[ how the lab works ]

Method, not magic.

Every experiment runs through the same six phases. Some die at sketch. Most die at sandbox. The ones that ship were shaped by the discipline, not by the inspiration.

01Notice

02Sketch

03Sandbox

04Ship

05Live

06Decide

Phase 01 · Notice

What problem keeps reappearing?

Every experiment starts with an observation. Not a market opportunity — an observation. A friction in a real day, watched closely.

I keep a notebook for these. It runs on paper, then on Apple Notes, then back to paper. The medium doesn't matter; the seriousness of looking does.

The Notice that became FoodNode: a procurement manager in Pune calling thirteen suppliers to find one verified turmeric source. The cost of the call was the cost of trust missing from the internet.

Phase 02 · Sketch

What is the smallest possible version that contains the idea?

A sketch is not a prototype. A sketch is a piece of code, a paper diagram, a single screen — wide enough to be wrong about, narrow enough to ship in a week.

The sketch's job is to answer one question: is this idea worth the prototype? If the sketch already feels like a slog, the idea is shaped wrong. Reshape, sketch again.

Quiet timer's first sketch was an iPhone screen showing a timer in 64-point text. No buttons, no menu. The sketch already felt finished — that was the signal.

Phase 03 · Sandbox

Can I live with it for two weeks?

The sandbox is the lab's name for the prototype-with-real-data stage. The thing has to run for me, on my devices, against my real problem, for at least two weeks.

If I stop using it after a week, the idea isn't real. If I keep coming back to it, even when it's ugly, the idea is real and the polish gap is the only thing between me and shipping.

SunnyERP ran on my own factory's data for three months before any external pilot. The first thing I shipped to the pilot was the thing I'd been using daily.

Phase 04 · Ship

Is this finished enough that I would link it from this site?

Shipping is the polish phase. It's the boring one. It's also the one that separates the experiments people pay for from the experiments people sigh at.

I ship when the rough edges are sanded — when the export button is pretty, the empty state is real, the keyboard shortcuts are consistent. Shipping isn't a release; it's a *standard*.

Focus companion stayed in beta for nine months because the complication scheduling logic wasn't calm enough to ship. Ship date moved when the data did.

Phase 05 · Live

What does maintaining this well look like?

A live experiment has its own metabolism. Bug reports, occasional emails, a steady trickle of feature requests that the lab takes seriously but rarely actions wholesale.

Living-well means: small refinements, not big rewrites. A live experiment is past the rewrite stage. If it needs a rewrite, that's a signal it should pause or sunset.

Quiet timer has shipped 14 point updates. Each one ≤ 3 changes. The product's shape hasn't moved.

Phase 06 · Decide

Pause, continue, or sunset?

Every quarter, every live experiment goes through one of three decisions: pause (work has slowed, audience small, but worth keeping the lights on), continue (live, healthy, worth the metabolism), or sunset (the idea has run its course, the page becomes an honest record).

The honesty of this is the brand. A product page that says paused is more premium than a product page pretending to be alive.

Two experiments sunset last year. Each has a page that explains the decision, links to the successor, and credits everyone who built it.

[ what we don't build ]

The honest exclusions.

Gambling, betting, sports trading.
Surveillance tooling — corporate, governmental, parental.
Dark-pattern attention apps. Engagement is not a feature.
Anything that depends on opaque AI training data.
Tools whose growth depends on users' inability to leave.

[ open source ]

What's open, what isn't, and why.

The lab's posture on open source is case-by-case. Three rules guide it.

Firmware is open. Anything that runs on hardware the user owns — Energy dashboard's ESP32 firmware, the watchOS complication scheduling pattern — ships under MIT.

Tooling is open. Build tools, design systems, dev utilities. If it makes a craftsperson's life easier, the lab doesn't hoard it.

Products stay closed. FoodNode's verification logic, SunnyERP's costing model, Quiet timer's anti-features. These are the lab's craft; opening them would only invite cheap clones that didn't earn the audience.

Every experiment page tells you which bucket it sits in.

Watch the method work in real time.

The newsletter publishes build logs as experiments move through the phases. Subscribe to see the work between the ships.

See the method applied →