What the dark has caught so far. Each card is one domain's round-trips, shown as the three states the engine reads — zero-point, unobserved, observed.
This is the catch, not a grade: we record the structure the engine caught in the dark — not a score of how the answer reads. The evidence is public — read it yourself.
What we've confirmed
confirmed
The Entanglement Bond
When two graders both say “Same,” that agreement is a signal, not the boring middle. Pre-registered, confirmed across model families, replicated on a 15-pair grader matrix.
refuted, and kept
The inverse-scale reading
We predicted smaller models would show the dark structure more. They didn't. We keep the falsification.
In-house research, pre-registered — not independently replicated.
What the dark has caught so far. Each card is one domain's round-trips, shown as the three states the engine reads — zero-point, unobserved, observed.
This is the catch, not a grade: we record the structure the engine caught in the dark — not a score of how the answer reads. The evidence is public — read it yourself.
What we've confirmed
confirmed
The Entanglement Bond
When two graders both say “Same,” that agreement is a signal, not the boring middle. Pre-registered, confirmed across model families, replicated on a 15-pair grader matrix.
refuted, and kept
The inverse-scale reading
We predicted smaller models would show the dark structure more. They didn't. We keep the falsification.
In-house research, pre-registered — not independently replicated.