A record of changing our mind.
A common worry about any framework like this is that its authors will reinterpret everything to stay right. So we keep this page from the start: a plain, running record of what we’re learning and refining as we go — and we expect to keep adjusting it as the evidence comes in.
Most of what’s below is a design-stage correction — a change we made while building and reviewing the framework, before any large-scale data — and a couple are open questions we’re holding until the data settles them. The bigger changes the data will force — archetypes merging, relocating, or being dropped — will be added here as they come, including the ones that turn out to vindicate a call we made. That’s the point of the page: the corrections and the open questions on record next to the framework, not buried.
- April 2026World
Separated the Arena back out of the Valley.
We had folded these two worlds together for a practical reason: both were expected to be relatively rare, so merging them looked tidy on the numbers. In practice it muddied the questions too much and didn't work — a sovereignty-and-power level and a kinship-and-belonging one sit on opposite sides of a self-versus-group divide, and collapsing them lost a real distinction. It also cost the secondary read, which matters more than it sounds: holding one of them as a secondary world is fairly common, and the merge made that impossible. We restored the Arena as its own world.
- May 2026What
Restored the Bear as its own archetype.
We had merged the Bear into the Dolphin on the strength of one type being hard to classify in isolation. Looking at the trait data directly, the two are about as far apart on extraversion and openness as other pairs we keep separate. The original merge was over-justified; we reversed it.
- June 2026What
Split the double-barrelled questions apart.
An honest one: we're a software team who care a lot about psychology but aren't formally trained in it, and double-barrelled items are exactly the kind of mistake that catches people like us. Reader feedback flagged questions that quietly asked two things at once — perceive something, then act on it — which someone could honestly answer yes to one half and no to the other. We took five of them apart so each asks a single thing. Generative writing doesn't catch this; an adversarial read, asking 'could someone agree with one clause but not the other?', does.
- June 2026What
Removed items that pulled in two directions.
Some questions loaded onto more than one archetype, blurring the result. That was a logical slip on our part: the archetypes can feel similar to each other, but the whole point of the test is clean, specific measurement of one motivation at a time — an item that points at two defeats it. We cut them. The AI reviews credited it as the right structural call: a thin, clean signal beats a fuller, muddier one.
- June 2026What
A hypothesis about what separates the closest pairs.
We had been separating the most similar archetypes — the Stag and the Wolf, for instance — along a material-versus-personal line. We now think the cleaner divide may be individual-versus-collective: the Stag holds to the shared order, the Wolf stands by its own people, and the same axis would separate two other fragile pairs. It's a hypothesis, not a finding — the data has to confirm it before we'd claim it, and it's one of the things the validation plan sets out to test.
- A test we want to runWhat
The Beaver — following the evidence.
We added the Beaver — the builder — to name a real institution-architect motivation. Measured on personality facets alone, it might dissolve into a blend of others, and we understand that. But the question we actually want the data to answer is whether an archetype that feels like a real, recognisable person stands on its own — and we're glad to follow the evidence wherever it lands. The Beaver stays, with the test for keeping or folding it set out in advance.