{"id":"e62b1dc7-30f4-49f6-be7a-70a1d43ee085","arxiv_id":"2508.18481","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Opaque targets and showing the real tool occluding the hologram improve depth accuracy and usability in optical see-through AR at arm's distance.","lead":"This paper reports a user study with ten participants testing how target transparency and tool visualization affect depth perception in optical see-through augmented reality on the HoloLens 2. Opaque targets and real-time occlusion by the real tracked tool produced the best depth accuracy and usability for arm's-length guidance.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Occlusion is confounded with real-tool visibility: the 'real tool' condition bundles occlusion with physical-tool presence and proprioception, so the causal claim that occlusion cues drive accuracy is not uniquely supported.","rationale":"This stress-test focused on the internal validity of the causal attribution. The central claim quoted by the reader says that opaque rendering and real-time occlusion by the real tool are critical. In Experiment 2, the 'real tool' condition is described as occluding the virtual target, but it also differs from the alternatives by the presence of a physical tool, the user's own hand, and natural occlusion cues from a real object at the task plane. The design therefore cannot tell whether 'occlusion handling' or 'real-tool visibility/proprioception' drove the effect. This is more load-bearing than the sample-size concern because even with a larger n the confound would remain. The proposed control condition would settle it. I found no internal inconsistency or evidence of misconduct; the abstract is clear and appropriately cautious about target transparency, but the occlusion conclusion overreaches the 2×3 design. Because no full text or data were available, the verdict should be conditional: the central claim should be accepted only if the occlusion-specific effect is demonstrated by an additional analysis or control condition.","tokens_in":773,"tokens_out":5351,"duration_ms":64728,"concrete_test":"Request the participant-level data and run a follow-up within-subjects condition with the same real tracked tool positioned immediately beside the target so it is visible but does not occlude it, with target opacity matching the low-transparency condition. If mean target-localization error remains comparable to the real-tool-occluding condition, the benefit is attributable to real-tool visibility/proprioception rather than occlusion; if error significantly increases without occlusion, the occlusion claim is supported. A cheaper analytical check: compare the virtual-tool-hologram condition to the real-tool condition after masking the tool region in recorded scene video—if the accuracy difference disappears when only the occluding silhouette is considered, occlusion is the key factor.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim's causal core is that correct occlusion (opaque target occluded by the real tool in real time) is critical for depth perception and precision. In Experiment 2, the 'real tool' mode provides the actual physical tool, and because the tool is held in front of the virtual target, it also occludes it. The 'virtual tool hologram' and 'no tool tracking' conditions lack the real tool entirely. Thus the independent variable 'tool visualization mode' simultaneously changes: (a) presence/visibility of the real tool, (b) occlusion of the virtual target by the real tool, (c) proprioceptive and haptic feedback from the user's own hand, and (d) vergence/accommodation cues from a real object at the task plane. Any one of these could explain the reported accuracy and usability advantage. The paper's own phrasing—'showing the real tool (occluding the virtual target)'—acknowledges the bundle but does not isolate occlusion. Without a condition where the physical tool is present but does not occlude, or a virtual tool that occludes identically, the headline recommendation to 'prioritize robust tool tracking and occlusion handling' overstates what the 2×3 design can establish. The n=10 sample and the absence of reported effect sizes/confidence intervals compound the concern, but the confound is the load-bearing issue.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports a two-experiment user study on optical see-through AR (HoloLens 2) at arm's distance. Experiment 1 compares high- vs. low-transparency target rendering in a depth matching task; Experiment 2 uses a simulated surgical pinpoint task on a frontal bone model with a 2×3 design (two target transparencies, three tool visualization modes: virtual tool hologram, real tool, and no tool tracking). The authors report that low-transparency targets produce significantly lower depth estimation error, that the real-tool condition (which occludes the virtual target) yields the highest accuracy and usability and the lowest workload, and that no tool tracking yields the worst outcomes. They conclude that correct occlusion, opaque rendering, and real-time tool tracking are critical for depth perception and precision in OST-AR.","tokens_in":1126,"tokens_out":2399,"duration_ms":33117,"significance":"If the findings hold, the paper provides practical, easily actionable guidance: rendering virtual targets opaque and ensuring real tools occlude them can improve depth perception in hand-scale AR tasks. The work is empirical rather than theoretical, with the strength of using a commercial OST-AR device and two complementary tasks (perceptual matching and a precision interaction task). The comparison across tool visualization modes is a useful contribution to a space where occlusion cues are often neglected. However, the causal claim about occlusion is not uniquely established by the experimental design, and the small sample size and lack of reported precision metrics