The computed tax, independently checked line by line.
Passing the test deck shows the return was accepted. It doesn’t show that every line is right for every filer. Every bracket moves every year, and the refund should be able to explain itself.
The explainer · 3 min 31
The annual release, the test decks as cases, the back-test, an engine move, and one mismatch on a stale standard deduction. Deduction per IRS, every other number invented.
The result
VeriQuant computes an independent tax for every case from the schedules your content team approves, and grades the engine’s output against it. Every case is a match, or a mismatch that names the first line where the two calculations part ways.
Here the engine is still on last year’s standard deduction and brackets. Seven lines move, the refund is short, and the report points at the first row that differs.
Drive the product’s own interview or return screens per case and match the computed lines against the rulebook.
Call the calculation API in a non-production environment and grade every returned line.
Exchange a sheet of cases and computed results in your field names. VeriQuant grades the sheet.
Brackets, deductions, phase-outs and credit limits load as data with a tax year, and a mid-year act lands as a new edition with its effective date. Every case re-grades against the edition in force for its filing. The report names the first line that moved and the row that moved it, the way a reviewer would.
IRS Assurance Testing System scenarios and state test decks load as cases without reshaping. So do your own: the filers your support team hears from, the edge cases nobody trusts. Each is graded line by line, and anything unexplained stays open rather than counted as passed.
Honest fit
It carries no tax content of its own. It computes an independent result from the schedules you approve and grades what your engine produced. It verifies the engine. It does not replace it.
Forms round per line, some jurisdictions truncate, and worksheets carry more precision than the form they feed. Each convention is declared in the rulebook, and agreement to the cent is the bar.
Your tax content never leaves your network. VeriQuant runs inside it as a container or a single Java file, with an offline licence, no telemetry and your own LLM for the drafting step.
Who runs it
The content analyst edits a bracket or a phase-out as data, with an author and a diff. Nobody opens a test script. Engineering gets the same check in CI, on every build.
Consumer and professional tax software
Individual and business returns, graded line by line before filing season opens.
Transaction tax engines
Sales, use and VAT determination checked per transaction against the rate and rule tables in force.
In-house tax platforms
One independent check that survives the annual release and the next legislative change.
The test deck shows the return was accepted for the scenarios the agency chose. It does not show that every line is right for every filer you serve. VeriQuant grades the engine’s output line by line, on the agency’s cases and on yours.
No. Drive the screens, call the calculation service in a non-production environment, or exchange a sheet of cases and computed results with no API access at all.
No. It computes an independent result from schedules you approve and grades what your engine produced. It carries no tax content of its own.
They load as an edition with a tax year. Every case re-grades against it, and the report names the first line that moved and the row that moved it.
A form, schedule or jurisdiction we don’t list? We build it during your evaluation.
The next step
The annual release. A mid-year act. A phase-out nobody trusts. In 45 minutes, a senior member of our team walks your calculation lead, content team and CIO through how the check is built for that case, and how your team runs it in your own non-production environment.