Writing study: Exploratory Causal Inference in SAEnce

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Tommaso Mencattini, Riccardo Cadei, Francesco Locatello · 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: Learning & optimization algorithms (primary) · Theory & formal foundations.

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

48 physical pages

Before references / appendix

≈ 11.11 page-equivalents

Main figures / tables

6 / 0

Attached appendix figures / tables

31 / 2

Attached appendix span

≈ 32.9 page-equivalents

Before references / appendix: ≈ 23.15%References / intervening material: ≈ 8.31%Attached appendix: ≈ 68.54%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	≈ 1.42 pp.	2.96%
2 TREATMENT EFFECT ESTIMATION IN RANDOMIZED CONTROLLED TRIALS	3	≈ 1.29 pp.	2.69%
3 EXPLORATORY CAUSAL INFERENCE VIA NEURAL REPRESENTATIONS	4	≈ 2.11 pp.	4.4%
4 NEURAL EFFECT SEARCH	6	≈ 1.61 pp.	3.35%
5 RELATED WORKS	8	≈ 0.6 pp.	1.25%
6 EXPERIMENTS	8	≈ 1.93 pp.	4.02%
7 CONCLUSION	10	≈ 0.65 pp.	1.35%
ACKNOWLEDGMENTS	11	≈ 0.2 pp.	0.42%
ETHICS STATEMENT	11	≈ 0.15 pp.	0.31%
REPRODUCIBILITY STATEMENT	11	≈ 0.46 pp.	0.96%
REFERENCES	12	≈ 4 pp.	8.33%
Appendix	16	≈ 1 pp.	2.08%
A PROOFS	17	≈ 6 pp.	12.5%
B NEURAL EFFECT TEST	23	≈ 1.05 pp.	2.19%
C MINIMAL PYTHON IMPLEMENTATION SNIPPET	24	≈ 1.95 pp.	4.06%
D EXPERIMENTS DETAILS	26	≈ 2 pp.	4.17%
E ADDITIONAL EXPERIMENTS	28	≈ 20.89 pp.	43.52%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 3
• Figure 3 · p. 6
• Figure 4 · p. 7
• Figure 5 · p. 9
• Figure 6 · p. 10

Attached appendix

• Table 1 · p. 27
• Table 2 · p. 27
• Figure 7 · p. 29
• Figure 8 · p. 29
• Figure 9 · p. 30
• Figure 10 · p. 30
• Figure 11 · p. 31
• Figure 12 · p. 31
• Figure 13 · p. 32
• Figure 14 · p. 33
• Figure 15 · p. 33
• Figure 16 · p. 34
• Figure 17 · p. 34
• Figure 18 · p. 35
• Figure 19 · p. 35
• Figure 20 · p. 36
• Figure 21 · p. 37
• Figure 22 · p. 37
• Figure 23 · p. 38
• Figure 24 · p. 38
• Figure 25 · p. 39
• Figure 26 · p. 39
• Figure 27 · p. 40
• Figure 28 · p. 41
• Figure 29 · p. 42
• Figure 30 · p. 43
• Figure 31 · p. 44
• Figure 32 · p. 44
• Figure 33 · p. 45
• Figure 34 · p. 46
• Figure 35 · p. 46
• Figure 36 · p. 47
• Figure 37 · p. 48

Argument and organization

1. Motivation and failure

   A concrete randomized-trial problem motivates a formal entanglement paradox before proposing the algorithm.

2. Algorithm, guarantees, evidence

   Recursive stratification is followed by a proof sketch, controlled simulations, and an ecological trial.

Problem, claim and contribution

The introduction connects hypothesis-driven trials to exploratory discovery using a concrete ant-behaviour example rather than an abstract foundation-model promise.

Pages 1–2, introduction and motivating example. · Main paper

Argument and section order

Named failure theorems precede Neural Effect Search, making the recursive method a response to a demonstrated inferential problem.

Pages 4–7, sections 3–4. · Main paper

Figures, captions and visual explanation

The first-page four-stage diagram separates measurement, selection, and interpretation; compact ant thumbnails connect the statistical pipeline to observable phenomena.

Page 1, Figure 1. · Main paper

Tables, comparisons and evidence

Aligned precision/recall/IoU panels reuse method colours and a shared legend while independently varying sample size and effect magnitude; the caption names the intended trade-off.

Page 9, Figure 5. · Main paper

Appendix and reproducibility

A dedicated contents page routes readers to proofs, algorithms, implementation, settings, and extensive sensitivity tests rather than one undifferentiated supplement.

Page 16 and appendices A–E. · Main paper

Clarity and writing conventions

Limitations distinguish useful exploratory signals from biological conclusions and acknowledge representation/identifiability assumptions.

Pages 10–11, section 7; appendix E.1. · Main paper

What to reuse—and what not to infer

Transferable lesson: Build a method narrative around an explicit failure mechanism, then reuse the same visual comparison grammar to test whether the proposed remedy actually addresses it.

Boundary: Keep assumptions and exploratory status visible: the small ant trial omits Bonferroni adjustment and also identifies a background-position artefact. Audit appendix consistency and executable snippets; the minimal Python example passes list.append's return value into recursion.

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 1–11, sections 1–7 and ethics/reproducibility statements; references pages 12–15.

Appendix / supplement

Pages 16–48: contents, A proofs, B testing algorithm, C minimal code, D settings, E assumptions/ablations/additional baselines.

Visually inspected PDF pages

1, 9

• ICLR 2026 final proceedings · 48 pages