{"id":"219abc2c-367d-4a70-a445-8e03aef0eaf5","arxiv_id":"1908.08532","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper presents a systematic, step-by-step calibration procedure for an AR haploscope, based on a decade of the authors' lab experience.","lead":"This paper describes how to build and calibrate an augmented reality haploscope, an optical device that superposes virtual images on the real world. It offers a step-by-step procedure for aligning lenses, beamsplitters, and users' eyes, a resource the authors say has been missing from the literature.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 3, Eq. 1 inverts the vergence-angle ratio, so following the printed procedure rotates each wing through the wrong angle and breaks the claimed collinearity of optical axes at finite viewing distances.","rationale":"The reader's condition is reasonable, but the weakest spot is stronger than the schematic-eye assumption. The paper claims a general invariant 'regardless of rotation'; that invariant is a geometric proposition about the angle at which each wing is locked. The printed formula for that angle is the cotangent rather than the tangent of the required half-convergence angle. Since the formula is used to set a key degree of freedom, the central claim is not supported as written. This can be fixed with a one-line correction (and ideally a validation measurement), so a conditional verdict is apt. I disagree with the reader's choice of weakest assumption: the schematic-eye model is explicitly stated and is a common approximation; the Eq. 1 inversion is an internal error that does not need any outside evidence to be exposed. Credit where due: the paper is transparent about limitations in Section 5 and provides substantial procedural detail; those are real strengths but do not cure the geometry problem.","tokens_in":6618,"tokens_out":11549,"duration_ms":112413,"concrete_test":"Use d = 1 m and IPD = 0.065 m. The printed Eq. 1 gives α ≈ 88.1°, while the inward rotation needed to intersect the target is θ = arctan((IPD/2)/d) ≈ 1.9°. Ray-trace (or draw) two wing axes separated by 0.065 m rotated inward by each angle; only the corrected angle makes both axes pass through a point 1 m in front of the user. A revised Eq. 1 or a worked numerical example would settle whether the paper's central alignment procedure is internally consistent.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central invariant ('each point on the user's optical axes... is presented with the same optical stimuli, regardless of rotation') requires the haploscope wing to be rotated to the same convergence angle as the user's eye. Section 3 specifies that angle as α = arctan(object distance / (IPD/2)). As written, this is the angle between the line of sight and the interocular baseline: it approaches 0 as the object approaches the eyes and approaches 90° as the object recedes to infinity. The inward rotation of one eye needed to fixate a target at distance d is θ = arctan((IPD/2)/d), which is the reciprocal of the printed fraction. For example, at d = 1 m and IPD = 0.065 m, the printed value is about 88.1° (near perpendicular to the baseline), while the required half-convergence angle is about 1.9°. Rotating the rails by the printed angle would send the two wing axes far behind the user rather than to the intended fixation point, so the optical axes would not remain collinear with the user's axes. This is an internal geometric error independent of the schematic-eye assumption; it is not merely a missing validation.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper describes the design, assembly, and calibration of an augmented reality haploscope, an optical bench device that presents virtual images at controlled accommodative and vergence distances while allowing the user to see the real world. The authors claim to provide the first systematic, published account of such calibration procedures. The central geometric claim is stated in Section 3: because each haploscope wing rotates about a point coincident with the modeled center of rotation of the user's eye, the optical axis of each wing remains collinear with the user's optical axis for any rotation angle, so that 'each point on the user's optical axes, including the nodal point, is presented with the same optical stimuli, regardless of rotation.' The calibration procedure consists of five steps: leveling the table and establishing a laser-based coordinate plane; centering optical elements and monitors on the rail axes; collimating the monitor image using the thin lens equation and verifying with a dioptometer; adjusting the beamsplitters so that the reflected optical axis passes through the pivot point; and adjusting and verifying the interpupillary distance (IPD) for each user. The paper is written as a procedural manual and includes photographs, a ray diagram, and references to prior theses and publications from the authors' laboratory.","tokens_in":6835,"tokens_out":3988,"duration_ms":37635,"significance":"If correct, the paper would provide a detailed, reproducible recipe for a laboratory instrument that can precisely control stereoscopic presentation parameters, which is valuable for research on depth perception, accommodation-vergence mismatch, and AR perception generally. The authors are honest about the scope: they describe an existing device and its calibration, not new perceptual results. The procedural detail is substantial, and the use of standard optical tools (laser level, dioptometer, thin lens equation) grounds the method in established practice. However, the significance depends entirely on the correctness of the geometric relationship between the rotation angle of the haploscope wings