Synthetic Form 1065-X - Amended Return or Administrative Adjustment Request Data
Synthetic training data — no real PII, fully coherent identities
Generate synthetic Form 1065-X amended partnership returns and administrative adjustment requests, built as real as-filed / net-change / corrected triples across five pages. The partnership counterpart to the 1040-X, with far denser currency columns.
267
Fields per document
5
Pages
Tax
Category
What this document is
Form 1065-X is the amended return and administrative adjustment request for partnerships. Like the 1040-X it uses a three-column structure — the amount as originally reported, the net change, and the corrected amount — but applies it across the full partnership income statement: ordinary business income, rental income, other rental income, guaranteed payments, and the rest of the distributive-share lines, over five pages.
Why generate synthetically
Partnership restatements are high-stakes and low-volume: they matter enormously to fund administrators and BBA audit workflows, and there is no realistic way to assemble a training corpus from real filings. Synthetic 1065-Xs give you the amendment structure at partnership scale — hundreds of currency cells arranged in signed triples — with exact ground truth on every column.
What makes synthetic data useful
Each 1065-X is generated as a genuine restatement of a coherent partnership: the as-filed column is a plausible original return, the net-change column carries signed adjustments that are frequently negative, and the corrected column reconciles. Because the underlying entity is the same one that can produce a Form 1065, you can generate the original and the amendment as a matched pair rather than two unrelated documents.
Training challenges
This form has the densest signed-currency layout in the library relative to its size: the great majority of its fields are currency cells arranged in as-filed / change / corrected triples, so column assignment errors are both easy to make and expensive. Net-change values are routinely negative — the interquartile ranges on the major income lines straddle zero — and a mis-read sign inverts the direction of the entire adjustment. Its annotation graph is larger than its field count, because single labels govern multiple columns, so a naive one-label-one-value assumption breaks immediately. The five pages repeat visually similar three-column blocks, which page classifiers tend to collapse.
Generate synthetic Form 1065-X - Amended Return or Administrative Adjustment Request data
Start with 500 free credits. No credit card required.
Generate NowWho uses this data
Fund and partnership administrators handling restatements, BBA centralized-partnership-audit workflows, K-1 correction and re-issuance pipelines, tax-controversy and audit-defense platforms, and accounting-firm document automation. The signed before/after/delta pattern also transfers directly to restated financial statements and contract amendment review outside tax.
Document complexity profile
267 fields across 5 pages, dominated by 178 currency amounts arranged in as-filed / net-change / corrected triples, plus 53 text, 32 checkbox targets, 2 date, and 2 numeric fields. 425 annotation relations — well above the field count, because a single printed line label governs all three columns of its row. Binding depth is shallow (2 arithmetic bindings, 3 function calls), since the difficulty is column geometry and sign handling rather than expression evaluation.
Key stats from our synthetic corpus
Quantitative characteristics of the Form 1065-X - Amended Return or Administrative Adjustment Request documents our generator produces.
| Metric | Value | Detail |
|---|---|---|
| Population coverage | 9% | 9% of identities in our 1,000-identity corpus belong to a partnership and are therefore 1065-X eligible. Requesting the form constrains the population to partnership entities for a full-size corpus. |
| Currency cells per document | 178 | 178 currency cells per return, the great majority of them in three-column triples. Column assignment across those triples is the dominant evaluation axis, well ahead of character-level OCR accuracy. |
| Median ordinary income adjustment | $394 | The median net change on ordinary business income is $394, but the p25–p75 range runs from -$9,597 to $9,224. Half the corpus carries downward adjustments, which is what makes it useful for sign-sensitive training. |
| Annotation relations | 425 | 425 label-to-value relations against a smaller field count, because each printed label governs three columns. Relation extraction, not box detection, is the harder half of this document. |
| Fields carrying a value | 85% | 227 of the form's fields carry generated values on a typical filing; the remainder render blank as they do on a real restatement. The populated share is far higher than on the original Form 1065, because an amendment restates lines rather than leaving optional schedules empty. |
How this document co-occurs with others
Rates at which identities in our corpus that produce a Form 1065-X - Amended Return or Administrative Adjustment Request also produce other documents.
| Correlation | Rate | Detail |
|---|---|---|
| Amends a matching 1065 | 100% | Every 1065-X identity can also produce the original Form 1065 it amends, giving you as-filed and restated partnership returns under one entity. |
| Also files payroll corrections | 100% | The same partnership can produce a Form 941-X, so a single entity yields both an income-tax restatement and a payroll-tax correction — the combined workload a tax-controversy platform actually sees. |
| Files quarterly payroll returns | 100% | Every synthetic partnership files Form 941 quarterly, so wage figures restated on the 1065-X can be checked against the payroll filings for the same period. |
| Partners file personal 1040s | 100% | Every partnership identity produces a personal Form 1040 as well. A partnership restatement flows through to amended partner returns, and this pairing lets you model that chain end to end. |
| Entity W-9 on file | 100% | Every partnership can co-generate a W-9 carrying the entity name and TIN that appear on the 1065-X header — the identity anchor that ties a restatement to the right entity record. |
| Married principals | 51% | 51% of synthetic 1065-X principals are married, and 8% have a self-employed spouse — related-party partnership structures that de-duplication logic must not collapse. |
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 1065-X figures measured over the partnership-eligible subset. No real partnership or partner 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_f1065x_standard_2024`.
Frequently asked questions
- What data format do synthetic Form 1065-X documents include?
- Each generated identity produces a rendered PDF plus a structured JSON annotation file with bounding boxes and ground-truth values for all 267 fields across five pages, with the as-filed, net-change, and corrected columns annotated as distinct fields. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.
- Are the net-change columns genuinely signed?
- Yes. On ordinary income, rental income, and guaranteed payments the interquartile range of the change column spans negative to positive values, so roughly half the adjustments in the corpus reduce a reported figure. Sign handling is the most common defect in amendment extractors, and this corpus exercises it in both directions.
- Can I generate the original 1065 alongside the amendment?
- Yes. The same partnership identity produces both, so you get an as-filed return and its restatement as a matched pair — the configuration you need to train or evaluate change-detection rather than single-document extraction.
- Why are there more annotation relations than fields?
- Because one printed line label governs three columns. Each column is its own annotated field, and each carries its own relation back to the shared label, so the relation count exceeds the field count. That structure is precisely what a relation-extraction model has to learn on a three-column form.
- Can I use this data commercially?
- Yes. All synthetic data is generated from statistical models, contains no real partnership or partner data, and is licensed for commercial use including ML model training and benchmarking.