Synthetic WH-381 Notice of Eligibility and Rights & Responsibilities (FMLA) Data

What this document is

Form WH-381 is the notice an employer must give an employee within five business days of a leave request under the Family and Medical Leave Act. Unlike the WH-380 certifications, which are filled out by a health care provider, the WH-381 is written by the employer: it states whether the employee is eligible, why they are or are not, what documentation is required and by when, whether accrued paid leave will run concurrently, whether the employee is a key employee, and how often they must report in while on leave.

Why generate synthetically

The WH-381 is where FMLA compliance failures actually happen. A missing certification deadline, an unchecked eligibility box, a paid-leave substitution rule that contradicts the employer's own policy — these are the findings that turn up in Department of Labor investigations, and they are findings a document model can be trained to catch. But the notices are internal employer records, and a corpus of them is a corpus of one company's leave decisions. Synthetic notices let a leave-management vendor train completeness and consistency checkers across many employer policy shapes rather than one.

What makes synthetic data useful

Each notice is generated as one side of a coherent leave file. The leave reason checkboxes agree with which FMLA certification the same employee is generating — an own-health-condition notice pairs with a WH-380-E and a family-health notice with a WH-380-F, and where the reason is family health the specific relationship checkbox names the family member who actually exists in that identity's household. Months of service and hours worked are computed from the simulated employment record rather than invented, so the eligibility determination on the form follows from numbers on the form. The certification return date, the paid-leave conditions and the reporting frequency are drawn together as a coherent employer policy.

Training challenges

The document is a decision record expressed almost entirely as checkbox pairs — eligible or not, key employee yes or no, paid leave will or will not be substituted, reporting required or not required — and the two members of each pair sit adjacent with near-identical labels. Getting the pair wrong inverts the meaning of the notice, which is a far more damaging error than dropping a field. Several free-text regions are conditional on the checkbox above them: the reporting-frequency text only exists if reporting is required, the paid-leave conditions only if substitution applies, so a model must learn presence as well as value. Contact-person blocks repeat three times across the four pages with different purposes (certification, benefits, insurance premiums) and identical structure, and the employee name repeats as a per-page header.

Who uses this data

Leave-management platforms and FMLA third-party administrators building notice-generation and notice-audit tooling, employment-law compliance vendors who review leave files for Department of Labor exposure, and HR document repositories that need to pair an employer notice with the provider certification it belongs to. The eligibility-and-deadline structure also generalises to state paid-family-leave notices.

Document complexity profile

41 fields across four pages — 22 text and 19 checkbox targets, with the checkboxes organised almost entirely as adjacent mutually exclusive pairs. Binding logic is modest (4 conditional bindings, 12 function calls, maximum expression depth 2) because the leave file's coherence is assembled by a registered computed-field module shared with the WH-380 certifications. The extraction difficulty is semantic rather than geometric: the document's meaning lives in which member of each pair is marked.

Key stats from our synthetic corpus

Quantitative characteristics of the WH-381 Notice of Eligibility and Rights & Responsibilities (FMLA) documents our generator produces.

How this document co-occurs with others

Rates at which identities in our corpus that produce a WH-381 Notice of Eligibility and Rights & Responsibilities (FMLA) also produce other documents.

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 and type counts come from the shipped WH-381 definition in the SymageDocs form library. No real employee, employer, or leave data was used at any stage.

Frequently asked questions

What data format do synthetic WH-381 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 41 fields across the four pages, including every checkbox in its checked or unchecked state. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.

Does the notice agree with the certification for the same employee?

Yes, by construction. The leave-reason checkboxes on the notice line up with which certification that identity produces, and where the reason is a family member's health condition the relationship checkbox names a person who exists in the same simulated household. That agreement is what lets you build a cross-document consistency checker and score it, rather than only checking fields within a single page.

Are the eligibility numbers actually consistent with the determination?

Yes. Months of service and hours worked in the preceding twelve months are computed from the simulated employment record, and the eligibility box reflects them. Every generated notice in the default corpus is an eligibility grant, so if you need denial cases for a completeness model you should treat this corpus as the positive class and construct negatives separately.

Which parts of the notice vary across the corpus?

The leave reason, the family relationship, the paid-leave substitution decision and its conditions, the reporting requirement and its frequency, the contact people, and all dates and durations vary. The key-employee determination and the calendar-year leave-year method are fixed in the current generator, so those two boxes read the same on every document.

Can I use this data commercially?

Yes. Every employee, employer, contact person, and policy detail is synthetic, describes no real organisation or worker, and is licensed for commercial use including model training, benchmarking, and redistribution inside your own products.