The Learning and Memory Lab
The model will change. The standard of judgment must survive it.
Cognition keeps the judgment you already earned inside the agents you already use, so the next session starts from the correction, the decision, and the standard, not a blank chat.
We let people off this list slowly. Jo approves every release before it reaches the product.
Loading
Release · September 16, 2026 · All news
Four defects found by reading one day's receipts against what happened: host markup was filed as judgment, a tie-breaker fired on every call, a retrieval miss counted as a skill failure, and a demonstrated win credited zero.
curl -fsSL https://www.lamlab.ai/api/install/claude-upgrade | shUpdate once so the marketing banner, terminal announcement, and stub version all say 1.6.94: the same story your agent already sees from /api/update-announcement.