temper the strength of the conclusions.","major_comments":[{"comment":"The real-tool condition bundles several independent cues: presence of the physical tool, occlusion of the virtual target by that real tool, proprioceptive/haptic feedback from the user's hand, and vergence/accommodation from a real object near the task plane. The virtual-tool and no-tracking conditions differ from it along all of these dimensions at once. Therefore the paper's central conclusion that 'correct occlusion cues' are the critical factor is not uniquely supported; the advantage could stem from any component of the bundle. To support the stated design recommendation, the authors need either an experimental condition that separates occlusion from tool presence (e.g., a real tool that does not occlude, or a virtual tool with identical occlusion geometry) or a substantially more cautious causal interpretation.","section":"Experiment 2 (abstract)"},{"comment":"The abstract repeatedly uses 'significantly' without reporting effect sizes, confidence intervals, or the specific statistical tests and multiple-comparison corrections used for the six-condition comparison. With n=10 participants, the precision of the estimates is a load-bearing concern; the strongest headline claims (e.g., 'lowest workload,' 'worst performance') should be accompanied by quantitative effect sizes and uncertainty bounds. If this information appears later in the full text, the abstract should summarize it; otherwise it should be added.","section":"Abstract / Results"}],"minor_comments":[{"comment":"The transparency levels are described only as 'high' and 'low'; please report the actual transmittance or opacity values used, as this is essential for replicating the rendering conditions.","section":"Abstract / Method"},{"comment":"Clarify how the 'virtual tool hologram' condition was implemented relative to the real tool: was the same geometric model used, and was its occlusion behavior matched to the real tool? This would help readers assess what the virtual tool condition isolates.","section":"Experiment 2"},{"comment":"The phrase 'showing the real tool (occluding the virtual target)' in the abstract is a characterization of the condition, not an analysis of which cue is responsible. Please rephrase to avoid implying that occlusion is the sole manipulated factor.","section":"Experiment 2"},{"comment":"The qualitative claim that 'making the target highly transparent... slightly impaired depth cues' is presented as if quantitative; if this is based on user comments, say so and avoid mixing qualitative impressions with significance statements.","section":"General"},{"comment":"Provide details on counterbalancing/order of the six conditions, since learning and fatigue effects are plausible in a within-subjects design with n=10.","section":"Method"},{"comment":"The external validity of the simulated surgical pinpoint task is acknowledged as a limitation in spirit but should be stated explicitly; arm's-length depth perception in a lab may not generalize to real surgical or industrial workflows.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The abstract alone would not normally support a full review; my assessment is based on the provided abstract and the described design. The confound in Experiment 2 is the central issue: unless the full methods already contain a separation of occlusion from real-tool presence, the conclusion as stated overreaches. If the full paper has additional conditions or a more modest framing, the manuscript may be salvageable with a major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper has a useful empirical finding hiding behind a causal claim it doesn't earn. The 2x3 design comparing target transparency and tool visualization is a reasonable extension of prior AR depth work, and the basic result—less transparent targets improve depth matching at arm's length—is plausible and consistent with existing literature. The usability data, with the real tool condition scoring best and no-tracking worst, is also believable.\n\nWhat's new: a concrete comparison of six combinations in a simulated surgical task. That's worth having.\n\nThe soft spot is the confound. The 'real tool' condition changes several things at once: physical tool presence, occlusion, haptic/proprioceptive feedback, and real-object vergence cues. Attributing the accuracy gain specifically to occlusion cues is not supported by the design. The abstract itself says 'showing the real tool (occluding the virtual target),' which acknowledges the bundle. To isolate occlusion you'd need a condition with the physical tool present but not occluding, or a virtual tool that occludes identically. Without that, the recommendation to prioritize occlusion handling overstates the evidence.\n\nAlso n=10 with no reported error bars or effect sizes makes the significance claims unverifiable from the abstract. That's a reporting gap, not necessarily a fatal flaw, but it matters for a between-subjects or mixed design.\n\nThe paper is for AR/surgical guidance designers. It deserves a serious referee: the question is practically important and the data are honest measurements, not curve-fitting. I'd send it out, but I'd ask the authors to either add a control condition that separates occlusion from tool presence or soften the causal language.