# `out/rooms/` (room notes)

One note per room: the analyst's prior, dated.

## What it holds

A1 fixes the form, under 350 words: a header (`# r/<sub>`, read date, coded share, off-platform pointers), then `doing`, `week`, `named`, `goal`, `unsettled`, `coder unsure` (what the coder said she could not place, with thread ids), `carrying` (three concept ids), `not this`, `ended`, `struck by` (the one free line, never cited), and `made`, written last from the canvas. A later stage appends one dated `later:` line. STANDING §2 also places D2's `model-D.md`, `verdict-D` and placements in `rooms/<sub>/`.

## Who writes it, and when

The analyst at A1, one note per room of the analysed set, in the order of `out/order-A1.txt` (`mem.py next` names the next room), written "before you open the next room" and logged. The harness refuses a note whose room's concepts were not read (`mem.py codes --sub`), and checks the note's form the moment it is written.

## Why it exists

STANDING §1: "Your room notes are your prior"; they are "queried (`mem.py note <sub>`), never read whole as a list", "never a source for a quote", and each later stage says what it found that they did not predict. A1: "a room you read and did not write about is a room you have not read."

## Where

[out/rooms/](../../../aggregator/runs/claude-v43-1/00_GIVE-THE-AGENT/out/rooms/).

Checked against: stages/STAGE-A1-reading-room-notes.md, STANDING-INSTRUCTIONS.md §1 and §2, tools/mem.py.
