Study T · Sessions & memory
The one thing statelessness can't do; and the three note lines that fix it
Conversation-carried context
Study Overview
The Missing Request Class
Every prior session study used self-contained requests, so Study T built the missing class: callback steps whose required fact lives only in earlier conversation; a declared codename a later rename must use, and a standing rule later inserts must apply unprompted (validated non-derivable from every pre-step tree).
A Total Dissociation
The dissociation is total: the stateless worked-examples recipe fails ALL 160 callback steps (0/80 per model) while scoring 160/160 on the ordinary steps of the very same sessions.
The Memo Fix
The registered fix; an app-maintained "Session notes" block, a memo instead of a transcript; recovers 80/80 on both models, ties full history on every paired comparison, beats it on end-state integrity (19–20/20 vs 17–19/20), and costs 1.02× stateless against history's 2.1×.
Guidance, Final Form
Session guidance, final form: fresh view per turn, two worked examples, and a memo of what the user declared.