How I think

I spent a decade reading people in therapy rooms, and I read systems the same way now — looking for where the load actually lands, what a structure rewards when no one is checking, and how far a product's promises sit from what it does to the people using it. That clinical training still runs underneath everything I build.

Case study Tradeoff Most comparison tools rank on price, which is rarely how people actually choose. This one is built around the trade-offs they really make. Read the case study → Case study Fair Pay Calculator Group practices argue about the percentage split, when the fairer question is what's left after overhead. This tool runs that math. Read the case study →

Process

The work tends to follow the same shape. I notice a pattern I can't stop poking at, build something that tests whether it holds, and then keep that thing honest enough to argue with.

The open question

Every "which one should I pick" tool ends at price, and real choices almost never do. People trade money against time, against friction, against how well a thing fits the person they already think they are, and they stop optimizing long before a spreadsheet would tell them to. I wanted to see what a tool would look like if it took all of that seriously.

The pattern

The behavioral science already had the parts — diminishing returns, identity-based choice, the tax that friction puts on every decision — but nobody had wired them together into something you could actually use. Once I set them next to how people describe their real decisions, most of the model wrote itself: weight the axes, let the weights differ from one person to the next, and treat "good enough" as a real stopping point rather than a failure.

What it became

Tradeoff scores options across cost, time, friction, identity-fit, and diminishing returns instead of ranking on price. Every weight is visible and editable, because a decision tool you can't argue with is just an opinion wearing a progress bar.

Source on GitHub →

What it demonstrates

The habit underneath it is finding the real unit of a decision and building around that, rather than the number everyone reaches for first.

The open question

Group practices argue about pay in percentages — 60/40, 70/30 — as though the percentage itself were the fair part. The question that actually matters comes a step later: once the practice covers what it really costs to keep a clinician in the room, who gets the margin that's left? That was the number nobody seemed to be calculating.

The pattern

Fairness starts after overhead, not before it. A split that looks generous on gross can go thin once you subtract rent, software, supervision, and payroll, and a split that looks stingy can turn out fine once you see how little surplus the structure actually throws off. The math also moves with volume, since fixed costs spread across more sessions, so a full caseload is genuinely cheaper to carry than a ramping one. Any honest tool had to price that tension in rather than pretend a single percentage fits every clinician.

What it became

It estimates collected revenue per session, turns monthly overhead into a cost per completed session, and benchmarks the clinician's share against the margin left after overhead. It starts from a different baseline for W-2 and 1099 work, and it raises the bar as the structure produces more surplus. It's a working tool, and it's here as a demonstration you can run your own numbers through.

The calculator needs JavaScript to run, but the idea it demonstrates is simple enough on its own: judge a pay split by the margin left after overhead rather than by the gross.

Educational estimate. It's here to show the method, not to replace legal, accounting, or contract review.

What it demonstrates

It's the same move as Tradeoff: find the real unit — here, the margin left after overhead — and build the tool around that.

Writing

Longer thinking lives on Substack, under “Mostly, Though.” A few places to start:

Read the full archive on “Mostly, Though” (Substack) or on LinkedIn.

Context

I lead clinical AI product work in behavioral health — the part where clinical judgment has to survive contact with a model and a shipping deadline. Before product, I spent about a decade in clinical and supervisory work as an LMFT-S.

That's still the lens I bring to any system: who carries the load, where it breaks, and what actually gets rewarded when no one is watching. I tend to trust what people do more than what they say they'll do, and that's why the tools I build are the kind that show their own work.

Get in touch

Speaking, board and licensing work, consulting, or a collaboration worth the email. If you're looking for therapy, that lives over at Arc Psychotherapy.