{"id":"7a1872e6-77d5-46cd-a5d9-eee3632d528f","arxiv_id":"2507.07890","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Hi-D maps visualize multi-dimensional categorical data by hierarchically cutting a regular polygon into area-proportional sub-polygons, one cut direction per dimension.","lead":"This paper introduces Hi-D maps, a new way to draw multi-dimensional categorical data inside a polygon by slicing it into smaller polygons, one slice per data dimension. It matters because it proposes a space-efficient alternative to treemaps, sunburst charts, and parallel sets, with interactive reordering and zooming.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central effectiveness claim — dimensional clarity plus treemap-like space efficiency — is asserted from design rationale but never measured; no user study or perceptual experiment supports it, so the core contribution remains unvalidated.","rationale":"I read the paper as a design contribution whose central claim is effectiveness: a new encoding that combines space efficiency with dimensional clarity. The algorithm's area proportionality is internally plausible: any convex polygon can be split along a fixed direction to realize arbitrary area fractions, so the construction itself is defensible. The load-bearing assumption is perceptual — that angle, hue, lightness, and marble counts can be decoded accurately enough to support real insight. The reader identified exactly this gap, and I agree. I find no internal inconsistency in the algorithm and no basis for rejection; the appropriate disposition is conditional acceptance pending empirical validation. This is not a manufactured concern: the paper's own conclusions list user studies and empirical comparisons as future work, and the related-work critique of treemaps ('level-counting') is itself an empirical claim that the paper does not substantiate. Thus the decisive test is a controlled user study measuring accuracy, speed, and learnability against the baseline techniques the paper compares itself to.","tokens_in":6516,"tokens_out":8519,"duration_ms":106487,"concrete_test":"Run a preregistered within-subjects study (N >= 24) comparing Hi-D maps, treemaps, sunburst, and parallel sets on identical 5- and 10-dimensional categorical datasets. Use three task types: (1) locate the region for a specified attribute intersection; (2) choose which of two highlighted intersections is larger; (3) identify which dimension and value a highlighted edge encodes. Measure accuracy and completion time, both without training and after a brief training session. If Hi-D maps do not significantly outperform treemaps on dimension-identification while matching or exceeding them on area-comparison, the claimed combination of treemap spatial efficiency and sunburst dimensional clarity is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim is that Hi-D maps 'retain the spatial efficiency of treemaps and dimensional clarity of sunburst visualizations' (Section 2) and help users analyze data 'quickly, effectively, and in an engaging manner' (Section 5). This requires that viewers can: (a) infer the dimension of an edge or cut from its orientation and hue, (b) infer the value from lightness, and (c) reliably compare leaf-polygon areas. None of these perceptual requirements are tested. The paper's limitations section concedes failure at about 30 dimensions and defers user studies and empirical comparisons to future work. The algorithm itself is internally coherent — for a convex polygon, binary-search area splits along a fixed direction can realize arbitrary proportions — so the weakness is not a mathematical error but an unvalidated perceptual and usability claim. Because the novelty resides primarily in the visual encoding, the absence of experimental evidence leaves the central contribution as a plausible design hypothesis rather than a demonstrated technique. The lack of released code or artifacts also prevents independent replication of the figures and algorithm behavior.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents Hi-D maps, a visualization technique for multi-dimensional categorical data. The data space is represented as a regular polygon; dimensions are assigned to polygon sides in a user-controlled order, and the polygon is recursively cut by lines parallel to those sides, with each leaf polygon's area proportional to the number of data records in the corresponding categorical intersection. The paper describes visual encodings (orientation, hue, lightness, saturation, countable 'marble' glyphs, and text), interaction mechanisms (dimension reordering, overview-detail, hierarchical browsing, dimension hiding, and animation), an algorithm with a stated time complexity, and a qualitative comparison with treemaps, sunburst visualizations, and parallel sets. The central claim is that Hi-D maps combine treemap-like spatial efficiency with sunburst-like dimensional clarity and help users analyze data quickly, effectively, and in an engaging manner, while acknowledging that the technique degrades around 30 dimensions.","tokens_in":6843,"tokens_out":7650,"duration_ms":89819,"significance":"The proposed layout is internally coherent and has design appeal. The recursive splitting of a convex polygon by lines parallel to chosen orientations can realize arbitrary leaf-area proportions, so the underlying area-proportionality claim is structurally