Federal tax & formal methods / Research prototype
An inspectable starting point for federal tax rules.
A Lean model of the 2026 ordinary income-tax rate schedule makes filing status, source revision and bracket arithmetic explicit.
Make the first scope concrete
The Internal Revenue Code combines definitions, calculations, exceptions, elections and rules that change with time. Formalizing it starts with a precise unit of work. This release chooses the ordinary individual income-tax rate schedule for tax year 2026 and two filing statuses: single and married filing jointly.
The input is reviewed taxable ordinary income. The model does not derive taxable income from receipts or decide which deductions a taxpayer can take. That keeps the initial arithmetic experiment inspectable while leaving the upstream classification work visible.
Statute and annual parameters
A current tax calculation needs more than a bare rate table copied from statutory text. The model binds its supported year and filing status to an explicit source revision, using the IRS’s published 2026 inflation-adjusted schedule as its numerical reference. The source manifest records the official publications and the exact modeled scope.
The relationship between statutory rules and annually published parameters is a useful formalization problem in its own right. A changed year must select a reviewed schedule rather than reuse plausible-looking thresholds from a different return period.
A calculation with an explicit boundary
The implemented calculator allocates taxable income across progressive brackets. Its integer arithmetic and stated final rounding convention make the result reproducible. The checker also requires the supported tax year, source revision, reviewed input and an in-scope filing status.
That result is a rate-schedule illustration. Actual return instructions can require tax tables, worksheets or other computation methods, and preferential capital-gain treatment can change the calculation. Credits, alternative minimum tax, other taxes, payments and withholding also affect a return’s final result. Those dependencies are outside this first model.
Proofs and boundary cases
The repository states which properties are proved about its acceptance guard and arithmetic. Its synthetic examples exercise supported cases, source and year mismatches, unsupported conditions and bracket boundaries. Inspecting those statements alongside the code is more informative than treating a test pass rate as tax accuracy.
A complete evaluation would need an independently implemented reference calculation and expert-reviewed cases under the applicable tax instructions. It should distinguish arithmetic errors, unsupported cases admitted, incorrect taxable-income facts and differences caused by the required return computation method.
Toward a larger tax model
The next modules could formalize individual definitions and their dependencies, with precise source citations and effective dates. A language model could propose typed facts and evidence spans; a checker could enforce the selected rules. Each new module needs its own acceptance criteria and evaluation.
The first contribution is a small executable rule model and a paper that states its assumptions. Wider tax-code coverage should grow from reviewed source-to-rule mappings, preserving the ability to replay a result and explain which facts and rule revision produced it.