Writing study: TabStruct: Measuring Structural Fidelity of Tabular Data

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Xiangjian Jiang, Nikola Simidjievski, Mateja Jamnik · Official ICLR 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Datasets, benchmarks & measurement resources (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

63 physical pages

Before references / appendix

≈ 10.86 page-equivalents

Main figures / tables

4 / 2

Attached appendix figures / tables

6 / 64

Attached appendix span

≈ 44.87 page-equivalents

Before references / appendix: ≈ 17.24%References / intervening material: ≈ 11.54%Attached appendix: ≈ 71.22%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1 INTRODUCTION	1	≈ 2.18 pp.	3.46%
2 RELATED WORK	3	≈ 0.8 pp.	1.27%
3 METHODS	4	≈ 2.53 pp.	4.02%
4 EXPERIMENTS	7	≈ 3.6 pp.	5.71%
5 CONCLUSION	10	≈ 0.4 pp.	0.63%
ETHICS STATEMENT	11	≈ 0.4 pp.	0.63%
REPRODUCIBILITY STATEMENT	11	≈ 0.18 pp.	0.29%
ACKNOWLEDGEMENTS	11	≈ 0.18 pp.	0.29%
REFERENCES	11	≈ 7.27 pp.	11.54%
Appendix	19	≈ 0.97 pp.	1.54%
A SUMMARY OF RELATED WORK	20	≈ 5 pp.	7.94%
B DESIGNS OF STRUCTURAL FIDELITY METRICS	25	≈ 3 pp.	4.76%
C RATIONALES FOR EVALUATION FRAMEWORK DESIGN	28	≈ 2 pp.	3.17%
D REPRODUCIBILITY	30	≈ 8 pp.	12.7%
E EXTENDED ANALYSIS AND DISCUSSION	38	≈ 11 pp.	17.46%
F EXTENDED EXPERIMENTAL RESULTS	49	≈ 14.9 pp.	23.65%

All detected figure and table locations

Main text

• Table 1 · p. 1
• Figure 1 · p. 2
• Figure 2 · p. 3
• Table 2 · p. 8
• Figure 3 · p. 9
• Figure 4 · p. 10

Attached appendix

• Table 3 · p. 23
• Table 4 · p. 24
• Figure 1 · p. 28
• Table 5 · p. 30
• Table 6 · p. 30
• Table 7 · p. 30
• Table 8 · p. 31
• Table 9 · p. 31
• Figure 5 · p. 32
• Table 10 · p. 33
• Table 11 · p. 33
• Table 12 · p. 34
• Table 13 · p. 34
• Table 14 · p. 34
• Table 15 · p. 35
• Table 16 · p. 35
• Table 17 · p. 35
• Table 18 · p. 36
• Table 19 · p. 36
• Table 20 · p. 36
• Table 21 · p. 37
• Table 22 · p. 38
• Table 23 · p. 39
• Figure 6 · p. 39
• Table 24 · p. 40
• Table 25 · p. 40
• Table 26 · p. 40
• Table 27 · p. 41
• Table 28 · p. 41
• Table 29 · p. 41
• Table 30 · p. 42
• Table 31 · p. 42
• Figure 7 · p. 43
• Figure 8 · p. 44
• Table 32 · p. 45
• Table 33 · p. 45
• Table 34 · p. 46
• Table 35 · p. 46
• Table 36 · p. 46
• Table 37 · p. 47
• Figure 9 · p. 48
• Table 38 · p. 49
• Table 39 · p. 49
• Table 40 · p. 50
• Table 41 · p. 51
• Table 42 · p. 51
• Table 43 · p. 51
• Table 44 · p. 52
• Table 45 · p. 52
• Table 46 · p. 53
• Table 47 · p. 53
• Table 48 · p. 54
• Table 49 · p. 54
• Table 50 · p. 55
• Table 51 · p. 55
• Table 52 · p. 56
• Table 53 · p. 56
• Table 54 · p. 57
• Table 55 · p. 57
• Table 56 · p. 58
• Table 57 · p. 58
• Table 58 · p. 59
• Table 59 · p. 59
• Table 60 · p. 60
• Table 61 · p. 60
• Table 62 · p. 61
• Table 63 · p. 61
• Table 64 · p. 62
• Table 65 · p. 62
• Table 66 · p. 63

Argument and organization

1. Motivating counterexample and metric construction

   Distinguish target-specific predictive utility from structure-wide fidelity, then introduce conditional-independence and SCM-free utility measures.

2. Four research questions and matched appendix sections

   Test benchmark validity, metric validity, generator behavior, and practical computation separately.

Problem, claim and contribution

A gravity example connects familiar physical variables to a causal graph and conditional-independence statements before contrasting conventional scores with structural violations.

Figure 1, page 2. · Main paper

Argument and section order

The introduction separates conceptual, technical, and empirical contributions; Q1–Q4 then provide an explicit contract that is mirrored by Sections 4.1–4.4 and Appendices E.1–E.4.

Page 3 contribution list; Section 4, page 7; Appendix E, pages 38–47. · Main paper

Figures, captions and visual explanation

The numbered motivating panels and long caption explain the full chain from physical system to score disagreement. The later heatmap and utility scatterplot divide metric validation from generator comparison.

Figures 1 and 3, pages 2 and 9. · Main paper

Tables, comparisons and evidence

The main table groups metrics by evaluation dimension and splits known-SCM from real-world data, explicitly marking unavailable causal scores N/A. Top-three colors aid comparison, but the dense typography requires zooming.

Table 2, page 8. · Main paper

Appendix and reproducibility

A linked appendix contents page leads to dataset IDs, preprocessing, generator-specific search spaces, downstream-predictor budgets, software, and all 29 dataset tables rather than only aggregate rankings.

Page 19; Appendix D, pages 30–37; Appendix F, pages 49–63. · Main paper

Clarity and writing conventions

The strongest qualification appears beside the central result: the correlation between global utility and global CI is empirical, not a formal equivalence. The future-work section returns to that unresolved theoretical bridge.

Section 4.2, page 8; Section 4.4, page 10; Appendix E.6, page 48. · Main paper

What to reuse—and what not to infer

Transferable lesson: For a new evaluation metric, start with a concrete counterexample, separate validity from usefulness and cost, and make the appendix follow the same research-question structure as the main paper.

Boundary: Global utility is a predictor-dependent empirical proxy, not identified causal structure or a privacy guarantee. Avoid upgrading the paper’s correlation and permutation-order arguments into universal causal claims; its normalization also distinguishes aggregate table values from raw per-dataset results.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-19

Reading status

Full text + selected visuals inspected

Main text

Pages 1–10, Sections 1–5; ethics/reproducibility statements on page 11; references on pages 11–18.

Appendix / supplement

Pages 19–63, complete Appendices A–F: related-work comparison, metric algorithms, design rationale, implementation, extended analyses, and every per-dataset results table.

Visually inspected PDF pages

2, 8, 9, 19

• ICLR 2026 official proceedings PDF, retrieved 2026-09-19 · 63 pages