interpretable user models are rare. The architectural decomposition is clear and the scenario traces are useful as illustrations of a possible design. However, the paper provides no user data, no quantitative metrics, no comparisons, and no runnable code; the two 'proof-of-concept' scenarios are scripted examples rather than measured demonstrations. The value of the paper is therefore as a conceptual blueprint, not as an empirical validation of adaptive personalization.","major_comments":[{"comment":"The closed-loop claim in the abstract and Section 1 is not supported by the described system. Section 3.3 states only that monitoring data 'can also serve as an additional source of information to update the Chronicle ... when appropriate thresholds and validated theories are met,' but no thresholds, update operations, or validation criteria are specified. In the two scenarios (Section 4.2, Step 6; Section 4.3), Chronicle updates are described as optional ('can optionally be used to update their Chronicle') and no before/after Chronicle content is shown. As written, the system is a one-shot query-and-render pipeline, and the adaptive closed-loop behavior that distinguishes PAiR from reactive personalization is not demonstrated. Please either implement and evaluate a concrete update path or explicitly re-scope the contribution to an architecture with a planned feedback mechanism.","section":"Section 3.3"},{"comment":"The paper claims to 'demonstrate its utility' through two proof-of-concept scenarios, but the demonstration consists of hand-authored step-by-step traces. There is no evidence that the Unity/OpenDome integration was executed, that the Reasoner LLM and situation detectors produced the shown outputs, or that the rendered scenes were presented to users. No logs, screenshots from a run, timing data, or comparisons against a non-personalized baseline are provided. If the scenarios are illustrations rather than executed implementations, the utility claim should be revised accordingly; if they were executed, please provide runnable artifacts or at least detailed logged traces.","section":"Section 4"},{"comment":"The utility of PAiR depends on the correctness of several components that are not validated in this manuscript: the off-the-shelf situation detectors mapping facial cues to semantic triples, the fine-tuned Reasoner LLM converting prompts and cues into schema-conformant queries, and the Chronicle content built by the pipeline referenced in [1]. Section 4.3 assumes, for example, that facial cues reliably produce <user, has_emotion, sad> and <situation_2, has_possible_cause, missing_someone>; a misclassification at any of these steps would yield a semantically wrong scene. Please add component-level validation or sensitivity analysis, or clearly state that these are assumed capabilities rather than demonstrated ones.","section":"Section 4.3"}],"minor_comments":[{"comment":"The text refers to 'Figure 3.1' but the figure is numbered Fig. 1 in the manuscript; please correct the cross-reference.","section":"Section 3.3"},{"comment":"There are numerous formatting and typographical issues, such as 'experiences—yetcurrentsystems' missing a space and missing spaces after colons in module lists; a careful copyedit is needed.","section":"Throughout"},{"comment":"The semantic similarity threshold θ is introduced as a free parameter but no value or selection procedure is given; please state whether it is fixed in the implementation or how it is tuned.","section":"Section 4.2, Step 3"},{"comment":"The 'off-the-shelf models' used for Situation & Context Detectors are not cited; adding concrete references would make the framework more reproducible.","section":"Section 3.2"},{"comment":"The Chronicle construction pipeline is summarized only by reference to [1]; a slightly more self-contained description of how Chronicles are learned and updated would help readers not familiar with the prior PAi work.","section":"Section 2.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a conceptual framework contribution rather than an empirical one; the gap between the claimed closed-loop adaptive system and the scripted scenario traces is the main obstacle. I encourage the editor to require that the revised version either provide a concrete, implemented Chronicle-update mechanism with evidence, or clearly qualify the claims as an unimplemented blueprint."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Enjoyable, clearly written framework paper—but keep expectations in check. PAiR is a plausible architectural sketch for giving XR systems an identity-level context layer via PAi's Chronicles, and the anchor-point reasoning flow in Section 4.2 is the most concrete thing here. The integration of PAi with XR is genuinely new relative to the cited literature, and the OpenDome instantiation gives a concrete sense of how the pieces fit. The paper earns credit for being explicit that the scenarios are proof-of-concept traces, not user studies.\n\nThe soft spot is structural, not cosmetic. The central selling point is a closed loop: Chronicles adapt based on evolving user behavior. That adaptation is never actually specified or evaluated. Section 3.3 mentions updating the Chronicle \"when appropriate thresholds and validated theories are met\" with no details, and in both scenarios the update is explicitly optional (\"can optionally be used\"). The stress-test note has this right: even a perfect Chronicle would not yield the claimed closed loop without a working update path. The lack of user data, metrics, and baselines is consistent with a design proposal, but then the abstract and conclusion should not say the framework \"demonstrates utility.