We collected the official blank PDFs for 137 government form entries<\/strong> (134 analyzed across 133 distinct PDF files, totaling 63.4 MB<\/strong>, 628 pages<\/strong>, and 19,512 terminal form fields<\/strong>) and ran a structural inspection using The full dataset is available as a static, no-JS CSV under CC BY 4.0<\/strong>: – Direct CSV Download (137 rows \u00d7 26 columns):<\/strong> https:\/\/justfill.app\/data\/justfill-government-forms-fillability-2026-09-02.csv<\/a> – Methodology, Caveats & Sortable Table:<\/strong> https:\/\/justfill.app\/research\/are-government-forms-fillable<\/a><\/p>\n Alongside the structural government form study, we also published a reproducible 5-row PDF mail-merge benchmark fixture and real user correction logs (5,674 detected fields across 43 flat\/scanned documents: 2.2% false-positive deletions, 1.2% missed-field additions, 3.5% geometry adjustments): License: CC BY 4.0<\/strong> (Attribution: JustFill, 2026-09-02). Feel free to use the CSV for visualizations, PDF parser testing, or civic-tech research!<\/p>\n<\/div>\n submitted by \/u\/mcsno <\/a> pypdf 6.6<\/code> to measure how government agencies actually construct and publish PDF forms.<\/p>\nSchema (26 Columns):<\/h3>\n
\n\n
\n \nColumn<\/th>\n Type<\/th>\n Description<\/th>\n<\/tr>\n<\/thead>\n \n slug<\/code><\/td>\nstring<\/code><\/td>\nCanonical form identifier (e.g., w-9<\/code>, i-129<\/code>, fl-150<\/code>, form-2848<\/code>)<\/td>\n<\/tr>\n\n formName<\/code><\/td>\nstring<\/code><\/td>\nOfficial form number\/title<\/td>\n<\/tr>\n \n issuer<\/code><\/td>\nstring<\/code><\/td>\nFull issuing agency name<\/td>\n<\/tr>\n \n group<\/code><\/td>\nstring<\/code><\/td>\nAgency cohort ( IRS<\/code>, USCIS<\/code>, SSA<\/code>, VA<\/code>, State Dept<\/code>, DOL<\/code>, CMS<\/code>, California courts<\/code>, etc.)<\/td>\n<\/tr>\n\n country<\/code><\/td>\nstring<\/code><\/td>\nJurisdiction ( United States<\/code>, United Kingdom<\/code>, Germany<\/code>, Spain<\/code>, India<\/code>)<\/td>\n<\/tr>\n\n status<\/code><\/td>\nenum<\/code><\/td>\nok<\/code> (134), blocked<\/code> (2 \u2014 HTTP 403 outside India), no_pdf<\/code> (1 \u2014 online form only)<\/td>\n<\/tr>\n\n pdfUrl<\/code><\/td>\nstring<\/code><\/td>\nDirect agency URL fetched on collection date ( 2026-09-02<\/code>)<\/td>\n<\/tr>\n\n pages<\/code><\/td>\nint<\/code><\/td>\nPage count of the official blank PDF (total: 628, median: 3.0)<\/td>\n<\/tr>\n \n sizeKb<\/code><\/td>\nint<\/code><\/td>\nFile size in KB (largest: State Dept DS-64<\/code> at 3,150 KB)<\/td>\n<\/tr>\n\n fields<\/code><\/td>\nint<\/code><\/td>\nTotal terminal AcroForm input fields (excluding pushbuttons)<\/td>\n<\/tr>\n \n textFields<\/code>, checkboxes<\/code>, radios<\/code>, choices<\/code>, signatures<\/code><\/td>\nint<\/code><\/td>\nBreakdown by widget type (14,207 text, 5,029 checkboxes, 202 dropdowns, 58 radio groups, 16 signatures)<\/td>\n<\/tr>\n \n xfa<\/code><\/td>\nbool<\/code><\/td>\nWhether \/AcroForm<\/code> contains an \/XFA<\/code> XML package<\/td>\n<\/tr>\n\n encrypted<\/code><\/td>\nbool<\/code><\/td>\nWhether the file carries owner encryption<\/td>\n<\/tr>\n \n modifyDenied<\/code>, annotateDenied<\/code><\/td>\nbool<\/code><\/td>\nParsed PDF permission bits<\/td>\n<\/tr>\n \n textLayer<\/code><\/td>\nbool<\/code><\/td>\nWhether page 1 yields >= 50<\/code> extractable characters<\/td>\n<\/tr>\n\n producer<\/code>, pdfVersion<\/code>, edition<\/code><\/td>\nstring<\/code><\/td>\nPDF producer metadata, version ( 1.5<\/code>\u20131.7<\/code>), and regex-extracted revision string<\/td>\n<\/tr>\n\n namedFieldsShare<\/code><\/td>\nfloat<\/code><\/td>\nShare ( 0.0<\/code>\u20131.0<\/code>) of terminal field names containing a readable 4+<\/code> letter word token<\/td>\n<\/tr>\n\n verdict<\/code><\/td>\nenum<\/code><\/td>\nxfa_hybrid<\/code> (114), acroform<\/code> (15), flat<\/code> (5), xfa_dynamic<\/code> (0), scanned<\/code> (0)<\/td>\n<\/tr>\n\n note<\/code><\/td>\nstring<\/code><\/td>\nCollection anomalies (e.g., landing page 404 while PDF path remained live, empty \/AcroForm<\/code> dicts)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\nSummary Statistics from the Snapshot:<\/h3>\n
\n
xfa_hybrid<\/code>:<\/strong> 67\/68 IRS forms, 24\/26 USCIS forms, 10\/13 California court forms, 4\/4 SSA forms, and 5\/5 VA forms include an Adobe \/XFA<\/code> layer alongside AcroForm widgets.<\/li>\nDesigner 6.5<\/code> and Adobe LiveCycle Designer ES 8.2 \/ 9.0 \/ 10.0<\/code> dominate federal and state forms.<\/li>\n<10%<\/code> readable):<\/strong> 61 of 68 IRS forms have namedFieldsShare < 0.10<\/code> (most are 0.0<\/code>, naming every field f1_1[0]<\/code>, f1_2[0]<\/code>, etc.), whereas USCIS, SSA, and VA use semantic camelCase names (LastName<\/code>, ClaimantSSN<\/code>).<\/li>\nmodifyDenied<\/code>), and 16 deny annotations (annotateDenied<\/code>).<\/li>\nForm 709<\/code> leads with 1,851 fields<\/strong> across 10 pages, followed by USCIS Form I-129<\/code> (980 fields<\/strong>, 38 pages), USCIS Form I-485<\/code> (760 fields<\/strong>, 24 pages), and IRS Form W-2<\/code> (568 fields<\/strong>, 11 pages).<\/li>\n<\/ul>\nBonus Dataset: Real-World AI PDF Field Detector Correction Telemetry<\/h3>\n
https:\/\/justfill.app\/research\/ai-pdf-form-filling-reliability<\/a><\/p>\n
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