sound. If the perceptual claims about orientation, hue, lightness, and marble counting are validated, the technique would be a useful addition to the multivariate categorical visualization toolbox, particularly because it keeps all dimensions within a single shared two-dimensional region. The paper is honest about limitations and clearly separates implemented features from planned evaluation. Its main weakness is that the central effectiveness claim is unmeasured: there is no user study, no perceptual experiment, and no empirical comparison with the named alternatives. The paper explicitly defers these to future work, so at present the core contribution is a plausible design hypothesis rather than a demonstrated technique. The absence of released code or supplemental material also limits independent replication of the figures and algorithm behavior.","major_comments":[{"comment":"The central claim that Hi-D maps 'help' users analyze data and 'retain the dimensional clarity of sunburst visualizations' is not supported by any evaluation. Section 5 explicitly lists user studies and empirical comparisons as future work, and the abstract's closing statement that 'Hi-D maps may add clarity before those limits are reached' is unfalsifiable as stated. The load-bearing assumptions are that viewers can (a) infer the mapped dimension from cut orientation and hue, (b) read value ordering from lightness, and (c) reliably compare leaf-polygon areas using area, saturation, and marble counts. None of these is tested. I recommend adding a controlled experiment, at minimum measuring dimension-identification accuracy, area-comparison error, and task completion time against treemaps, sunburst, and parallel sets. Alternatively, the comparative and effectiveness claims should be systematically toned down to design hypotheses.","section":"Abstract and Section 5 (Conclusions and Future Work)"},{"comment":"The split step assumes every sub-partition has positive size. The paper does not state how zero-count intersections are handled. In real categorical data, many high-dimensional intersections are empty; a zero-count partition would require a zero-area child polygon (a degenerate segment), and the recursive splitting of that degenerate polygon in subsequent dimensions is undefined. The paper should either specify a minimum-area or epsilon-based treatment for empty partitions, or explicitly restrict the method to data with no empty intersections and discuss the consequences for practical applicability.","section":"Section 4, Algorithm 1"},{"comment":"The claimed advantages over treemaps and sunburst visualizations are asserted qualitatively without operationalized measures. For example, the statement that Hi-D maps 'do not suffer as much from dimensional ambiguity' because they use angle and hue is a design rationale, not an empirical result; the same applies to the claim that cross-dimensional intersections are easier to find than in parallel sets. These comparative statements are central to the paper's motivation. If the evaluation requested in the first major comment is added, it should test these comparisons directly; if not, the comparative language should be revised to 'may' throughout.","section":"Section 3.2 (Comparison)"}],"minor_comments":[{"comment":"The complexity expression 'O(n(m + p) +pnnlgl)' is typographically garbled in the submission. It should be rendered as something like O(n(m+p) + p^n * n * log l), or the formula should be replaced with a clearly typeset derivation.","section":"Section 4"},{"comment":"The hue assignment is under-specified: the text states hue varies linearly from 0.0 to 0.4 and from 0.6 to 0.9, but does not state how hue values are assigned to individual dimensions, especially when the number of dimensions is odd or even, or how the unused last side of an even-dimensional polygon is treated in the color mapping.","section":"Section 3 (Design)"},{"comment":"The paper uses lexicographic value ordering for all dimensions, which imposes an arbitrary ranking on nominal categorical values. The authors should acknowledge that this ordering may bias interpretation or state that users can redefine value order.","section":"Section 3 (Design)"},{"comment":"The handling of the 'even number of dimensions' case (keeping the number of sides odd and leaving one side unused) is described in prose but not reflected in Algorithm 1 or in a precise initialization rule. Please make the mapping from dimensions to polygon sides explicit in the pseudocode or in an accompanying parameter table.","section":"Section 4, Algorithm 1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is well-written and honest about its limitations, but it reads as an early-stage technical report rather than a completed journal contribution. The lack of any evaluation of the central perceptual and usability claims is the main obstacle. If the venue accepts pure design papers without user studies, a minor revision with substantially softened claims might be appropriate; under the current stated claims, I recommend major revision. The authors should also be encouraged to release code and any synthetic datasets used in figures, as the current paper does not allow independent reproduction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper is a plausible design contribution with a clear algorithm and an honest limitations section, but its central claim—that Hi-D maps add clarity before perceptual limits are reached—is asserted, not demonstrated. If you read it for the algorithm, it's a reasonable layout scheme; if you read it for evidence, there is none.