Synthetic W-2 Wage and Tax Statement Data
Synthetic training data — no real PII, fully coherent identities
Generate synthetic 2025 vertical-stack W-2 sheets — four stacked copies on a two-page template whose second page carries no fields at all. Current-year wage base, plus the field-free trailing page that breaks naive page-classification and page-grouping logic.
144
Fields per document
2
Pages
Tax
Category
What this document is
This is the 2025 vertical-stack W-2: four full-width Wage and Tax Statements for the same employee printed one above the other, shipped as a two-page template. All the fields live on the first page; the second page carries the recipient instructions that accompany the statement and no data at all. That combination — a dense data page followed by a field-free page in the same document — is extremely common in real payroll output and almost absent from curated training corpora.
Why generate synthetically
Page classification is where multi-page document pipelines quietly fail. A page with no extractable fields is not an error, not a blank scan, and not a separator; it is a legitimate part of the document that must be recognised, kept with its sibling, and skipped for extraction. Corpora assembled by cropping statements out of payroll output throw those pages away, so models never learn them and then encounter them constantly in production. Generating the document whole, instructions page included, is the only way to supervise that behaviour with ground truth.
What makes synthetic data useful
Each document is one simulated employee's 2025 statement replicated in four full-width bands on the first page, with Box 3 Social Security wages capped at the 2025 wage base of $176,100 and Box 5 Medicare wages continuing past it for high earners. The second page is emitted as part of the same document with a page annotation and no field annotations, so the ground truth states explicitly that the correct number of extracted fields from that page is zero. Because the population is held constant with the 2023 and 2024 stacks, the three revisions form a controlled series in which only the wage base and the page count move.
Training challenges
Two problems compound here. The first is the stack itself: four copies at identical horizontal offsets, so copy assignment depends entirely on vertical banding, and identical values across copies mean a wrong assignment scores as correct under value-only evaluation. The second is the field-free second page. Document classifiers routinely label it as a different document, page-grouping logic routinely splits the pair, and extraction pipelines routinely report a confidence collapse on a page where zero fields is the right answer. A model that handles the stack but splits the document has still failed the submission.
Generate synthetic W-2 Wage and Tax Statement data
Start with 500 free credits. No credit card required.
Generate NowWho uses this data
Income-verification and lending platforms that receive multi-page payroll output as a single upload, document-management and IDP vendors whose page-classification and page-grouping stages need supervised negative examples, and payroll bureaus validating that their own two-page print output survives downstream ingestion.
Document complexity profile
144 fields across a two-page template, all of them on the first page: 72 currency amounts, 52 text, 12 checkbox targets, and 4 Social Security number and 4 employer identification number typed fields, arranged as one 36-field statement stacked four times. 108 annotation relations, 12 conditional bindings, 16 function calls, maximum expression depth 2. The second page is field-free by design and annotated as such — one of the few documents in the catalog where a page's correct extraction result is the empty set.
Key stats from our synthetic corpus
Quantitative characteristics of the W-2 Wage and Tax Statement documents our generator produces.
| Metric | Value | Detail |
|---|---|---|
| 2025 Social Security wage base | $176,100 | Box 3 Social Security wages cap at $176,100 while Box 5 Medicare wages continue to $316,900. The top rung of the three-year stacked ladder, and the current-year value a 2024-trained validator will reject. |
| Pages carrying fields | 1 of 2 | All fields sit on the first page; the second carries recipient instructions and no data. Ground truth for that page is an explicit empty set, which lets you score page classification and page grouping rather than assuming them. |
| Copies stacked per page | 4 | Four full-width statements at identical horizontal offsets, so copy assignment depends on vertical banding alone. Identical values across copies mean a mis-assignment is invisible unless the evaluation checks copy identity. |
| Blank field rate | 42% | 60 of the annotated fields render blank on a typical sheet, repeated across the four bands, on top of an entire page that is blank by design. Between the two, this is the catalog's densest concentration of correctly-empty regions. |
| Median Box 1 wages | $45,600 | Across 641 eligible synthetic identities the median Box 1 wage is $45,600, p25–p75 $27,200 to $74,400, spanning $4,000 to $316,900 — identical to the 2023 and 2024 stacks so the revision comparison stays controlled. |
How this document co-occurs with others
Rates at which identities in our corpus that produce a W-2 Wage and Tax Statement also produce other documents.
| Correlation | Rate | Detail |
|---|---|---|
| Prior revision, single-page template | 100% | The 2024 stack is the same layout on a one-page template with a lower wage base. The pair separates the page-structure change from the year change. |
| Same year, eleven-page packet | 100% | The official IRS packet for the same year carries its fields on one page out of eleven. Together the two documents bracket the multi-page problem: one field-free page here, ten there. |
| Same year, half-page single copy | 100% | The same 2025 wage data rendered as one half-page statement with no replication and no extra page. The simplest member of the family, and the natural baseline for the stacked and packet variants. |
| Employee's personal return | 100% | The wages on the statement flow to the employee's Form 1040. The W-2-plus-1040 pair is the most common multi-document bundle in consumer tax processing. |
Wage and rate figures above are corpus-derived: they were computed over the 641 employment-eligible identities inside a local synthetic corpus of 1,000 identities generated by SymageDocs' World Simulation Engine at seed 20260421. Field, page, type and relation counts come from the shipped 2025 vertical-stack W-2 definition in the SymageDocs form library. No real employee, employer, or payroll data was used at any stage.
Frequently asked questions
- What data format do synthetic 2025 vertical-stack documents include?
- Each generated identity produces a rendered PDF plus a structured JSON annotation file with bounding boxes, field types, and ground-truth values for all 144 fields — 36 fields per copy across four stacked copies, all on the first page — plus 108 label-to-value relations. The second page is part of the same document and carries no field annotations. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.
- Why does the template have a page with no fields?
- Because the real statement ships with recipient instructions on the reverse. Keeping that page in the document is deliberate: it supervises the case where the correct extraction result for a page is the empty set, which is a case most pipelines have never been trained on and most evaluations do not score.
- How does this differ from the 2023 and 2024 stacks?
- The Social Security wage base rises to $176,100, and the template gains the field-free second page that the earlier two revisions do not have. Holding the population fixed across all three means the trio isolates two variables cleanly — the year-indexed cap and the page structure — over an otherwise identical document.
- Is this the current W-2 revision?
- It is the current stacked revision at the time of writing. The 2026 revision splits Box 14 into 14a 'Other' and 14b for Treasury Tipped Occupation Codes; the catalog ships that change on the single-page and packet layouts. Any pipeline running in 2026 sees both Box 14 structures in the same inbox, so train and evaluate on both.
- Can I use this data commercially?
- Yes. Every Social Security number, name, address, employer and wage figure is synthetic, contains no real personal data, and is licensed for commercial use including model training, benchmarking, and redistribution inside your own products.