Notes from the lab
What 0.821 recall@5 means, and what it doesn't
Anjo's memory scores 0.821 recall@5 on LongMemEval-S. Here is how we ran it, which controls we checked, and the claims the number does not support.
Why we filter 'last time you said' out of Anjo's replies
A language model can fabricate a shared memory in one sentence. Anjo removes unsupported ones before display. How the guard works, and where it fails.
Measuring whether conversations leave you better, against our own engagement
Anjo tracks whether a conversation helped, not how long it ran. The wellbeing trend, the outcome ledger, and the rule that self-report alone moves nothing.