and the user's convergence. As described below, that relationship is stated incorrectly in Eq. (1), and this error is load-bearing for the central invariant claimed in Section 3. In addition, the paper contains no quantitative validation of the claimed alignment accuracy, despite the word 'Measurement' in the title. With the geometric error corrected and the validation gap addressed, the paper could be a useful methods reference.","major_comments":[{"comment":"Equation (1) defines alpha = arctan(object distance / (IPD/2)), which is the angle between the line of sight and the interocular baseline, not the inward rotation angle of the eye (or haploscope wing) needed to fixate a target at a given distance. For a target at distance d, the required half-convergence angle is theta = arctan((IPD/2)/d). For example, with d = 1 m and IPD = 0.065 m, the printed formula gives approximately 88.1 degrees, while the correct inward rotation is approximately 1.9 degrees. Following the printed procedure would rotate each wing nearly parallel to the baseline, directing the optical axes far to the side rather than at the intended fixation point. This inverts the geometric relationship and, as written, breaks the central claim that the wing axes remain collinear with the user's optical axes. The formula must be corrected and the surrounding text (including the description of alpha as 'the vergence angle or angle of binocular parallax') clarified to state whether alpha is the half-angle or the total angle between the two lines of sight.","section":"Section 3, Eq. (1)"},{"comment":"The central invariant stated in Section 3—'each point on the user's optical axes, including the nodal point, is presented with the same optical stimuli, regardless of rotation'—is asserted without any quantitative verification. The paper's title promises 'Measurement,' but no measurement of alignment accuracy, residual angular error, or verification of optical axis coincidence is reported. Section 5 itself concedes that 'refraction, tracking errors, optical distortions, and measurement errors could all potentially be detrimental to the accuracy of the final calibrated haploscope.' Since the paper's contribution is a calibration procedure intended to guarantee a precise geometric relationship, the absence of any validation data (e.g., laser-alignment measurements at multiple rotation angles, or an error budget) leaves the central claim unsupported. At minimum, the authors should either provide such validation or explicitly frame the paper as a methods description whose verification is future work, and adjust the title and abstract accordingly.","section":"Section 3 and Section 5"}],"minor_comments":[{"comment":"The abstract and conclusions state that the paper describes 'measurement' of the haploscope, but Section 5 (Future Work) is the only place where measurement errors are discussed; there is no measurement section or reported measurement data. The scope should be stated consistently.","section":"Abstract and Section 6"},{"comment":"The description of the optical path (e.g., 'a -10 diopter concave lens, producing a minified virtual image 5 cm in front of the lens') would be clearer if the sign conventions for object and image distances in the thin lens equation were defined explicitly, since these distances are central to the collimation step in Section 4.3.","section":"Section 3, optical path description"},{"comment":"The collimation step relies on the thin lens equation but does not state the actual distances used or the method for positioning the lenses (e.g., whether they are set to the back focal distances of the respective lenses). Without these values or a reference to a specific figure or table, the reader cannot reproduce this step without consulting the cited thesis [10].","section":"Section 4.3"},{"comment":"References [14] and [15] appear to describe the same paper (same title, same authors, same journal) with different citation details; this duplication should be resolved.","section":"References [14] and [15]"},{"comment":"The phrase 'shining straight down the center of the rail' is ambiguous; 'straight along the center of the rail' would be clearer.","section":"Section 4.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is a methods-oriented technical description. Its novelty is modest—it is a detailed account of an existing apparatus—but such accounts can be useful to the community if the technical content is correct. The Eq. (1) error is a serious, load-bearing flaw that must be corrected before the paper can be considered; the lack of validation is also a significant gap, though it could be addressed by reframing the paper's scope. I recommend major revision and careful re-review of the geometry."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis is a methods paper about building and calibrating an AR haploscope. The useful thing here is the detail: the step-by-step assembly and calibration with laser levels, dioptometer, IPD target, and the explicit assumption of a schematic eye. I believe the authors when they say they couldn't find a previous paper that spells out these details; this kind of procedural knowledge is normally lab folklore. If you are about to build such a device, this is the closest thing to a recipe you'll find.\n\nThe writing is clear and the authors are honest about the simplifications (schematic eye model, possible error sources in Future Work). The use of standard optics (thin lens, collimation) is unobjectionable.