\" The other heavy assumption is that Chronicles—borrowed wholesale from the authors' prior PAi papers—are accurate, reasoning-ready identity models. That is plausible but unvalidated here, and self-citation alone doesn't make it so.\n\nNone of this kills the paper as a blueprint. The architecture is internally coherent, the two scenarios are illustrative rather than deceptive, and the writing is honest about the absence of empirical validation in future work. Read it as a position paper for a research agenda, not as evidence that the system works.\n\nI'd send this to a venue that welcomes system/design proposals—a workshop, CHI late-breaking work, or an HCI venue with short-paper format—with a request to soften the utility claim and add an evaluation roadmap. It deserves serious referee engagement, so not a desk reject. I would not cite it as a validated system, but I might cite it as a representative framework for perspective-aware XR if writing a related-work section.","headline":"PAiR is a clear architectural blueprint for perspective-aware XR, but the closed-loop adaptation claim is asserted, not demonstrated, so the paper should be read as a design proposal rather than a validated system.","tokens_in":8651,"tokens_out":2548,"would_cite":false,"duration_ms":27558,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"PAiR creates a closed loop in which extended-reality scenes are driven by user-specific Chronicles, identity models learned from multimodal digital footprints.","keywords":["Perspective-Aware AI","Extended Reality","Chronicles","user identity modeling","knowledge graphs","human-AI interaction","interpretable personalization","closed-loop adaptive systems"],"falsifier":"Run two users through the same XR scene with identical prompts but deliberately different Chronicles, one with the pivotal memory removed; if the rendered scene is identical, the loop is not actually driven by the Chronicle. Separately, feed a recorded set of facial-expression frames through the Situation and Context Detectors and count how often the produced semantic triples match a human labeler's judgment; if accuracy is near chance, the emotional-cue branch of the loop has no reliable input.","tokens_in":7685,"feed_emoji":"🥽","tokens_out":5797,"duration_ms":63880,"temperature":0.7,"pith_summary":"PAiR is a proposed framework for making extended-reality (XR) experiences respond to a user's identity, not just their immediate behavior. The paper's central claim is that a closed-loop connection between Perspective-Aware AI and XR lets immersive scenes be driven by a Chronicle—a dynamic knowledge graph built from a person's multimodal digital footprint—so the experience reflects the user's cognitive and experiential history. The authors argue this goes beyond reactive personalization, which optimizes short-term engagement, by giving the system a reasoning-ready model of who the user is. Two proof-of-concept scenarios in an XR engine show the loop in action: a financial helper that places a personalized spending chart on a specified table, and a desk environment that, sensing sadness, retrieves a happy memory and renders it as a photo frame. The value, if the framework works, is that XR can become interpretable and adaptive in ways tied to user identity rather than surface cues.","feed_headline":"A closed loop ties XR scenes to your evolving identity","feed_subtitle":"Instead of reacting to gaze and clicks, XR scenes reason over a living model of who you are and what you remember.","key_machinery":"The central object is the Chronicle, a dynamic knowledge graph that encodes a user's cognitive, behavioral, and experiential evolution from multimodal digital footprints. The argument's mechanism is a closed loop: a fine-tuned Reasoner LLM turns prompts and sensed states into semantic triples, queries the Chronicle, the Object Synthesizer retrieves or generates matching media, an XR Scene Generator renders it, and the User State Monitor feeds reactions back to update the Chronicle. The Situation Graph ontology supplies the semantic vocabulary that keeps the loop interpretable, while the LLM Scribe translates between human language, symbolic reasoning, and executable XR scene configurations.","core_discovery":"The paper's discovery is that the Chronicle, originally designed for decision support, can serve as the reasoning substrate for immersive environments. PAiR routes both explicit user requests and passively sensed emotional cues through the same pipeline: a fine-tuned Reasoner LLM converts input into semantic triples, queries the user's Chronicle, applies spatial, temporal, ontological, and thematic reasoning to anchor the result in the 3D scene, and hands the output to a synthesizer that retrieves or generates media. The rendered scene is then displayed by an XR scene generator, while the user's reactions are fed back into the Chronicle, closing the loop. The paper demonstrates this with two implemented scenarios and claims this is a foundational blueprint for perspective-aware XR rather than a finished product.","pith_inferences":["The paper does not run this test, but