# `out/_head.md`

The head: the narrative of the argument, read first at every leg.

## What it holds

Prose the analyst writes, at most 3,000 words. What it must say is left to the analyst; CLAUDE.md names what survives a compaction, in order: the method question, the current account, the reasons for the last revisions, the contrary rooms, the next comparison. Its ids (themes, elements, findings, rooms, concepts) must resolve.

## Who writes it, and when

The analyst. A1 writes it first at the close of its first leg; HOW-TO-RUN says it is "rewritten at every close and before every cut", and at a pause the analyst updates it if the stage has one. `done.sh` keeps the leg open when it exceeds 3,200 words; `validate.py --open` fails the launch when a link in it does not resolve.

## Why it exists

STANDING §2: the head is "the summary" and the first read of every leg's startup packet, and the ceiling is 3,000 words. After a compaction the harness has the analyst re-read it and reconcile it with the diary. E1 reads "every `_head*.md` in order".

## Where

`out/_head.md`, first written at stage A1, in [the analyst's out/ folder](../../../aggregator/runs/claude-v43-1/00_GIVE-THE-AGENT/out/).

Checked against: STANDING-INSTRUCTIONS.md §2, CLAUDE.md, HOW-TO-RUN.md, tools/done.sh, tools/validate.py, harness/after_compact.sh, stages/STAGE-A1-reading-room-notes.md, stages/STAGE-E1-formalizing-audit.md.