We rebuild real companies' filings from public EDGAR data and score them against what was actually filed: 10-Qs and 10-Ks, Regulation A reports, proxies, and more. Every number links to the filed document. Check us.
And none of this files itself. Your own securities attorney reviews and signs every filing before it reaches the SEC; the software cannot skip that step.
234 / 234
TakePublicTakePublic proofRebuilt figures match the filed ones35 real companies. Generated 2026-08-10.
35 companies, one standard
Each card is a company rebuilt from its published data, as if it were a customer, then compared figure by figure with what it filed. Apple's 52/53-week fiscal calendar, MITT's bank-style REIT balance sheet: different shapes, same checks. Cards with the TakePublic chip link to a published draft you can read cover to cover.
No company matches. Clear the search to see all 35.
The 35 companies above get the full line-by-line treatment. The same derivation then runs across the whole filer universe: every SEC registrant with a 10-Q filed in the last six months and every registrant with a 10-K filed in the last fifteen months.
10,060
real filings rebuilt and checked
9,926
pass every check
0
showed a figure that contradicts the filing
9,926 pass every check134 safe declines, no statements produced
✓Balance sheet tiesFor every filer, rebuilt assets equal liabilities plus equity.
✓Net income matchesThe rebuilt figure matches the filed one, from the company's own public XBRL facts.
✓No line overclaimsNo rebuilt line claims more than a filed total.
✓Declines are safeWhen a filer's public data is missing or self-contradictory, the system refuses to produce statements rather than produce wrong ones.
Run 2026-08-07, repeated on a weekly schedule.
Every form, the same gate
The 10-Q was first. The same capture, draft, and score loop now covers 77 real filed documents across 8 form types, and every fixture must hit its full mechanical ceiling before any change merges.
79
gold fixtures in the regression gate
8
form types, one scoring standard
100%
of fixtures at their mechanical ceiling
10-Q
35 fixtures · gate 55 / 55Form 10-Q
Rebuilt from public XBRL facts and scored line by line against the filed quarterly report. Real-model narrative runs: 93.1 to 97.3 / 100 (35 runs, claude-fable-5). Apple Inc., JPMORGAN CHASE & CO, Hims & Hers Health, Inc., and 32 more.
Rebuilt from public XBRL facts and scored line by line against the filed annual report. Real-model narrative runs: 90.9 to 97.9 / 100 (27 runs, claude-fable-5). Apple Inc., JPMORGAN CHASE & CO, Hims & Hers Health, Inc., and 24 more.
Scored against comparative anchors extracted from the filed HTML; Regulation A filings carry no XBRL. Real-model narrative runs: 83.8 to 87.5 / 90 (5 runs, claude-fable-5). Ark7 Properties LLC, Worthy Property Bonds, Inc., Innovega Inc., and 2 more.
Scored against the registrant's filed XBRL fact history for the registration statement. Real-model narrative runs: 84.7 to 87.4 / 90 (2 runs, claude-fable-5). HyOrc Corp, MINERALRITE Corp.
Plan statements scored against the sponsor's filed 11-K, with both statement identities tied. Real-model narrative runs: 84.7 to 86.9 / 90 (3 runs, claude-fable-5). Walmart Inc., HOME DEPOT, INC., COCA COLA CO.
A deterministic gate on the generated Pink disclosure package; synthetic fixture, no filed twin.
The narrative, scored separately
96.7 / 100
HIMS full document
Model claude-fable-5Run 2026-08-06
The numbers above come from the books; the AI writes the prose. Scored against the filed report, all 103 stated figures grounded. Before any customer filing moves, their securities attorney reviews and signs it.
Read our output before you decide
Complete generated documents, scored line by line against what each company actually filed. Open one and read it cover to cover.
TakePublic drafts your next 10-Q from your books, your securities attorney signs it, and we file it to EDGAR, Inline XBRL included. From $2,999 per month.
These companies are not TakePublic customers; their filings are public records, chosen because anyone can check them. Method: XBRL facts pulled from EDGAR, a trial balance derived the way connected books feed the product, the 10-Q regenerated through the same pipeline customers use, and the output scored against the filed document. The universe check applies the same derivation to every registrant with a recent 10-Q or 10-K on EDGAR and asserts the balance sheet ties, net income matches the filed figure, and no line claims more than a filed total. Statement scores and universe checks are deterministic and involve no AI.