Writing study: Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Hao Tan, jun lan, Zichang Tan, Senyuan Shi, Ajian Liu, Chuanbiao Song, Huijia Zhu, Weiqiang Wang, Jun Wan, Zhen Lei · 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) · Datasets, benchmarks & measurement resources.

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

58 physical pages

Before references / appendix

≈ 10.21 page-equivalents

Main figures / tables

6 / 8

Attached appendix figures / tables

34 / 11

Attached appendix span

≈ 40.89 page-equivalents

Before references / appendix: ≈ 17.6%References / intervening material: ≈ 11.9%Attached appendix: ≈ 70.5%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	≈ 1.64 pp.	2.83%
2 RELATED WORK	3	≈ 0.59 pp.	1.02%
3 HYDRAFAKE DATASET	3	≈ 1.53 pp.	2.64%
4 METHOD	5	≈ 2.19 pp.	3.78%
5 EXPERIMENTS	7	≈ 3.23 pp.	5.57%
6 CONCLUSION	10	≈ 0.29 pp.	0.5%
ACKNOWLEDGMENTS	11	≈ 0.1 pp.	0.17%
REFERENCES	11	≈ 6.9 pp.	11.9%
A APPENDIX	18	≈ 13.72 pp.	23.66%
ETHICS STATEMENT	31	≈ 27.17 pp.	46.84%

All detected figure and table locations

Main text

• Figure 1 · p. 2
• Figure 2 · p. 4
• Figure 3 · p. 6
• Table 1 · p. 8
• Table 2 · p. 8
• Table 3 · p. 8
• Table 4 · p. 8
• Table 5 · p. 9
• Table 6 · p. 9
• Figure 4 · p. 9
• Figure 5 · p. 9
• Table 7 · p. 10
• Table 8 · p. 10
• Figure 6 · p. 10

Attached appendix

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

Argument and organization

1. Practical evaluation before model design

   A heterogeneous-training/OOD-testing mismatch motivates HydraFake's four evaluation splits before introducing the detector.

2. Cold start, preference alignment, online exploration

   The training pipeline maps directly onto component ablations and qualitative reasoning comparisons.

Problem, claim and contribution

The introduction identifies a protocol mismatch, not merely an accuracy deficit: single-source training underrepresents realistic mixed-source deployment.

Sections 1 and 3, pages 1–5; Figure 2 · Main paper

Argument and section order

Reward, backbone, reasoning-pattern and preference-negative ablations each answer a different design question, avoiding a single undifferentiated component-removal table.

Tables 2–7, pages 8–10 · Main paper

Figures, captions and visual explanation

Figure 3 uses two horizontal training stages, three dataset sizes, sample reasoning tags and a separate reward inset. Arrows convey data and checkpoint transitions rather than decorative sequencing.

Figure 3, page 6 · Main paper

Tables, comparisons and evidence

Table 1 separates small vision models, generic MLLMs and specialized detectors; its caption explains the restricted-training mini variant and omission of ID scores for one comparison block.

Table 1, page 8 · Main paper

Appendix and reproducibility

Data provenance, per-subset precision/recall, batch-level latency, complete prompt templates and failure examples are provided; these reveal low-resolution weaknesses obscured by aggregate accuracy.

A.1–A.6, Tables 10–19, Figure 22, pages 19–30 and 44; prompts pages 50–58 · Main paper

Clarity and writing conventions

Long colored reasoning traces make qualitative contrasts concrete but can overwhelm the visual evidence. Generated explanations include debatable claims, so fluency should not be treated as verified forensic reasoning.

Figures 16–22, pages 38–44 · Main paper

What to reuse—and what not to infer

Transferable lesson: Build the evaluation taxonomy around the claimed deployment gap, then align each training stage with a distinct ablation and preserve the exact averaging protocol in table captions.

Boundary: The qualitative traces and model-judged reasoning scores do not establish causal faithfulness. Percentage-point gains are sometimes written as percentages; use the tabulated values and do not equate Table 1's subset average with the four-split average elsewhere.

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 1–6; acknowledgements and references pages 11–17.

Appendix / supplement

Pages 18–58, A.1–A.12: data construction, implementations, reasoning formulation, annotation, full results, prompts, qualitative comparisons, failures, ethics and limitations.

Visually inspected PDF pages

6, 8

• ICLR 2026 final proceedings · 58 pages