if the closed-loop claim holds, an obvious evaluation is to compare scenes produced for the same prompt with a user's real Chronicle versus a shuffled or empty Chronicle; if the output does not change, the Chronicle is not doing the work.","Because the paper adds no measurement of Chronicle accuracy, a concrete next step is to benchmark whether the Reasoner's semantic triples and object retrievals match ground-truth user preferences; until then, the closed loop's benefit is an architectural promise rather than a measured effect.","The sharing of Chronicles suggests a new class of XR social applications, such as perspective-taking training, where the same scene is rendered through two different identity models and the difference between those renderings becomes the content.","A testable extension is to make the anchor-selection threshold and the emotion-to-situation inference measurable, so failures in scene grounding can be traced to either the semantic-similarity step or the detector step."],"forward_implications":["Immersive systems can move from reactive optimization of gaze, clicks, and heart rate to identity-level adaptation: the same prompt can produce different scenes for different users based on different Chronicles.","Personalization compounds over time, because user reactions are written back into the Chronicle, so each session should make the next one more aligned with the user's evolving state.","Explicit and implicit input become interchangeable: a typed request and a detected facial expression both enter the same semantic-reasoning path, so passive cues can trigger proactive experiences.","With consent, Chronicles can be shared, so XR can synthesize experiences from another person's perspective rather than only from one's own history.","The architecture is an engine-agnostic blueprint: only the XR Scene Generator needs to be adapted when moving from one XR engine to another."],"supporting_citations":[{"why":"Supplies the Chronicle construction pipeline—multimodal encoding, temporal segmentation, and knowledge-graph learning—that PAiR assumes as its identity model.","marker":"[1]"},{"why":"Defines the Situation Graph ontology and the embedding and query mechanics that the Reasoner uses to ground user states semantically.","marker":"[6]"},{"why":"Argues that Chronicles can be shared to synthesize experiences from others' perspectives, the capability PAiR extends into XR.","marker":"[3]"},{"why":"Provides the neural-symbolic translation approach behind the LLM Scribe's conversion of language and sensed cues into symbolic scripts.","marker":"[2]"},{"why":"Motivates using fine-tuned LLMs as symbolic translators between natural language and structured Chronicle queries.","marker":"[20]"}],"fun_headline_variants":["XR scenes adapt based on your evolving identity","Closed-loop XR reasons over a live model of you","PAiR: your identity drives immersive XR contexts","Chronicles as the reasoning core for adaptive XR"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The framework assumes a person's digital footprint can be turned into a reasoning-ready record accurate enough that the system's queries find the right memory or preference—an accuracy the paper does not measure.","fun_headline_variants_meta":{"raw":{"variants":["XR scenes adapt based on your evolving identity","Closed-loop XR reasons over a live model of you","PAiR: your identity drives immersive XR contexts","Chronicles as the reasoning core for adaptive XR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000188,"raw_usage":{"total_tokens":1279,"prompt_tokens":837,"completion_tokens":442,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":453,"completion_tokens_details":{"reasoning_tokens":379}},"tokens_in":453,"tokens_out":442,"duration_ms":5843,"temperature":1.0,"reasoning_tokens":379,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:43:03.375345+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run two users through the same XR scene with identical prompts but deliberately different Chronicles, one with the pivotal memory removed; if the rendered scene is identical, the loop is not actually driven by the Chronicle. Separately, feed a recorded set of facial-expression frames through the Situation and Context Detectors and count how often the produced semantic triples match a human labeler's judgment; if accuracy is near chance, the emotional-cue branch of the loop has no reliable input.","supporting_citations":[{"cited_title":"In: Proc","cited_arxiv_id":null,"evidence_quote":"Supplies the Chronicle construction pipeline—multimodal encoding, temporal segmentation, and knowledge-graph learning—that PAiR assumes as its identity model."},{"cited_title":"Information15(12), 766 (2024)","cited_arxiv_id":null,"evidence_quote":"Defines the Situation Graph ontology and the embedding and query mechanics that the Reasoner uses to ground user states semantically."},{"cited_title":"TechRxiv (2024)","cited_arxiv_id":null,"evidence_quote":"Argues that Chronicles can be shared to synthesize experiences from others' perspectives, the capability PAiR extends into XR."},{"cited_title":"In: AAAI 2021 Symposium on Combining Machine Learning and Knowledge Engineering (2021)","cited_arxiv_id":null,"evidence_quote":"Provides the neural-symbolic translation approach behind the LLM Scribe's conversion of language and sensed cues into symbolic scripts."}],"review_version":1}