The AI never does the math.
Every number in your plan—income tax, Social Security benefits, Roth conversion effects, debt payoff dates, retirement projections—comes from a calculation engine: a software built specifically to do financial arithmetic the way a tax preparer's software or an actuary's model would. It follows fixed formulas and published tax rules. Given the same inputs, it produces the same answer every time.
The AI, the chatbot you're actually talking to, never touches that arithmetic. Its job is to understand what you're asking, translate it into the right question for the engine, and explain the result back in plain language. Think of it as a translator standing between you and a calculator: the translator can explain what a number means and answer follow-up questions, but the calculator is what actually did the math.
Why this distinction matters
Large language models, the technology behind tools like ChatGPT, are built to predict likely-sounding text, not to compute exact figures. Ask a general-purpose AI chatbot to calculate your tax bill or your retirement savings, and it can produce numbers that sound plausible but are subtly, or often badly, wrong, and it won't tell you when it's guessing. There's no way to check its work, because there's no fixed formula behind the answer, just a prediction of what a good answer would look like.
Delta Prime's engine works the opposite way. It's ordinary, deterministic software: the same kind of rules-based code that runs payroll systems, tax software, and actuarial models. It doesn't predict, estimate, or improvise. It calculates.
How we validate it
- Federal income tax is validated to the cent against NBER TAXSIM, the academic gold-standard tax model used in hundreds of published studies. Our engine reproduces its federal income-tax figures exactly—so the federal numbers match an independent reference calculator, not just our own code.
- State income tax is a flat-rate approximation, not a full state-by-state computation, and it is not independently checked against a tax oracle the way the federal numbers are. It gives a reasonable estimate, but the state portion should be treated as an approximation.
- Social Security estimates are checked against the Social Security Administration's own published benefit factors.
- Retirement projections and Monte Carlo simulations use standard actuarial and statistical methods used across the financial planning industry, not a language model's approximation of them.
What the AI is actually for
The AI layer exists to make the engine usable in plain conversation: asking the right follow-up question, catching when a scenario needs a number you haven't given yet, and explaining a result in a way that makes sense for your specific situation. It's the interface. The engine is the math.
Projection, not a guarantee. Educational illustration, not financial advice.