\n\nWhat's actually new: the specific combination of hierarchically cutting a regular polygon with lines parallel to an ordered sequence of sides, with leaf area proportional to intersection size, plus dimension reordering and interactive browsing. I don't see that exact recipe in the treemap/sunburst/parallel sets literature they cite. The algorithm is simple and credible: binary-search area splits realize arbitrary proportions, and the recursive structure is clearly described. The authors also add multiple visual cues (hue, lightness, saturation, marbles, text) and interactive features like overview-detail, dimension hiding, and hierarchical browsing.\n\nThe soft spot is the effectiveness claim. The paper states it retains treemap-like space efficiency and sunburst-like dimensional clarity, and that it enables quick and effective analysis—but no user study, no perceptual experiment, and no empirical comparison support any of that. The limitations section concedes failure around 30 dimensions and defers user studies to future work. That's honest, but it means the core value is a hypothesis. I also note the marble-counting cue is a heuristic that may not help viewers compare areas, and the algorithm's exponential dependence on values per dimension limits it to small categorical domains, though they acknowledge that. No code or data are released, so independent replication of the figures isn't possible.\n\nThat said, the paper is internally coherent, the writing is clear, and the authors don't overclaim. The binary-search splitting logic is sound; there's no mathematical error I can find. The citation list looks appropriate, and the self-citation to their agent-based animation work is fine.\n\nWho's this for? A reader working on categorical visualization or space-filling layouts could get ideas from the design. It's not yet a demonstrated technique, but it's a concrete, testable design proposal. I'd send it to peer review—a referee could push for a user study or at least a controlled comparison, and the design is specific enough to be worth that effort. I wouldn't cite it as evidence for effectiveness until evaluation exists, but I'd cite it as prior art in related work.\n\nRecommendation: engage with it, but treat the effectiveness claims as open questions.\n\nBest,","headline":"A concrete, testable design for categorical visualization whose effectiveness claims are all unverified—worth refereeing, but not worth believing yet.","tokens_in":7216,"tokens_out":4263,"would_cite":true,"duration_ms":37615,"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":"This paper introduces Hi-D maps, a space-efficient interactive visualization that maps multi-dimensional categorical data onto a regular polygon by hierarchical cuts parallel to its sides.","keywords":["Hi-D maps","multi-dimensional categorical data","space-filling visualization","regular polygon partition","hierarchical visualization","interactive visualization","treemaps","sunburst"],"falsifier":"A controlled experiment in which naive viewers are shown Hi-D maps of synthetic data and asked to name the dimension each cut represents and estimate relative intersection sizes; if accuracy falls steeply as dimensions grow or as polygon angles become similar, the central clarity claim fails.","tokens_in":6315,"feed_emoji":"🗺️","tokens_out":4405,"duration_ms":45122,"temperature":0.7,"pith_summary":"Dimensional categorical data — demographic records, survey answers, product attributes — is hard to see when each record combines several categories. The paper proposes Hi-D maps, a visualization that packs all such data into one regular polygon, cutting it hierarchically with lines parallel to each dimension's side so that every leaf polygon's area equals the fraction of records in that combination of categories. The motivation is practical: existing space-filling methods such as treemaps lose track of which dimension a cut belongs to, while radial methods such as sunbursts waste space by giving each dimension its own ring. Hi-D maps aim to keep the space efficiency of the former and the dimensional clarity of the latter in a single view, with interaction for re-ordering, zooming, and hiding dimensions.","feed_headline":"Hi-D maps fold many dimensions into one polygon","feed_subtitle":"Each polygon's area shows the size of one categorical intersection, keeping treemaps' efficiency and sunbursts' clarity.","key_machinery":"The central object is a regular polygon with a user-controlled, ordered sequence of sides, one per dimension. The construction cuts the polygon hierarchically with lines parallel to the chosen side, using binary search to place each cut so the child polygon areas match the sizes of the data partitions; recursion continues until all dimensions are used, and the leaf polygons are the categorical intersections. When the number of dimensions is even, the algorithm keeps the polygon's side count odd and leaves one side unmapped, avoiding parallel sides that would blur the angular encoding. This single recursive splitting rule is what carries both the space efficiency and the dimensional clarity.","core_discovery":"Hi-D maps is a space-partitioning visualization that maps the full