{"id":"51775ac6-2140-4312-9fd1-42226c197267","arxiv_id":"2505.05147","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Two stacked gratings in a contact lens form moiré fringes that shift with viewing angle, giving a measured 0.28 degree resolution for lens orientation from ordinary photographs.","lead":"A contact lens with two stacked microscopic gratings creates moiré fringes whose shift reveals the lens's viewing angle, and the authors report 0.3 degree accuracy from ordinary camera photos. This passive marker could give AR/VR headsets a low-power, light-independent way to track eye position.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed 0.28° precision ignores the angular gradient across the finite-size label; at 40 cm the gradient bias is the same order as the claimed error and grows closer to the eye.","rationale":"Both the reader and this pass identify the same soft point: the precision figure is computed from a model that treats the observation angle as constant across the label. The paper's own text flags the gradient as a known but unquantified effect, and the Fourier phase extraction averages over the strip, so the measured period and phase are affected by the chirp. A first-order derivation shows that the normalized shift is only invariant to a common scale factor; since the two patterns have different moiré periods (1.1 mm vs 0.5 mm), the gradient produces a bias that is not absorbed by the linear calibration. The magnitude is plausibly 0.2-0.3° at the experimental 40 cm distance, i.e., of the same order as the reported 0.28°, and it would grow with proximity to the camera, undermining the AR/VR deployment claim. This does not refute the central physical mechanism; a lens sample was clearly demonstrated. But the precision number is not yet supported. The reader's CONDITIONAL verdict is appropriate, with the requested condition being a quantitative gradient analysis or a cropped-window validation. I see no reason to move the verdict; UNCHANGED.","tokens_in":14151,"tokens_out":14861,"duration_ms":150035,"concrete_test":"Re-analyze the original 31 photographs with the analysis window for each moiré pattern cropped to the central 20% of its width (reducing the angular span by ~5x), using the same Fourier algorithm, and compare the estimated angles to the full-window values at the same true angles. If the median absolute shift exceeds 0.1° or the RMSD changes by more than 30%, the finite-angular-size gradient materially affects the precision claim; if the shift is negligible, the concern is refuted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is the unquantified finite-angular-size effect. The precision estimate of 0.28° (Section 'Angle measurements and precision estimation', Eq. (2)) is derived under the zero-angular-size approximation, which the authors explicitly invoke in 'Label operating principle' and Supplementary Note 1 ('assume the camera or observer is positioned at a sufficient distance ... all its parts are viewed at the same angle'). At the experimental distance of 40 cm, a label a few millimeters wide subtends roughly a degree, i.e., the same order as the claimed precision. Because the two patterns used in Eq. (2) have different periods (a1~1.1 mm, a4~0.5 mm), a linear gradient in tan(Θ) across the label does not cancel in the normalized relative shift (x4-x1)/(a1+a4): the apparent period and phase extracted by the Fourier algorithm (Note 2) are each biased by a factor that depends on the local moiré period, leaving a residual term of order (H/pA)*g*(a1+a4) in angle. A rough estimate for H=250 μm, pA=31.6 μm, d=400 mm gives a bias near 0.2-0.3°, comparable to the reported RMSD. The authors themselves list 'accounting for moiré pattern gradients' as future work, confirming this effect is not included in the 0.28° figure. If the bias is real, the experimental precision is worse than claimed, and the extrapolation to close-range AR/VR cameras (where the gradient is several times larger) is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a passive optical eye-tracking label for contact lenses, consisting of two superimposed gratings with a slight period mismatch. The observation angle is encoded in the parallax-induced relative shift of the moiré fringes, as described by Eq. (2). The authors fabricate a PDMS contact lens with four moiré patterns, image it from ~40 cm on a rotation stage, and extract fringe shifts and periods by Fourier analysis. They report an averaged RMS deviation of 0.28° between true and estimated angles and claim an angular resolution better than 0.3°, with potential for further improvement, targeting AR/VR eye-tracking applications.","tokens_in":14455,"tokens_out":5057,"duration_ms":48483,"significance":"If the central claim holds, the method offers a simple, passive, illumination-independent alternative to camera-based eye tracking that avoids infrared illumination, scale bars, and perspective correction. The theoretical derivation in Supplementary Note 1 is clear, and the experimental realization involving a laser-engraved bilayer grating in a contact lens is a substantial technical effort. The multi-pattern differential design is a sensible way to remove common-mode errors. However, the precision claim is currently supported only by an in-sample calibration residual, and the zero-angular-size approximation is violated at the experimental distance in a way that is acknowledged but not quantified. The approach is promising, but the metrological validation needs strengthening.","major_comments":[{"comment":"The reported σ_Θ = 0.28° is the root-mean-square