Stuff I've learned

The first audit of my career happened on a garbage route.

High-school summers, I collected trash across a sprawling acreage of apartment complexes — picking and stashing loose wrappers and cans. My supervisor ran quality control from a pickup. If he spotted trash in my section, he drove me back to the start of the route.

What that teaches you isn’t don’t skip sections. It’s that half-baked work is not a savings — it’s a loan, and the interest is more work later. The audit wasn’t an event at the end of the season. It was live, continuous, and idling a hundred feet behind me.

Garbageman, bookkeeper, billboard salesman, CPA, product manager, founder. It’s an atypical path, and telling it as a sequence would miss what made it useful.

The jobs weren’t steps. They were the same handful of lessons, arriving in different uniforms.

---

The bookkeeping came in college, part-time, back when bank reconciliations were still done by hand. That’s where I learned how cash actually behaves — and why the balance on the bank statement had to be “adjusted” at month end before it told the truth. The bank’s number was right. The books’ number was right. The work was the relationship between them.

The same job taught me what a few disciplined policies can do to volume. We turned millions of lines of raw music sales — CDs were still a thing — into royalty calculations tied to contract terms. Whether an artist had licensed five of an album’s eleven songs or six meant thousands of dollars in pro rata revenue. I learned that one the hard way: a long email thread finally resolved whether a song was in scope, and I forgot to change the math. The data hadn’t moved. The words had. Business is never linear, and the numbers are downstream of the agreements.

---

Later in college I sold billboards, and the calendar came apart.

I sold them in August. I wasn’t paid until the following January, after the boards went up for the Christmas campaign. The sale, the work, and the cash were three different events in three different months — each with a defensible claim to being the one that counted. Anyone who has waited out the gap between a commitment and its money understands performance obligations in a way no textbook delivers.

---

Then the CPA years, at a Big 4 firm, where I was handed the rule I’ve repeated most since: never write the same number in pen in two different places. I was chastised — correctly — for a summary sheet with values typed into it rather than formula-driven. A typed number looks identical to a derived one, right up until something upstream changes, and then it lies with total confidence. So I made everything derive, gradually turning my manual processes into long ifna(index(match),match) chains — if you know, you know. The workbook got less impressive to look at and much harder to fool.

---

Building technology inside financial institutions scaled all of this up.

A loan feed wasn’t valid because it arrived. It was valid when the sum of its dollars and the count of its disbursements both zeroed against the cover sheet’s stated amounts. Two independent traces, agreeing — that was the test, and the contents justified it: a loan was never just disburse-and-repay. Partial payments, failed disbursements, and write-offs ruled the day.

A stock trader was never just a stock trader, either. Margin enabled or not, pattern day trader or not, in-flight trades or none — all of it bore on whether the next trade could be fulfilled, in a world where trades settle in microseconds. You can’t meet that with a pile of if-thens. You need policy and you need technology, and they have to be in sync: the rule legible in the system, the system answerable to the rule.

And whether tendered shares of a private security are taxable — and how — turned out to be a function of the type of stock, where the company is located, where the employee lives, and the date they sold. Every one of those answers had to be not just calculated but provable. Right wasn’t enough. Showably right was the job.

---

Read those back and a shape emerges that I never planned.

Audits work best live, because half-done work compounds. The numbers are downstream of agreements, and the agreements keep changing. The sale, the work, and the cash are different events in different months. A number gets written once and derived everywhere else. A feed earns trust when independent traces zero against each other. Policy and technology move together or not at all. And an answer that can’t prove itself isn’t finished.

When I finally sat down to build TallyUp — a record meant to hold what a business knows, with the statements falling out as byproducts — I didn’t have to invent its principles. They were already there, in the uniforms: the supervisor on the route, the adjusted bank balance, the January check for the August sale, the summary sheet that had to derive, the cover sheet that had to zero.

I just had to write down stuff I’d learned.