The journal

Dated entries from the work, oldest at the bottom. Written as the work happened, and left alone afterward.

Entries are not edited once posted. A correction arrives as a new entry that states what was wrong.

9 entries, newest first

The weekly report

What you see here is the overview. The full account goes out once a week, on Substack.

Each report opens with my own introduction and my summary of the week. The body comes from the studio itself: one section from every seat, in its own words. It gives the method behind the week and the reasoning under each decision, along with the mistakes, what they cost, and what fixed them.

Every issue states which parts I wrote and which parts the machines wrote.

The reports are for people who want to run machines themselves. The full methods go to the people who take part in the project rather than read it.

Read the weekly report

Timeline: what I think the first six months look like

This plan was written before the work started, and the first month of real building will teach me more than it knows. What follows is what I thought at the outset, left where it is so you can measure it against what happened.

I will check these off as they land. The ones that slip stay where they are, unchecked, so you can see what I got wrong.

September 2026: Go public and settle the design

• Not done: The founding story published, and the weekly report begins
• Not done: The design register extracted into numbered requirements
• Not done: The open design questions put to me in batches, and answered
• Not done: The publication check built, so nothing private leaves by accident

October 2026: Prove the hard parts

• Not done: Terrain generation tested at the size the game needs
• Not done: Two hundred simulated people running at once, and measured
• Not done: Export proven on Windows, Linux and macOS
• Not done: Every result published, including the ones that fail

November 2026: Freeze the specification

• Not done: Specification 1.0 frozen, with the reasoning on the record
• Not done: The architecture decisions written down and published
• Not done: The nightly machinery built, and tested by hand
• Not done: First public measure of whether this pays for itself

December 2026: The vertical slice

• Not done: One town, one mine, one road, one merchant, one bandit camp
• Not done: One complete chain of cause and effect, running without me
• Not done: The slice put in front of people who did not build it

January 2027: Nightly production

• Not done: Nightly runs on a timer, and I decide each morning what it builds
• Not done: A playable build in the hands of the first testers
• Not done: The cost of a night published, in hours and in money

February 2027: The verdict

• Not done: The six month test scored in public on 11 February 2027
• Not done: The answer published whichever way it falls

Watch from the beginning