\n\nNow the problem. Equation 1 in Section 3 defines α = arctan(object distance / (IPD/2)). That is the angle between the line of sight and the interocular baseline, not the angle through which each eye rotates from the primary position. For a target at 1 m with IPD 65 mm, the printed value is about 88°, while the actual half-convergence rotation is about 1.9°. The text explicitly calls α the angle the eyes rotate to and says the haploscope wings must rotate to that α. If a reader follows that, the wings end up pointing nearly perpendicular to the intended target, and the claimed collinearity of optical axes (the central invariant of the design) breaks completely. This isn't a validation gap; it's an internal geometric error. The irony is that the later calibration procedure (Section 4.5) uses a target placed at IPD separation and laser levels to align the rails, which does not explicitly rely on Eq. 1. So the practical steps might actually be right, but the paper as written mixes a correct mechanical alignment procedure with an incorrect theoretical framing. A referee should demand both a corrected equation and an explicit statement of how the wing rotation angle is computed in practice.\n\nThe second soft spot is the lack of quantitative validation. The paper claims the calibrated instrument aligns the optical axes, but reports no measurement of the achieved accuracy. For a methods paper, a simple verification—e.g., measuring angular error over a range of distances—would make the recipe trustworthy.\n\nWho is this for? Labs that need a haploscope and want to replicate a working setup. It is niche, but for that audience the detail is valuable. It deserves peer review, but it should come back with major revisions: fix Eq. 1, clarify how the rotation angle is actually set in the assembly, and add whatever accuracy data they can collect. Without the fix, the central claim does not hold as written.","headline":"A rare, detailed haploscope calibration write-up that contains a genuine geometric error in its vergence-angle formula—fixable, but the paper as written would misdirect anyone who follows Eq. 1.","tokens_in":7331,"tokens_out":6943,"would_cite":false,"duration_ms":67005,"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":"A five-step calibration procedure aligns an AR haploscope's optics with the viewer's eye.","keywords":["augmented reality","haploscope","calibration","optical see-through display","vergence","interpupillary distance","schematic eye","stereoscopic vision"],"falsifier":"A direct test: place a camera with its entrance pupil at the modeled nodal point and photograph a calibration target through each wing while rotating the wings across the full vergence range. If the target's image position or focus changes with rotation angle, or equivalently if the binocular parallax of the target shifts as the wings turn, then the claim that every point on the user's optical axis receives the same stimulus regardless of rotation fails at that rotation.","tokens_in":6434,"feed_emoji":"👁","tokens_out":16570,"duration_ms":131236,"temperature":0.7,"pith_summary":"This paper gives a step-by-step recipe for building and calibrating an augmented reality haploscope, an optical instrument that superimposes precisely controlled virtual images onto the user's view of the real world. The central claim is that the five-step calibration makes each rotating wing's optical axis collinear with the user's modeled optical axis, so that every point on the user's visual axis, including the nodal point, receives the same optical stimulus regardless of how the wings rotate. The authors report that they have not found any earlier publication describing haploscope calibration at this level of detail, which would make this the first systematic account. If correct, the paper gives other perception laboratories a replicable method for running AR experiments on depth perception, accommodation-vergence conflict, and related visual phenomena.","feed_headline":"Five-step recipe aligns AR haploscope optics with the eye","feed_subtitle":"Lets perception labs overlay virtual images on the real world with controlled focus and binocular angle.","key_machinery":"The argument is carried by a three-stage optical chain and a laser-defined calibration coordinate system. The monitor image passes through a $-10$ diopter minimization lens that creates a minified virtual image $5$ cm in front of the lens; a $+10$ diopter collimating lens placed $10$ cm away turns that image into collimated light, and an accommodation lens of adjustable negative power refocuses it to a chosen focal distance before a beamsplitter reflects it into the user's eye, while lens positions follow the thin-lens equation $\\frac{1}{f} = \\frac{1}{u} + \\frac{1}{v}$. The vergence angle is set by $\\alpha = \\arctan\\left(\\frac{\\text{object distance}}{\\mathrm{IPD}/2}\\right)$, and the wings rotate about the modeled eye position so the system's optical axes stay collinear with the user's. The load-bearing model is the schematic eye — a center of rotation, a single nodal point, and a pupil, all on one line — and the ground truth of the whole calibration is the gravity-balanced laser level, which defines a coordinate plane through which each rail's optical axis, each beamsplitter's reflected ray, and the user's modeled axis are brought into coincidence.","core_discovery":"On the paper's own terms, the discovery is procedural: an AR haploscope assembled from off-the-shelf optics can be calibrated so that its optical axes remain collinear with a user's modeled optical axis across the full range of vergence rotation. The calibration has five discrete steps: level the table and establish a laser-defined coordinate plane; center every optical element on its rail's axis; collimate the monitor image and verify it with a dioptometer; adjust each beamsplitter so the reflected ray passes through the wing's pivot point; and set