June closes the first chapter. January's proposition was "define the screen on the server"; June's is "the person who knows what to build stands in front." The two are the front and back of one story.
What went out
| Date | Track | |
|---|---|---|
| 06-04 | Horizon | The Age of Domain Experts |
| 06-11 | Build | All of It on One Runtime |
| 06-18 | Field | One Person, the Work of Many Hands |
| 06-20 | Field | The Accountant Who Never Misses a Deadline |
Counted whether it is one thing
Attach a device a month for half a year and the good stories accumulate, but you do not learn whether it is one thing or several that resemble each other. So June built no new device and counted what was there.
Comparing version numbers is not evidence — a version is a number typed by hand and copies that drifted while keeping it are ordinary. So the bytes of the three files that actually draw and hold a screen were compared.
27 copies of the runtime · renderer, widget_factory and state_manager
identical byte for byte · 0.5.1Not one line apart. And the devices behind them differ across three kinds: real boards, C simulators and Dart servers.
What the first half-year proved
"This is what such a tool looks like, this is how short it is, and this is when it can be trusted." No more than that.
Still open
- Nothing has run for long. Each sample's log is one or two seconds. What leaks after weeks is unknown
- Field scale — one screen holding five nodes and one holding five hundred are different problems
- A model's tool-selection accuracy — the model in the LLM piece was a deterministic stub
- Whether readers write it themselves — that 45 lines are readable is what was confirmed; how long a newcomer takes needs a newcomer
Next chapter
From July the subject is the order you build things. From raising one server through serving screens, permission, duplicates, retries and subscriptions, in one line.
makemind.dev "Signal" — the 2026-06 close. The end of the first half-year.