Synthetic Form 7004 - Application for Automatic Extension Data
Synthetic training data — no real PII, fully coherent identities
Generate synthetic Form 7004 business extension requests, including the two-digit form code printed one character per box that identifies which return is being extended. Compact on the page, deepest binding logic in the catalog.
25
Fields per document
1
Page
Tax
Category
What this document is
Form 7004 is the application for an automatic extension of time to file a business income tax return. A single entity uses it to extend any of dozens of different returns, and which one is indicated by a two-digit form code printed into individual character boxes at the top of the page. The rest of the form is the entity's identity, a tentative total tax, total payments and credits, and the resulting balance due.
Why generate synthetically
The 7004 is a routing document: getting the form code wrong sends an entity's extension to the wrong return entirely. It is also a comb-field extraction problem — one digit per box — which is a distinct OCR task from reading a free-form numeric field, and one that public form corpora barely cover. Synthetic 7004s give you labeled character-cell extraction plus the entity data around it.
What makes synthetic data useful
The form code is generated consistently with the entity type behind the document, so the code identifies a return the entity could actually file rather than a random two-digit pair. Tentative tax, payments, and balance due are computed from the entity's financial model, which means a 7004 and the corporate or partnership return it extends can be generated as a matched, arithmetically consistent pair.
Training challenges
The form code is split across two adjacent single-character boxes, so a model has to read each cell independently and then join them in the right order — the classic comb-field failure mode, where transposition produces a valid-looking but wrong code. Despite its small field count the 7004 carries the deepest binding logic of any form we publish, because the tentative tax computation branches on entity type before it computes; that means the document's values are tightly coupled and a partial extraction is not merely incomplete but internally inconsistent. Eight checkboxes covering entity and filing circumstances sit on the same compact page as the numeric block, mixing extraction modes at close range.
Generate synthetic Form 7004 - Application for Automatic Extension data
Start with 500 free credits. No credit card required.
Generate NowWho uses this data
Business tax-prep and compliance platforms, entity-management and corporate-secretarial software, accounting-firm workflow tools that track extension deadlines across a client book, and IDP vendors building tax-mail routing. The comb-field pattern also transfers directly to any document that prints codes one character per box — remittance forms, government applications, and claim forms all do it.
Document complexity profile
25 fields on a single rendered page: 14 text, 8 checkbox targets, and 3 numeric fields, including the two single-character form-code cells. 25 annotation relations. Binding logic is disproportionately deep for the size — 16 conditional bindings, 5 arithmetic bindings, and 35 function calls at a maximum expression depth of 4, with the highest average binding depth of any form in our catalog — because the tentative tax branches on entity type before computing.
Key stats from our synthetic corpus
Quantitative characteristics of the Form 7004 - Application for Automatic Extension documents our generator produces.
| Metric | Value | Detail |
|---|---|---|
| Population coverage | 64% | 64% of identities in our 1,000-identity corpus are attached to an entity and therefore 7004 eligible. The remaining 36% have no business relationship. |
| Average binding depth | 1.4 | The highest average expression depth of any form we publish, at a maximum depth of 4. Small on the page, deep in the logic — the opposite of the usual size-to-complexity relationship. |
| Most common form code | 43% | The most frequent form code accounts for 43% of the corpus, the second 38%, and the third 19%. Extension traffic is genuinely skewed, and a routing classifier should learn that prior. |
| Median tentative total tax | $4,615 | Median tentative total tax is $4,615, p25–p75 of $2,168 to $11,317, maximum $765,685. The figure is computed from the entity's financial model, so it agrees with the return being extended. |
| Median balance due | $748 | Median balance due is $748 against median payments of $3,732. Unlike the individual Form 4868, most business extensions carry a real payment — a distinction worth training a classifier on. |
How this document co-occurs with others
Rates at which identities in our corpus that produce a Form 7004 - Application for Automatic Extension also produce other documents.
| Correlation | Rate | Detail |
|---|---|---|
| Extends a corporate return | 100% | Every 7004 entity can also produce the Form 1120 the extension covers, giving you extension-and-return pairs whose form code should match the return type. |
| Extends a partnership return | 9% | 9% of identities are attached to a partnership and can produce the Form 1065 the extension covers. That subset is where form-code routing actually matters, because the wrong code sends the extension to the wrong return type. |
| Extends an association return | 100% | The same entity can produce a Form 1120-H, another return the 7004 can extend — useful for training a model to distinguish extension targets within the 1120 family. |
| Files quarterly payroll returns | 100% | Every entity in the corpus files Form 941 quarterly, so an extension request can be generated alongside the payroll filings that establish the entity is active. |
| Entity W-9 on file | 100% | Every entity can co-generate a W-9 carrying the name and TIN on the extension header — the identity anchor for matching an extension to an entity record. |
All stats above are corpus-derived: they were computed on a local synthetic corpus of 1,000 generated identities produced by SymageDocs' World Simulation Engine, with 7004 figures measured over the entity-eligible subset. No real entity or financial data was used. `make corpus-stats` regenerates the shipped corpus_stats.json for the site's analytics form set, which does not include this edition; reproduce these figures with `python -m symagedocs.corpus_analysis --corpus-size 1000 --form-id irs_f7004_standard_2024`.
Frequently asked questions
- What data format do synthetic Form 7004 documents include?
- Each generated identity produces a rendered PDF plus a structured JSON annotation file with bounding boxes and ground-truth values for all 25 fields, including the two form-code character cells annotated separately. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.
- How is the form code represented in the annotations?
- As two separate single-character fields, each with its own bounding box and ground-truth digit. That is what lets you train and evaluate comb-field reading directly, including the transposition errors that a joined-string annotation would hide.
- What form codes appear in the corpus?
- Three code values are present across the corpus, at roughly 43%, 38%, and 19% frequency, each matching the entity type generated behind the document. The imbalance is deliberate: real extension traffic is dominated by a few return types, and a classifier should be trained on that skew rather than a uniform distribution.
- Why does such a small form have complex bindings?
- Because the tentative tax computation branches on entity type before it computes. The 7004 carries 16 conditional bindings, 5 arithmetic bindings, and 35 function calls at a maximum expression depth of 4 — the highest average binding depth of any form we publish, on one of the smallest field counts.
- Can I use this data commercially?
- Yes. All synthetic data is generated from statistical models, contains no real entity or financial data, and is licensed for commercial use including ML model training and benchmarking.