the interpupillary distance using two laser levels and a calibration target, a step that also re-verifies every earlier alignment. The authors claim primacy for the level of descriptive detail: after 180 years of haploscope use since the original stereoscope was reported in 1838, they have not previously found a paper describing these calibration specifics. The paper itself notes, in its future-work section, that refraction, tracking errors, optical distortions, and measurement errors still bound the accuracy of the final calibrated system.","pith_inferences":["Because the recipe is written against generic hardware concepts rather than proprietary parts, it plausibly transfers to other optical see-through displays with rotating optical paths, though the paper does not demonstrate such a transfer.","The schematic-eye premise sets an implicit boundary on the claim: real eyes have a slightly moving rotation center and more complex optics, so the 'same stimulus at every rotation' invariance should degrade at extreme vergence angles, and measuring that degradation would map the model's limits.","The paper's own list of residual error sources suggests a natural follow-on: quantifying refraction, distortion, tracking, and measurement errors would turn the recipe into a metrology with error bars, letting laboratories compare calibration quality across devices."],"forward_implications":["With this recipe, a research group can reproduce a calibrated AR haploscope from off-the-shelf parts without rediscovering the alignment steps.","Perceptual AR experiments that need independent control of focal demand and vergence angle, such as depth matching or accommodation-vergence studies, become repeatable procedures.","The final IPD verification doubles as a check on all earlier calibration steps, because two laser levels must simultaneously bisect the calibration target, each rail's centerline, and each wing's center of rotation.","Because the wings rotate about the modeled eye position, vergence angle can be changed without introducing the optical distortion that a fixed-axis display would produce."],"supporting_citations":[{"why":"Base design of a haploscope with rotatable optics and dynamic focus that this device iterates on.","marker":"[13]"},{"why":"Source of the detailed step-by-step calibration procedure the paper summarizes.","marker":"[10]"},{"why":"Underlies the schematic-eye assumption by specifying how the optical image on the retina is controlled.","marker":"[20]"},{"why":"Supplies a schematic eye model for virtual environments, supporting the collinear rotation-center, nodal-point, pupil assumption.","marker":"[7]"},{"why":"Documents the depth-perception errors that optical misalignment causes, motivating the interpupillary-distance calibration step.","marker":"[9]"},{"why":"Supports the design goal of rotating about the modeled eye position and using a beamsplitter to overlay real and virtual views.","marker":"[15]"}],"fun_headline_variants":["Five-step calibration aligns AR haploscope optics","AR haploscope: 5-step calibration recipe","Detailed AR haploscope calibration after 180 years","How to calibrate an AR haploscope in 5 steps"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the schematic-eye model: the calibration assumes a human eye behaves as if its center of rotation, a single nodal point, and a pupil all lie on one straight line, so if real eyes rotate about a moving center or have more complex optics, every aligned axis in the device inherits the mismatch.","fun_headline_variants_meta":{"raw":{"variants":["Five-step calibration aligns AR haploscope optics","AR haploscope: 5-step calibration recipe","Detailed AR haploscope calibration after 180 years","How to calibrate an AR haploscope in 5 steps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000173,"raw_usage":{"total_tokens":1254,"prompt_tokens":899,"completion_tokens":355,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":515,"completion_tokens_details":{"reasoning_tokens":294}},"tokens_in":515,"tokens_out":355,"duration_ms":3853,"temperature":1.0,"reasoning_tokens":294,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:47:14.148179+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test: place a camera with its entrance pupil at the modeled nodal point and photograph a calibration target through each wing while rotating the wings across the full vergence range. If the target's image position or focus changes with rotation angle, or equivalently if the binocular parallax of the target shifts as the wings turn, then the claim that every point on the user's optical axis receives the same stimulus regardless of rotation fails at that rotation.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Base design of a haploscope with rotatable optics and dynamic focus that this device iterates on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the detailed step-by-step calibration procedure the paper summarizes."},{"cited_title":"Westheimer","cited_arxiv_id":null,"evidence_quote":"Underlies the schematic-eye assumption by specifying how the optical image on the retina is controlled."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies a schematic eye model for virtual environments, supporting the collinear rotation-center, nodal-point, pupil assumption."},{"cited_title":"Lee and H","cited_arxiv_id":null,"evidence_quote":"Documents the depth-perception errors that optical misalignment causes, motivating the interpupillary-distance calibration step."},{"cited_title":"Singh, S","cited_arxiv_id":null,"evidence_quote":"Supports the design goal of rotating about the modeled eye position and using a beamsplitter to overlay real and virtual views."}],"review_version":1}