I recently solved a problem that has been bothering me for almost a year now. For reasons no one wants me to get into, I had been attempting to model US federal marginal income tax brackets for individuals under 26 USC §1. The math is pretty straightforward. The brackets are adjusted each year according to the average CPI-U for the period September TY-2 through August TY-1. There are some quirky rounding rules, but I had managed to get all the numbers in the model to match the actual tax brackets since 1993 without difficulty. Except one--the 35.5% bracket, which the math kept saying should have had an upper limit of $297,300, but which had actually applied up to $297,350. No matter how I tweaked the rounding rules, I couldn't make it match without throwing a dozen other things out of whack.

As it turns out, BLS, who publish the CPI-U, had in late-2000 discovered a computing error that affected the reported CPI-U from January 1999 through August 2000. While BLS published corrections for January-August 2000, they never published the corrections for January-December 1999. But they gave the corrected numbers to the IRS under a tax statute revision that passed Congress late in 2000 for IRS to use in computing the 2001 tax brackets. I managed to find the unpublished figures here. Using the corrected numbers in calculating the September-August average CPI-U according to the IRS rule finally corrected my erroneous bracket!

And there was much rejoicing!