PaycheckLink

I've been building PaycheckLink, a free paycheck/take-home-pay calculator for all 50 US states. A few weeks in, I decided to actually verify the tax math instead of assuming it was right — and I'm glad I did.

What I found:

• Federal withholding was underpaying by about 15% on a standard biweekly paycheck — turned out the engine was using stale IRS Pub 15-T constants (a Step 1g deduction figure that was nearly double what it should've been).

• California was missing the state's personal exemption credit entirely — a small but real ~$150/year gap that compounds across every CA user.

• Georgia's tax rate was stale at 5.19% when the actual current rate is 4.99% — the state passed a mid-year rate cut (HB 463, signed May 2026, retroactive to January 1) that my data source hadn't caught up to yet. Same pattern hit Utah and South Carolina.

• New York City's local tax was silently missing from the default calculation — a real gap since NYC is one of the most-searched paycheck queries there is, and roughly 40%+ of NY State's population lives there.

What I learned:

The scariest bug wasn't any single wrong number — it was realizing that "tax rate for state X" isn't a fact you look up once and cache forever. States pass legislative changes mid-year, sometimes retroactively. A calculator that's "correct" in January can be quietly wrong by June.

So beyond fixing the specific bugs, I built:

• A golden test suite with hand-verified benchmarks for every major scenario (single/MFJ, no-tax states, two-layer local tax stacks)

• An automated content-drift check that scans every hardcoded rate/dollar claim in the site's copy and fails the build if it doesn't match the live tax engine — because it turns out static prose drifts out of sync just as easily as code

• The same lockfile-based sync check between the website and the Android app I'm building, so the mobile version can't silently diverge from the web engine either

Why I'm sharing this:

If you're building anything that touches "current law" or "current rates" — tax, shipping costs, interest rates, whatever — a one-time data pull isn't enough. The accuracy work isn't a launch task, it's an ongoing process, and building the automated checks early saved me from finding these the hard way, in public, after launch.

Happy to go deeper on any of this — the tax engine architecture, the drift-detection tooling, or the SEO/backlink side I've been working on in parallel. What's the trickiest "this data goes stale" problem you've run into in your own product?