deviation of the 31 measured angles from a linear fit performed on the same 31 data points, and the gap H is estimated from the slope of that same fit. This makes the reported error an in-sample calibration residual rather than a prediction error on independent angles. To substantiate the claimed 0.3° precision, the authors should validate with a held-out procedure (e.g., fit on one half of the angular range and test on the other) or use an independently measured H and phase offsets, and report the resulting out-of-sample error.","section":"Angle measurements and precision estimation (Fig. 4)"},{"comment":"The zero-angular-size approximation is explicitly assumed in Note 1 ('all its parts are viewed at the same angle'), but at the experimental distance of 40 cm, a label a few millimeters wide subtends approximately 0.5°–1°, i.e., the same order as the claimed 0.28° error. The authors acknowledge the gradient issue in 'Label operating principle' and list 'accounting for moiré pattern gradients' as future work, but they do not quantify its effect on the Fourier-extracted phase or on the normalized shift in Eq. (2). Because the four subgratings have different periods, a linear gradient in tan Θ across the label does not cancel in (x4-x1)/(a1+a4); a rough estimate for H = 250 μm, p_A = 31.6 μm, and a 1° angular span gives a bias of order 0.2°–0.3°, comparable to the reported precision. This systematic effect must be quantified or removed before the 0.28° figure can be accepted as the method's precision.","section":"Label operating principle; Supplementary Note 1"},{"comment":"The experimental demonstration is performed at a camera distance of ~40 cm, while the claimed AR/VR applicability would place the observation camera much closer to the eye (centimeter scale). At such distances, the angular gradient across the label is several times larger, further violating the zero-angular-size approximation and the paraxial linear model. No experiment, simulation, or error analysis is provided at AR/VR-relevant distances, so the statement that the method 'meets typical AR/VR application requirements' is not supported by the presented data.","section":"Discussion; Measurements"}],"minor_comments":[{"comment":"Eq. (2) uses p_B in the prefactor, while the derivation in Note 1, Eq. (S4), gives p_A. Since p_A and p_B differ by up to 6%, the notation should be made consistent or the approximation stated.","section":"Eq. (2) vs Supplementary Note 1"},{"comment":"The abstract states 'angular resolution exceeding 0.3°', while the Discussion says 'angular precision better than 0.3°'. Resolution, accuracy, and precision are distinct terms; the manuscript should use them consistently.","section":"Abstract and Discussion"},{"comment":"The text states the error is 'approximately 1/25 of the label's angular period' in one place and '1/37' in the Discussion; these values should be reconciled.","section":"Results, precision paragraph"},{"comment":"The sentence 'even if the image of the label is distorted due to large viewing angles or some optical aberrations these distortions to find the normalized relative shift between gratings' is incomplete and should be rewritten.","section":"Results, last paragraph before Fig. 4"},{"comment":"The fabricated stack is described as 400-µm-thick PDMS, but the inferred optical gap is H ≈ 250 μm. The relationship between the physical stack thickness, the ink layers, and the effective optical gap should be clarified, since H is a key parameter in Eq. (2).","section":"Materials and Methods, Fabrication"},{"comment":"The claim of 'enhanced precision of approximately 0.2° or better for a narrow angle range' is not quantified; please specify the angle range and the method used to derive this value.","section":"Fig. 4d"}],"recommendation":"major_revision","confidential_remarks":"The manuscript presents a plausible and technically interesting concept, but the central precision claim is currently supported only by an in-sample residual and an unquantified systematic effect. The authors should be asked to provide a held-out validation and to quantify the finite-angular-size bias before publication. The paper is within scope for the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a solid engineering demonstration, not a breakthrough. The genuinely new bit is using the relative shift between four moiré patterns with different mismatches as an angle readout, and showing sub-degree precision from a standard camera on a bench setup. That is a reasonable extension of existing moiré contact lens work (IOP sensors, refs 34 and 36), and the paper is honest that the mechanism is classical.\n\nThe authors get credit for a clean experiment: rotation stage as independent ground truth, 31 images, Fourier extraction of phase and period, and explicit discussion of limitations like the zero-angular-size approximation and the fact that gradient handling is future work. The rendering comparison in Fig. 3 is a nice check that their interpretation of the fringe shifts is right.\n\nSoft spots, in rough order of importance. First, the 0.28° (and 0.41° for the single pair) is the RMS deviation of the known stage angles from the linear fit on the same data. That is an in-sample residual, not a validation on held-out data. It tells you how well the linear model fits, not how accurately you can predict an unknown angle. The paper does not make that distinction sharply. Second, the finite angular size of the label at 40 cm spans about 1°, the same order as the claimed precision. The authors acknowledge the gradient in principle but do not quantify its effect on the extracted Fourier phase. The stress-test note's estimate that this could bias the result by 0.2–0.3° is plausible; it does not destroy the demonstration, but it means 0.28° is optimistic. Third, the flat-interface Snell model for a curved contact lens is an approximation, and there is no on-eye validation; the leap to 'satisfactory for gaze detection in AR/VR' goes beyond the data. Fourth, no data or code are released; for a measurement paper, that matters more than usual because the error analysis is the main result.