Synthetic W-2 Wage and Tax Statement Data
Synthetic training data — no real PII, fully coherent identities
Generate synthetic reissued 2019 W-2 statements in the borderless quarter-sheet layout — the replacement copy an employer prints when an employee loses the original. The oldest revision in the catalog and the only W-2 rendered without printed box outlines.
148
Fields per document
1
Page
Tax
Category
What this document is
This is a reissued 2019 W-2: the replacement statement an employer produces when an employee needs a copy of a prior-year wage statement they no longer have. It uses the quarter-sheet four-copy layout, but rendered without the printed box outlines that frame every field on a standard W-2 — the boxes are defined by label position and whitespace alone. Reissued statements circulate constantly in mortgage underwriting, immigration filings, benefits appeals and back-tax work, where the document needed is several years old.
Why generate synthetically
Every income-verification pipeline claims to handle prior-year documents and almost none of them are evaluated on any. Training corpora are assembled from whatever was recent when the corpus was built, so models systematically overfit to one or two revisions and degrade on anything older — which is precisely the population a reissued statement comes from. A 2019 revision generated today gives you a genuine out-of-distribution test with full ground truth, including the era-specific Social Security wage base that a model trained on current documents will normalise incorrectly.
What makes synthetic data useful
Each statement carries a simulated employee's 2019 wage record with the wage base and rate structure of that year: Box 3 Social Security wages cap at $132,900, well below every subsequent year, while Box 5 Medicare wages continue past it for high earners. The four copies on the sheet carry identical values with distinct coordinates, as on any perforated sheet. The absence of printed box rules is a rendering property of the definition, not an artefact — it reproduces the borderless output some payroll print streams actually produce, where the only cue distinguishing Box 4 from Box 6 is the printed label above it.
Training challenges
Removing the box outlines removes the feature most layout models lean on hardest. Detectors that segment on printed rules have nothing to segment; models that rely on the visual grid to associate a label with the value beneath it have to fall back to pure spatial proximity, in a quarter-page layout where the next label is a few millimetres away. On top of that, the four-copy replication means four instances of every label with identical values, so a cross-copy merge is invisible to value-level scoring. And the 2019 wage base sits far enough below current caps that a model which learned the cap as a constant will flag valid documents as inconsistent.
Generate synthetic W-2 Wage and Tax Statement data
Start with 500 free credits. No credit card required.
Generate NowWho uses this data
Mortgage and consumer-lending platforms handling prior-year income documentation, immigration and benefits-appeal document processing where multi-year wage histories are required, back-tax and amended-return preparers, and any income-verification vendor that needs a defensible out-of-distribution benchmark rather than a same-revision holdout.
Document complexity profile
148 fields on a single page: 72 currency amounts, 56 text, 12 checkbox targets, and 4 Social Security number and 4 employer identification number typed fields, arranged as one 37-field statement replicated four times. 125 annotation relations, 12 conditional bindings and 16 function calls at a maximum expression depth of 2 — a slightly lighter binding graph than the current four-up sheets, reflecting the era's simpler Box 12 handling. Rendered without printed box outlines.
Key stats from our synthetic corpus
Quantitative characteristics of the W-2 Wage and Tax Statement documents our generator produces.
| Metric | Value | Detail |
|---|---|---|
| 2019 Social Security wage base | $132,900 | Box 3 Social Security wages cap at $132,900 — the lowest cap of any revision in the catalog — while Box 5 Medicare wages run to $316,900. The gap between Box 3 and Box 5 on a high earner is far wider here than on any current-year statement, which is exactly what a cap-hardcoding validator gets wrong. |
| Printed box outlines | None | The borderless rendering removes the printed grid that layout models use to associate labels with values. Field association has to come from spatial proximity and label semantics alone, in a layout where four copies place the nearest competing label a few millimetres away. |
| 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. The wage distribution is held constant across revisions so that differences you measure between years come from the document, not from the population. |
| Blank field rate | 43% | 64 of the annotated fields render blank on a typical sheet, repeated across the four copies. With no printed outlines, a blank field is genuinely invisible in the rendered image — the annotation is the only evidence it exists, which is what makes this corpus useful for blank-field recall. |
| Statements per sheet | 4 | Four complete statements with identical values and distinct coordinates. Copy-level annotation makes the deduplication error measurable rather than silent. |
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 |
|---|---|---|
| Modern sheet, same four-up geometry | 100% | The 2024 four-up sheet uses the same quadrant geometry with printed box outlines and a current wage base. Holding geometry fixed and varying only the outlines and the year isolates which of the two your model is actually depending on. |
| Other historical-revision document | 100% | The 1988 Form 1040 is the catalog's other deliberately old revision. Together they form a small historical-document benchmark for pipelines that claim multi-decade coverage — a claim almost never tested. |
| Current single-copy layout | 100% | The same wage semantics rendered as a modern half-page single statement with printed outlines. The contrast between this and the reissued sheet is the widest layout gap available inside one form family. |
| Issuing employer's quarterly returns | 100% | The employer that issued the reissued statement also files Form 941. Reconciling a prior-year employee statement against employer returns is the exact task a back-tax or amended-return preparer performs. |
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, type and relation counts, and the borderless rendering, come from the shipped 2019 reissued 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 reissued W-2 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 148 fields on the sheet — 37 fields per copy across four copies — plus 125 label-to-value relations. Annotations are present for every field whether or not a printed outline surrounds it. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.
- Why does this layout have no box outlines?
- Because the shipped definition reproduces the borderless print stream some payroll systems produce for replacement copies. It is the hardest variant in the W-2 family for layout-driven extraction, and it is the one most likely to arrive as a fax or a photocopy of a photocopy. If your model's field association depends on printed rules, this is the document that will show you.
- Is a six-year-old revision actually useful for training?
- It is most useful for evaluation. Holding a prior-year revision out of training and scoring against it is the only honest measurement of whether a W-2 model generalises across revisions or memorised the current one. Prior-year documents are a large share of real mortgage and immigration submissions, so the score you get on this document is closer to production accuracy than the score you get on the current year.
- What was the 2019 Social Security wage base?
- $132,900. Box 3 caps there while Box 5 Medicare wages continue uncapped, so a 2019 statement for a high earner shows a much wider Box 3 to Box 5 gap than a current-year one. Validation logic that hardcodes today's cap will reject these documents outright.
- 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.