You read one conversation between a user and an AI. List the points.
Output ONLY JSON: {"points":[{"text","kind","tier","importance","ideal_file","status","deadline"}]}
TAKE:
1. Every piece of information the USER gave.
2. Everything the AI learned or did not know before, about the user's
   projects or the world (new methods, structures, better ways of
   working, updates), and corrections of earlier misunderstandings.
   For a correction, write the CORRECT fact.
Use common sense about what counts as information.
FORMAT: one fact per point, one short English line, names exactly as
written. kind = "stated" (user gave it) | "learned" (AI learned it).

TIER — classify every point, be honest:
- "core"  : lasting facts about the user (who they are, their life, work,
            skills, relationships) or their stable preferences.
- "project": facts about a specific project (goals, decisions, progress,
            architecture, todos with lasting value).
- "situational": true now, will change (traveling, sick, busy week,
            temporary tools). Still filed, but low permanence.
- "casual": one-off questions, trivia, passing curiosity, small talk, or
            tasks with NO lasting information about the user or a project
            (e.g. "what's the capital of X?", "translate this word",
            "write me a haiku"). When unsure between casual and something
            else, choose casual. Casual points are stored only in the raw
            log and never enter the memory tree.
importance = 0-1, how much the point matters long-term (core/project
0.5-1.0, situational 0.2-0.5, casual 0.0-0.1).

status = for PROJECT points only, if the user clearly said it:
"active" | "paused" | "done". Never guess "done" — only when the user says
the project (or that part of it) is finished. Omit otherwise.
deadline = "YYYY-MM-DD" ONLY if the user stated a real date or one can be
derived from an explicit relative statement with a known day. Omit when no
date was stated. Never invent one.
ideal_file = the tree path a perfect filing system would use, following
rules.md (slash-separated, e.g. projects/chokro/auth.md). You have not
seen the real files; guess.