\n\nThe citation pattern looks fine, and the self-citation to the patent is declared. I do not see a load-bearing flaw in the physics: the parallax-moiré mechanism is well understood and the bench data are consistent with it. The paper is best read as a proof-of-concept for a passive label that removes illumination and perspective-correction burdens, with precision still to be nailed down.\n\nWho is this for? Engineers working on lens-integrated AR/VR tracking and smart contact lens folks. It should go to peer review; a good referee will ask for held-out validation, a quantitative gradient analysis, and ideally an eye-model demonstration with realistic geometry. I would not cite it in my own work yet, but I would follow the follow-up.","headline":"A credible bench demonstration of a passive moiré-parallax contact-lens label for eye tracking; the 0.28° precision is a calibration residual and the unquantified angular-gradient bias likely makes it optimistic, but the core idea is sound and worth a serious referee.","tokens_in":15087,"tokens_out":1767,"would_cite":false,"duration_ms":17172,"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 passive label inside a contact lens encodes viewing angle in moiré-fringe shifts, recovering lens orientation with 0.28° accuracy from one photograph.","keywords":["eye tracking","contact lens","moiré pattern","parallax","augmented reality","virtual reality","angular measurement","Fourier phase"],"falsifier":"Compare angles recovered from moiré phase with the rotation stage while changing only the camera distance, say from 40 cm to 10 cm; if the recovered angle drifts at fixed stage angle, the label's finite angular size is biasing the phase. A more direct check is to crop the left and right halves of the label and compute their Fourier phases separately: if their implied angles differ by about the label's angular width, the zero-angular-size assumption is violated.","tokens_in":13947,"feed_emoji":"👁️","tokens_out":9755,"duration_ms":87498,"temperature":0.7,"pith_summary":"The paper proposes and tests a contact-lens eye-tracking label that works without infrared light, scale bars, colour bars, or perspective correction. The label is a passive stack of two gratings with slightly different periods; the gap between them creates parallax, so tilting the lens shifts the visible moiré fringes. Measuring the relative shift of four moiré patterns in a single photograph recovers the lens orientation, and the authors report an angular error of 0.28° after averaging four pattern pairs and better than 0.2° near the centre of the range. If this holds, AR/VR systems could track gaze and lens rotation with an ordinary camera and a cheap passive in-lens label, with the same platform offering a path to much finer precision.","feed_headline":"Moiré fringes on a contact lens track eye angle to 0.28 degrees","feed_subtitle":"A passive grating stack reads gaze direction from a standard camera, with no infrared light or scale bars.","key_machinery":"The central object is a bilayer grating stack with a deliberate period mismatch: a reference grating of period $p_A=31.6\\,\\mu\\mathrm{m}$ and a top grating divided into four zones with periods 29.7, 30.7, 32.5, and 33.5 $\\mu\\mathrm{m}$, separated by a gap $H\\approx250\\,\\mu\\mathrm{m}$. When the stack is viewed at an angle, the parallax between the two gratings shifts the moiré fringes; the operative quantity is the relative shift of two oppositely moving patterns $\\tilde{x}_4-\\tilde{x}_1$ normalized by the sum of their periods $a_1+a_4$. That normalized-relative-shift construction is what decouples the measurement from illumination, magnification, and perspective, and the four-zone design supplies four independent estimates whose average reduces the error from 0.41° to 0.28°.","core_discovery":"On its own terms, the paper establishes that the viewing angle of a contact-lens label can be recovered from the relative phase of moiré patterns rather than from any absolute feature. In the label, a reference grating of period $p_A=31.6\\,\\mu\\mathrm{m}$ is stacked with a top grating subdivided into four zones with periods 29.7, 30.7, 32.5, and 33.5 $\\mu\\mathrm{m}$, separated by $H\\approx250\\,\\mu\\mathrm{m}$. The resulting fringes have period $a_x=p_Ap_B/|p_B-p_A|$ and shift with the internal angle, giving $\\tan\\Theta_{\\mathrm{lens}}=\\frac{p_A}{H}\\frac{\\tilde{x}_4-\\tilde{x}_1}{a_1+a_4}+C$, where the constant absorbs fixed phase offsets. Because the angle depends on a dimensionless ratio of shift to fringe period, the readout is invariant under image scaling, label position, and mild perspective distortion. The paper reports that over a ±15° rotation range imaged at 40 cm, a single pair yields 0.41° RMS error, averaging the four opposite-shifting pairs gives 0.28°, and the central ±10° region gives about 0.2°.","pith_inferences":["Editorial inference: if the finite-angular-size bias identified in the paper is suppressed, the same scheme should reach well below 0.1°, moving it closer to the precision of scleral