Can a prompt replace Gusto?

Build a payroll *model*. It computes and warns; it never files and never pays. Say that in the first line of the README, and say that tax rules change and this is not advice.

Stack: a local application, SQLite, one command.

Gross-to-net calculator with every rate loaded from a data file **you maintain**, dated, so a historical run reproduces exactly what the rates were then. Ship the file empty with a documented shape and a worked example, rather than baking in numbers that will be silently wrong next year.

Compute and show every line separately, never a single net figure: gross, each employee-side deduction, each employer-side contribution, and net. The employer cost total is the number founders consistently underestimate and it should be the largest thing on the screen.

Scenarios, which is what makes this worth building:
- What a raise of X actually costs, employer-side included.
- Salary versus contractor for the same work, with the caveat that classification is a legal question and not the calculator's to answer.
- The fully loaded cost of the next hire, including the employer contributions and any per-person software fee.

Calendar: a filing and payment schedule you enter once per obligation, with a recurrence and a lead time, producing a list of what is due in the next thirty days and an alert as each approaches. It tracks; it does not file.

Run records: each modelled pay run stored with its inputs, the rate-file version used, and the outputs, so a figure can be reproduced months later.

A reconciliation view: modelled totals against what your real provider actually charged, imported from a CSV, with the differences highlighted. That check catches a wrong rate or a missed deduction before an auditor does.

Write tests for gross-to-net against hand-computed examples, for the dated rate file resolving correctly for a past period, and for the calendar firing on the right day across a month boundary.

Do not attempt filing, payments, or benefits enrolment.