Research company · Portland, Oregon
Anything genuinely complicated is too big to hold in your head at once. So you need buckets to pour it into — few enough to keep track of, in a place you can find again. There are four of them here. They are simple enough to explain to a ten-year-old, and the same four have held a chess game, a global crisis thesis, a memory system, a world generator and a city.
The buckets
Everything on this site is these four buckets pointed at something. If the buckets are the wrong shape, all of it is wrong together — which is exactly why they are worth writing down this plainly.
Ask anyone to hold nine different things in mind at once and two of them quietly become the same thing. Seven is roughly where a list stops being a list and starts being a blur.
So nothing here is ever divided into more than seven pieces at one level. If you need more, you have not found the right seven yet — or you need another level.
Every piece gets a fixed address — a location you go to, not a label you search for. Two things sitting next to each other are genuinely next to each other.
An address written down two years ago still points at the same spot today. That is what makes it possible to check an old decision instead of taking someone's word about it.
Almost every system grows by adding room on the outside — and everything already inside shifts to make space. Addresses stop meaning what they meant.
This one opens a room inside a room. The outside never changes shape, and nothing that already had a place loses it. You can keep going down forever.
Every level looks exactly like every other level. Not similar — identical in structure, with its own seven parts and its own addresses.
Write the rules for reading, scoring and checking once, and they work at any depth, forever. There is only ever one kind of thing to handle.
Forty-nine squares. Forty-eight of them hold something. The amber one in the corner is a door.
Click the door and you fall inside that square — and find another forty-nine waiting, exactly the same. You can keep going as long as you like. The grid you left did not move, did not resize, and did not renumber. Everything you already put somewhere is still exactly where you put it.
Watch the address at the top get longer as you descend. That string is the whole location, forever — and you could write it on a napkin today and still find the same square in ten years.
Shown flat so you can click it. The real one is a cube — seven by seven by seven, so 342 places per level instead of forty-eight, with the door at (6,6,6). The addresses work exactly the same way, just with a third number.
1 × 3 × 7 when the thing is simple. 3 × 3 × 7 for most things. Let it recurse when the thing is genuinely complex. That is the whole sizing decision. A goal breaks into three, each into three, each into seven — sixty-three places to put work. A crisis breaks into three lenses over seven domains across three scales. Same rule, three settings.
The honest part
There is no proof that seven is the right number of buckets. There is no theorem that says a cube must be seven wide. Seven came from a chess variant that needed a true centre — an eight-wide board has no middle square, and a seven-wide one does. It stayed because every time it was tested against a new problem it kept being enough and not too much.
We could dress that up as derivation. It would be more impressive and it would be false. These are engineering choices that survived contact with real problems, which is a weaker claim than proof and a much more honest one.
It is also a falsifiable one. If a domain genuinely needs nine irreducible parts, the first bucket is the wrong shape and we would rather know. The Arc Lab exists for exactly that — it measures whether a chosen set of parts is actually independent, and it will report your basis as bad if it is.
A number you chose can be checked. A number you claim to have derived invites everyone to audit the derivation instead of the results — and if the derivation is thin, the results go down with it.
Stating the choice openly puts the weight where it belongs: not on whether seven is metaphysically correct, but on whether a system built this way works.
The part that counts
Not "they are elegant." Not "they feel right." The same four buckets were used by the same person on five unrelated problems over two years, and in each case the structure held and can be checked by someone else. Here is what that looks like when you count it.
| A chess variant | Needed a centre square, so the board went from eight wide to seven.
The centre is still in the code as ORIGIN_COORD = [3,3,3]. This is where the
rules came from, not where they were applied. |
|---|---|
| A crisis thesis | Three lenses — what is true, what people feel, what we owe each other — over seven crisis domains, across three scales. Sixty-three places to put an analysis. Written September 2024. |
| A memory system | 342 places per level and one reserved door that opens a whole new level beneath. Every address tested individually in both directions — 741 assertions, exhaustive rather than sampled, because an off-by-one in an odometer is silent. |
| A world generator | Three galaxies, three systems each, seven worlds each. 63 of 63 addresses generated and sitting in a directory tree anyone can count. |
| A city simulator | The same decomposition pointed at terrain, roads, emergency services and utilities. Days old, builds clean, and — stated plainly — has no tests yet. |
One person applying one rule five times is consistency, not independent replication, and we are not going to inflate it. What makes it worth your attention is that every row above is checkable without asking us: the constants are literals in source, asserted by tests, or countable as files on a disk. The evidence page publishes the failures alongside the passes, including the run where our own harness wrongly reported us broken.
Two halves of the same coin
The rules were written down before they were code — which means the code can be checked against them instead of rationalised backwards into them. Read either side first.
Where the rules were worked out.
A crisis-scoring thesis begun September 2024, a 228-page manual on urban resilience, a fractal template and a weighting formula that scales by an order of magnitude per tier. None of it peer reviewed, none of it offered as a credential.
Read the research →Where they had to survive.
A running cognition engine, an addressable memory lattice, a laboratory that tests whether a chosen basis is any good, a certified-inference module, and a testbed built to falsify the whole thesis.
Read the architecture →Go deeper
The four rules stated as engineering: what recursive, holographic and superimposed actually mean for a running system, and the honest trade that growing inward makes.
342 places per level, six faces to a cell, decay and consolidation over time, and a door at (6,6,6) that opens a whole new level underneath.
Three by three by seven. How a goal too large to act on becomes sixty-three pieces small enough to hand to a tool.
The instrument that tests the first bucket. Bring your own seven — it measures whether they are actually independent, and tells you when they are not.
Addresses for generated content, with the ranges enforced at the boundary. 63 of 63 generated and countable.
The chess variant that needed a true centre. First project, and the reason everything else is seven wide.
Standing apart
The one thing here that shares none of the above. It makes an expert's sign-off on an AI system stop expiring silently: a model tag is re-pulled, a driver updates, and the system answers differently while every health check stays green.
It wraps a model you already run, has no dependencies, and requires you to adopt none of the other ideas on this site. That independence is deliberate.
“We aren't teaching AI to think. We're giving it a place to stand.” G-ynthetic Labs · Portland, Oregon