data space of multi-dimensional categorical data onto a single regular 2D polygon of constant circumradius. Each dimension is assigned to one oriented side of the polygon, ordered clockwise, and the polygon is cut recursively with lines parallel to the current side, so the area of each resulting sub-polygon is proportional to the number of data points in that categorical intersection. The display uses orientation, hue, lightness, line thickness, and countable 'marble' glyphs as reinforcing cues, and supports interactive re-ordering, hierarchical browsing, and dimension-hiding. The paper's claim is that this retains the spatial efficiency of treemaps while providing the dimensional clarity of sunburst visualizations: all dimensions share the same space, but their identity is encoded by angle and color rather than by level counting or separate rings.","pith_inferences":["A natural next step the paper leaves implicit is a formal comparison of angle-plus-hue dimension identification against level-counting in treemaps; if angle perception degrades more slowly than level counting, the method's clarity claim gains quantitative support.","The use of an odd number of polygon sides whenever dimensions are even suggests a general design rule: reserve one side to avoid parallel ambiguity, a trick that could be exported to other angular space-filling layouts.","Because the paper reports failure around 30 dimensions from circularity and crowding, one testable extension is a hybrid approach that adaptively switches to a different polygon or layout when angular separation falls below a perceptual threshold."],"forward_implications":["If Hi-D maps work as claimed, a viewer can see the relative size of every combination of categories in a single bounded 2D view, without following links or traversing rings.","The same display supports interactive dimension re-ordering, hierarchical filtering by clicking edges, and hiding uninteresting dimensions, so the space partition can be explored rather than just inspected.","Because the polygon is cut recursively, the technique extends naturally to hierarchical data, with innermost polygons as leaves.","The algorithm's cost is dominated by $p^n$ for $n$ dimensions with up to $p$ values each, so the method targets categorical data with few values per dimension, not numerical data."],"supporting_citations":[{"why":"Treemaps, the space-filling hierarchy that Hi-D maps aim to match in spatial efficiency.","marker":"[17]"},{"why":"Sunburst visualizations, whose dimensional clarity Hi-D maps aim to retain.","marker":"[23]"},{"why":"Parallel sets, the categorical-data technique used as a comparison baseline.","marker":"[4]"},{"why":"Survey of multivariate visualization used to identify the gap Hi-D maps addresses.","marker":"[6]"},{"why":"Survey of high-dimensional visualization techniques framing the design space.","marker":"[14]"},{"why":"Survey of visual database exploration techniques that informs the related-work discussion.","marker":"[19]"}],"fun_headline_variants":["Hi-D maps pack many dimensions into one polygon","One polygon, many dimensions: Hi-D maps","Interactive polygon maps for multi-dimensional categorical data","Hi-D maps: all dimensions in one polygon"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes that viewers can read which dimension a cut represents from its angle and hue, and judge intersection sizes from area, saturation, and marble counts, without training or perceptual bias; no user study in the paper yet tests that assumption.","fun_headline_variants_meta":{"raw":{"variants":["Hi-D maps pack many dimensions into one polygon","One polygon, many dimensions: Hi-D maps","Interactive polygon maps for multi-dimensional categorical data","Hi-D maps: all dimensions in one polygon"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000749,"raw_usage":{"total_tokens":3305,"prompt_tokens":883,"completion_tokens":2422,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":2364}},"tokens_in":499,"tokens_out":2422,"duration_ms":19824,"temperature":1.0,"reasoning_tokens":2364,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:28:16.940842+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A controlled experiment in which naive viewers are shown Hi-D maps of synthetic data and asked to name the dimension each cut represents and estimate relative intersection sizes; if accuracy falls steeply as dimensions grow or as polygon angles become similar, the central clarity claim fails.","supporting_citations":[{"cited_title":"Johnson and B","cited_arxiv_id":null,"evidence_quote":"Treemaps, the space-filling hierarchy that Hi-D maps aim to match in spatial efficiency."},{"cited_title":"Stasko and E","cited_arxiv_id":null,"evidence_quote":"Sunburst visualizations, whose dimensional clarity Hi-D maps aim to retain."},{"cited_title":"Bendix, R","cited_arxiv_id":null,"evidence_quote":"Parallel sets, the categorical-data technique used as a comparison baseline."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Survey of multivariate visualization used to identify the gap Hi-D maps addresses."},{"cited_title":"Grinstein, M","cited_arxiv_id":null,"evidence_quote":"Survey of high-dimensional visualization techniques framing the design space."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Survey of visual database exploration techniques that informs the related-work discussion."}],"review_version":1}