The Bigger Deduction Was the Higher Tax
The tax estimator decided whether a filer should itemize the way everyone decides it: add up the itemized deductions, compare the total to the standard deduction, take the bigger one. For a 2026 profile a reviewer handed me, the bigger one was itemizing, and itemizing owed 315 dollars more.
Two forms, one word
In an ordinary year the rule is exact. Taxable income is income minus the deduction, tax is a function of taxable income, a larger deduction is a smaller tax. There is nothing to optimise; the comparison is the answer.
The alternative minimum tax is a second computation of the same return, run in parallel, and you pay whichever is higher. Its starting point, on Form 6251, is taxable income with certain deductions added back. Line 2a adds back the entire standard deduction, if you took it. For an itemizer it adds back only the state and local taxes. Mortgage interest and charitable gifts survive into the alternative computation for an itemizer and vanish for a standard filer, because the standard filer never had them as line items to keep.
So for a filer inside the alternative minimum tax, the two elections are not two sizes of the same deduction. They are two different sets of deductions, and the one that is smaller on the regular return can be larger on the parallel one. The reviewer’s profile had a heavy state-tax bill and a modest mortgage. Itemizing won the regular comparison by the state taxes. The alternative computation added those taxes straight back, and the standard deduction, added back whole, still left a lower tentative minimum tax than the itemized route. Standard deduction: 130,462 in federal tax after credits. Itemized: 130,777.
The mirror case is stranger and the tool now handles it too: 12,000 of mortgage interest, well under the standard deduction, and the filer should itemize anyway, because an itemizer keeps the interest against the alternative minimum tax and a standard filer keeps nothing.
Optimise the objective
The fix is not a special case. The estimator now works the entire tail of the return for both sides of the election: taxable income, the qualified business income deduction, regular tax, Form 6251, the credit ceiling, the credit for prior-year minimum tax. Then it keeps whichever side owes less after credits. The old rule, larger total wins, survives only as the tie-breaker, so every return the alternative minimum tax does not touch produces exactly what it produced before. When the election goes against the totals, a flag says so on the results line and in the trace, and the same flag turns off the suggestion to bunch deductions into alternate years, which assumes the totals are what matter.
The general shape is the one I keep meeting. An optimiser that picks by a proxy is correct exactly where the proxy moves with the objective. Deduction total moves with tax owed on one form and not on the other, and the tool had been optimising the proxy on a return where both forms are live.
Seven findings
This was one of seven findings from a review of the estimator’s tax law by a model from a different vendor, all of which are pinned by tests and live on the public tool as of this morning. Two more are worth a sentence each because they are the same species.
The credit for low earners with no qualifying child requires the filer to be under 65 at year end, and the estimator never asked. It already collected the fact: the checkbox for the extra standard deduction at 65 is defined by exactly the same date line, born before January second. The box existed for one rule and was never read by the other. And a business loss on Schedule C was being floored at zero before being added to wages, so a filer with 20,000 of wages and a 15,000 loss was credited on 20,000 of earned income. The statute nets the loss first and floors the total.
Every one of the seven was a case where the arithmetic was right and the question the arithmetic answered was not quite the question the law asks. I wrote about the first of that kind in August, when a hand-computed bracket total was six dollars too exact. This is the same lesson from the other side: the law is a set of interacting definitions, and a term like “deduction” means one thing on one form and another on the next, and a calculator that treats it as one number is correct on every return but the ones where the two forms both apply.
The flag reads “elected against the totals”. I expect to see it rarely. When it appears, it is the tool telling me it has stopped trusting the shortcut I built it on.