search coils without their invasiveness.","Editorial inference: the relative-phase readout is not inherently limited to rotation; translational slip of the lens along the eye also changes the parallax projection, so a label with more than four zones could separate rotation from translation in one image.","Editorial inference: the pairwise analysis is forced by uncontrolled random phases between grating zones; a fabrication process that fixes those phases would allow one global fit over all zones, improving noise averaging beyond simply averaging independent pairs.","Editorial inference: the multiple period mismatches form a Vernier scale; combining zones with different angular periods could yield an absolute, not merely incremental, angle encoder over a range larger than the single-pattern angular period."],"forward_implications":["Averaging the four opposite-shifting moiré pairs gives an angular error of 0.28°, and about 0.2° within ±10° of the range centre, which the paper states is sufficient for typical AR/VR gaze detection.","Because the angle is read as a dimensionless ratio of fringe shift to fringe period, the same algorithm works without scale bars, colour bars, or corrections for image perspective or magnification.","Phase-based readout makes the measurement insensitive to ambient lighting and camera exposure, removing the infrared illumination normally needed for video oculography.","The label is passive and read by an ordinary camera, so continuous eye tracking adds little power and no extra hardware to an AR/VR device that already has a camera.","Reducing the grating periods shrinks the angular period and therefore multiplies sensitivity; the paper estimates this could improve precision by more than an order of magnitude."],"supporting_citations":[{"why":"Prior demonstration that moiré patterns on a contact lens can encode a physical quantity; the present label adapts that idea to encode viewing angle.","marker":"[34]"},{"why":"Shows embedded moiré structures can be built into soft contact lenses, the fabrication route the present module follows.","marker":"[36]"},{"why":"A current lens-integrated eye-tracking solution against which the passive label approach is positioned.","marker":"[50]"},{"why":"Supplies the high-precision eye-movement benchmark that motivates the 0.3° accuracy target.","marker":"[52]"},{"why":"Prior contact-lens photodetector-based eye tracking; the present approach is contrasted by needing only an external camera.","marker":"[55]"},{"why":"Survey establishing AR/VR gaze tracking as the application driving the required precision and constraints.","marker":"[12]"}],"fun_headline_variants":["Contact-lens moiré fringes track eye angle to 0.28°","Passive moiré contact lens reads gaze direction with 0.28° accuracy","Moiré-pattern lens tracks eyes precisely without infrared or scale bars","Contact lens moiré yields sub-degree eye tracking for AR/VR"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the whole label is seen from one viewing angle; in the experiment a label a few millimetres wide at 40 cm spans about a degree of viewing angle, so points across it are actually viewed at slightly different angles, and the paper does not quantify how much this spreads the measured fringe phase.","fun_headline_variants_meta":{"raw":{"variants":["Contact-lens moiré fringes track eye angle to 0.28°","Passive moiré contact lens reads gaze direction with 0.28° accuracy","Moiré-pattern lens tracks eyes precisely without infrared or scale bars","Contact lens moiré yields sub-degree eye tracking for AR/VR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000676,"raw_usage":{"total_tokens":3079,"prompt_tokens":956,"completion_tokens":2123,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":572,"completion_tokens_details":{"reasoning_tokens":2038}},"tokens_in":572,"tokens_out":2123,"duration_ms":15562,"temperature":1.0,"reasoning_tokens":2038,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:11:56.407006+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare angles recovered from moiré phase with the rotation stage while changing only the camera distance, say from 40 cm to 10 cm; if the recovered angle drifts at fixed stage angle, the label's finite angular size is biasing the phase. A more direct check is to crop the left and right halves of the label and compute their Fourier phases separately: if their implied angles differ by about the label's angular width, the zero-angular-size assumption is violated.","supporting_citations":[{"cited_title":"C., Ho, C","cited_arxiv_id":null,"evidence_quote":"Prior demonstration that moiré patterns on a contact lens can encode a physical quantity; the present label adapts that idea to encode viewing angle."},{"cited_title":"& Que, L","cited_arxiv_id":null,"evidence_quote":"Shows embedded moiré structures can be built into soft contact lenses, the fabrication route the present module follows."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"A current lens-integrated eye-tracking solution against which the passive label approach is positioned."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the high-precision eye-movement benchmark that motivates the 0.3° accuracy target."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Prior contact-lens photodetector-based eye tracking; the present approach is contrasted by needing only an external camera."},{"cited_title":"B., MacNeilage, P","cited_arxiv_id":null,"evidence_quote":"Survey establishing AR/VR gaze tracking as the application driving the required precision and constraints."}],"review_version":1}