\n\nRecommendation: engage, but treat the occlusion claim as a hypothesis until the confound is addressed.","headline":"Opaque targets help depth perception in OST-AR, but the real-tool condition bundles occlusion with physical tool presence, so the headline causal claim is overreach.","tokens_in":1511,"tokens_out":1165,"would_cite":false,"duration_ms":13251,"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":"Opaque target rendering and real-tool occlusion reduce depth errors in optical see-through AR at arm's length.","keywords":["optical see-through AR","depth perception","occlusion","transparency","usability","HoloLens 2","surgical guidance","tool tracking"],"falsifier":"Run a preregistered replication with 40+ participants and a realistic surgical drill task on cadavers or phantoms; if opaque target rendering with real-tool occlusion fails to produce lower depth error and higher usability than a transparent target, the central claim is false. Alternatively, a field study in an operating room comparing error rates under the same visualization conditions would settle it.","tokens_in":773,"feed_emoji":"🥽","tokens_out":2764,"duration_ms":28917,"temperature":0.7,"pith_summary":"The paper tests how target transparency and tool visualization affect depth perception and usability in optical see-through AR (HoloLens 2) for arm's-distance tasks. Across two experiments with ten participants, it finds that rendering the target opaque rather than highly transparent significantly lowers depth estimation error, and that showing the user's real tracked tool occluding the virtual target produces the highest accuracy and usability with the lowest workload. Failing to track the tool yields the worst performance. The authors argue that correct occlusion cues and opaque rendering are critical for precision in OST-AR, and that transparency should be used only when tool tracking is unavailable.","feed_headline":"Opaque rendering plus real-tool occlusion wins for AR depth","feed_subtitle":"In see-through headsets, hiding holograms behind real tools beats transparency for precision at arm's length.","key_machinery":"The experimental manipulation is the central object: a 2x3 design crossing target transparency (high vs. low) with tool visualization mode (virtual hologram proxy, real tracked tool, no tool tracking). The load-bearing mechanism is occlusion: the real tool hiding part of the target in real time, combined with opaque target rendering, provides a monocular occlusion cue that anchors depth judgment at near-field arm's distance.","core_discovery":"The central discovery is a systematic ordering of visualization conditions: at arm's distance, a low-transparency (opaque) holographic target is estimated with significantly lower depth error than a high-transparency target, and the best overall performance comes from rendering the virtual target opaque while letting the real tracked tool pass in front of it, occluding part of the hologram. The real tool condition outperforms both a virtual tool proxy and a no-tracking condition on localization error, system usability, and perceived workload. The paper interprets this as evidence that occlusion of the virtual target by the real tool supplies the strongest depth cue among the tested options,","pith_inferences":["The occlusion advantage likely generalizes to any manipulated tool (e.g., pointers, drills, styluses) at arm's distance, not just surgical pinpoints.","A larger sample with more diverse participants may reveal interactions between transparency and tool mode, since the paper notes the transparent-target/real-tool condition slightly impaired depth cues.","The findings suggest that future OST-AR systems should expose occlusion handling as a first-class API, not an optional rendering flag.","A testable extension: measure vergence or accommodative responses to see whether occlusion reduces depth error by easing stereo-conflict, which would explain the observed error reduction."],"forward_implications":["OST-AR designers should render targets opaque and invest in real-time occlusion of virtual content by real tracked tools.","If robust tool tracking is unavailable, transparency can be a fallback, but it will not recover depth accuracy or usability.","Depth estimation errors increase when a target is highly transparent, so see-through rendering should not be used for precision-critical tasks.","Surgical and industrial guidance systems at arm's length should prioritize tool tracking over transparency features.","User workload and usability ratings align with objective accuracy, so subjective ratings can serve as a proxy for depth-quality failures."],"supporting_citations":[],"fun_headline_variants":["Opaque holograms plus real tools improve AR depth precision","Real-tool occlusion beats transparency for AR depth","Occlude holograms with real tools for best AR depth","Real tool occluding opaque target yields best AR precision"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The results rest on ten participants performing two controlled laboratory tasks, and if those tasks do not capture the visual and motor demands of real arm's-distance AR use, the reported differences may not transfer to actual surgery or industrial work.","fun_headline_variants_meta":{"raw":{"variants":["Opaque holograms plus real tools improve AR depth precision","Real-tool occlusion beats transparency for AR depth","Occlude holograms with real tools for best AR depth","Real tool occluding opaque target yields best AR precision"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000588,"raw_usage":{"total_tokens":2648,"prompt_tokens":843,"completion_tokens":1805,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":587,"completion_tokens_details":{"reasoning_tokens":1739}},"tokens_in":587,"tokens_out":1805,"duration_ms":13759,"temperature":1.0,"reasoning_tokens":1739,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T16:25:59.050341+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a preregistered replication with 40+ participants and a realistic surgical drill task on cadavers or phantoms; if opaque target rendering with real-tool occlusion fails to produce lower depth error and higher usability than a transparent target, the central claim is false. Alternatively, a field study in an operating room comparing error rates under the same visualization conditions would settle it.","supporting_citations":[],"review_version":1}