THE METHOD, as an operating runbook (as of 30 Sep 2026)

THE ONE TRUE SYSTEM

• Canon is D:\general-practices\methods\THE-METHOD.md: the full text, with the war stories that bought each law. This runbook is the OPERATING half. When the two disagree, the doc wins and this file gets fixed.
• GRAVEYARD:
  • D:\task-4ce\docs\METHOD.md and D:\rook-worlds\rook-worlds-product\docs\THE-METHOD.md are earlier generations.
  • So is RookWorlds' own follow-the-method runbook.
  • If one of them is not yet a pointer here, do not edit it: edit the canon.
• The Notion mirror ("THE METHOD :: How Daniel and Claude ship together") is downstream by rule. Edit the canon, then refresh Notion. Never the reverse.

The two failure modes everything here fights

1. Knowledge evaporates. The reasoning lives in a chat log nobody reads again.
2. Work rots silently. A replaced system still compiles, a migration never runs, a doc describes the old version.

The answer is never "more discipline". Put knowledge where the next session TRIPS OVER it, make rot LOUD, and make failure impossible rather than remembered.

WHERE DOES THIS GO?

What you have	Its ONE home
Where we are, what is next, what bites	HANDOFF.md, capped and checked
What shipped and when	docs/CHANGELOG.md, one line per arc
How a system works	docs/concepts/<system>.md
A trap found by building	NOTES.md, appended, never rewritten
A module or architecture decision	docs/DECISIONS.md
Why, a preference, a correction	Claude memory, one fact per file
How to OPERATE something	a runbook, .claude/skills/<job>/SKILL.md
Content or data that changes	the database, never code or repo files
Something that would help EVERY project	D:\general-practices (or D:\foundation, if every new code project should start with it)

Route, then DELETE from the source. Cite the canon; never restate it.

The handoff test: would a session six weeks from now be worse off if this line sat in a concept doc behind a pointer? If no, it does not go in the handoff.

A missing destination is the real failure mode. If something has nowhere to roll off to, CREATE THE HOME FIRST.

Splitting work

A task is a whole idea; a subtask is an atomic part of it, never a session. One task = one session = one branch = one PR. Before filing something new, ask: a thing on its own, or a piece of a bigger thing? A loose end hangs off its parent task.

The loop

talk → plan (plan-a-build) → build on a branch → prove on real things (prove-it) → close gate + independent review (escape-hunt) → ship (ship-it).

Done means SEEN working. A green check alone is never done.

Writing a runbook

A system ships WITH its runbook, unprompted. Anatomy:

1. THE ONE TRUE SYSTEM + GRAVEYARD, with real symbol names
2. Entry points, verbatim
3. Knobs, each as file:line with a one-line meaning
4. Verification: which artifact, and what PASS literally says
5. Traps, each with its consequence
6. OPEN QUESTIONS

The description is the only line every session sees. Write it as "load when <situations>", with the trigger phrases in it.

The laws (short form; the scars are in the canon, §9)

• Loud fallbacks. Verify from artifacts. A control that cannot fail proves nothing.
• Test the property, not the name, and grep for the bad predicate everywhere. Name the graveyard.
• Data in the database. One authority per fact. Make it impossible, not remembered.
• Do not re-litigate a proven fix on one unverified symptom.
• Say what you did NOT do. Label a guess as a guess. Excellent over more.

Close ritual

1. Update the handoff STATE and RESUME HERE against the real git tip.
2. Add the changelog one-liners, and roll the oldest handoff entry out.
3. Check that durable knowledge reached its home. No mechanism checks this step; it is yours.
4. Write or update the runbook for anything operable.
5. Run the cap check. PASS = CAP CHECK: PASS.

Traps

• Silence from a stop-hook nudge is its PASS signal, so a broken hook and a clean session look identical. After a big session with no nudges, run the hook by hand once.
• A start hook that injects the handoff does not excuse skimming it.
• The handoff's own hashes are not evidence. Compare them against git log -1.

OPEN QUESTIONS

• Nothing yet verifies that a runbook's file:line knobs still point at real lines. In RookWorlds, one set of cites was off by 9, and only a human caught it. A cite checker is the obvious next tool.
• Close step 3 (durable knowledge reached its home) has no signal at all.