Synthetic MA Verification of Self-Employment Income (INVF) Data
Synthetic training data — no real PII, fully coherent identities
Generate synthetic self-employment income verification forms — the two-page declaration a benefits agency uses to verify gig and cash-economy earnings that no W-2 or 1099 covers. Training data for public-benefits eligibility automation.
35
Fields per document
2
Pages
Healthcare
Category
What this document is
This is the self-employment income verification a state health and human services agency sends when an applicant reports earnings that no employer will confirm. The household head lists each business or gig, how long in the year it runs, gross income, monthly expenses, average monthly income and net for the year, then totals across all of them and signs. Massachusetts calls its version the INVF; every state runs an equivalent, and it is the document that decides whether a household keeps its coverage.
Why generate synthetically
Cash and gig income is the hardest income to verify and the most common reason an eligibility determination stalls. The forms that carry it are self-reported, frequently handwritten, filled out by people under time pressure, and covered by benefits-confidentiality rules that put a real corpus out of reach for every vendor building against them. Meanwhile the volume is enormous and rising with the gig economy. Synthetic verification forms let eligibility automation be trained and measured on the exact document that gates the decision, without touching a real household's file.
What makes synthetic data useful
Each generated form is one simulated household's genuine self-employment picture. The trades are the ones that actually appear on these forms — handyman work, tutoring, pet sitting, market stalls, hair styling, rideshare driving, house cleaning, landscaping, food delivery, snow plowing, freelance design — rather than generic business names. Timeframes are realistic: 77.7% of gigs run year-round and the rest are seasonal windows that match the trade, so snow plowing is a winter entry and landscaping is a spring-to-fall one. Gross income, monthly expenses, average monthly income and the annual net are consistent with each other and roll up into the household total.
Training challenges
The form has four repeated business blocks and almost nobody fills more than two: the first is always populated, the second on 27.7%, and the third and fourth are empty on every document in the corpus. Two entirely blank labelled blocks per page is the strongest hallucination bait in the catalog. Everything is typed as free text, including the money and the date, so amounts must be recognised from context. And the two supplementary fields behave the way real supplementary fields do — the income-frequency box fills on 14.5% of forms and the household notes line on 15.3%, each with a single fixed value — which means a model sees them almost never and must still not invent them the other 85% of the time.
Generate synthetic MA Verification of Self-Employment Income (INVF) data
Start with 500 free credits. No credit card required.
Generate NowWho uses this data
Public-benefits eligibility platforms and state health-agency vendors automating Medicaid, SNAP and subsidised-coverage verification, income-verification services covering the gig and cash economy where payroll data does not reach, and case-management systems that must extract a household income total from a self-reported declaration.
Document complexity profile
Fields spread across two pages, both carrying content: a household header, four repeated six-field business blocks, a household total with two conditional supplementary fields, and a signature block — with one label-to-value relation per field. All but the form date and the signature are typed as free text, and two of the four business blocks are empty on every document, so the page's labelled structure is substantially larger than its populated content.
Key stats from our synthetic corpus
Quantitative characteristics of the MA Verification of Self-Employment Income (INVF) documents our generator produces.
| Metric | Value | Detail |
|---|---|---|
| Second business reported | 27.7% | 27.7% of households report a second gig; none report a third or fourth. Two fully labelled but permanently empty blocks make this the corpus's strongest test of whether a model will populate a section that has no content. |
| Year-round gigs | 77.7% | 77.7% of businesses run year-round; the rest are seasonal windows — May to September, April to October, June to August, December to March, or the school year — matched to the trade. Seasonality changes how an agency annualises the income, so the field is decision-relevant, not decorative. |
| Trade categories | 12 | Businesses are drawn from twelve real informal trades at roughly 8 to 10% each, including handyman services, tutoring, pet sitting, market stalls, hair styling, rideshare driving, house cleaning and snow plowing. A flat twelve-way categorical in a free-text field is a target models tend to collapse. |
| Income-frequency field populated | 14.5% | The income-frequency box fills on only 14.5% of forms, always with the same value, and the household notes line on 15.3%. Rarely-populated fixed-value fields are where precision is genuinely tested — the correct answer is nothing on roughly six forms in seven. |
| Signature present | 100% | Every form carries a head-of-household signature and signature date. Together with the forms in the catalog whose signature blocks are always empty, this gives a signature detector both classes from documents that are otherwise structurally similar. |
How this document co-occurs with others
Rates at which identities in our corpus that produce a MA Verification of Self-Employment Income (INVF) also produce other documents.
| Correlation | Rate | Detail |
|---|---|---|
| Self-employment income on the personal return | 100% | The same households file a Form 1040 reporting the same self-employment activity. Agencies cross-check a declared gig income against the return, so the pair is the realistic verification bundle. |
| Healthcare intake for the same household | 100% | The coverage this form protects is used at a practice that captures a registration sheet. Following one household across the benefits and clinical sides is the entity-resolution case state vendors actually face. |
| Revenue evidence for the reported gig | 100% | Freelance invoices are the supporting documentation an applicant attaches to substantiate the income declared here. Declaration plus evidence is a two-document verification task, not a single-form extraction. |
| Statements the household receives | 100% | Households in this corpus receive itemized medical bills for their care. Income verification and cost-sharing land in the same case file and are processed by the same intake queue. |
Prevalence rates and category distributions above are corpus-derived: they were computed over all 1,000 identities in a local synthetic corpus generated by SymageDocs' World Simulation Engine at seed 20260421 — every identity is eligible for this form, so no sub-sampling applies. Field, page, type and relation counts, the four-block structure and the empty third and fourth blocks come from the shipped self-employment income verification definition in the SymageDocs form library. No real household, benefits, or income data was used at any stage.
Frequently asked questions
- What data format do synthetic self-employment verification documents include?
- Each generated identity produces a rendered two-page PDF plus a structured JSON annotation file with bounding boxes, field types, and ground-truth values for every field on both pages, with one label-to-value relation per field. Fields are typed as text apart from the form date and the head-of-household signature. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.
- Which agency's form is this?
- The layout follows the Massachusetts health-benefits verification of self-employment income. The structure — repeated business blocks, gross and expense columns, an annual net, a household total and a signature — is common to the equivalent forms used by other state agencies, so a model trained here transfers to them far better than one trained on a generic income statement.
- How many businesses does a typical household report?
- One on every form, a second on 27.7%, and never a third or fourth in the current corpus. If your pipeline must handle three or four concurrent gigs, treat that as uncovered by this corpus rather than assuming the blocks are exercised.
- Is the form signed?
- Yes. Unlike several other forms in the catalog, the head-of-household signature and its date are populated on every generated document, so this is one of the corpus's reliable positive sources for signature detection and signature-date extraction.
- Can I use this data commercially?
- Yes. Every household, member identifier, business and income figure is synthetic, contains no real benefits-recipient data, and is licensed for commercial use including model training, benchmarking, and redistribution inside your own products.