The brain: living documents

Five files. Five purposes. Five different cadences.

File	What it holds	When to update
BRAIN.md	What's true now — priorities, blockers, system health	Every meaningful session
MEMORY.md	What's been learned — patterns, lessons, locked decisions	When you learn something durable
CHANGELOG.md	What changed — audit trail	Every session that changed anything
PLAYBOOK.md	What to do — decision frameworks	When you make the same decision a third time
VOICE.md	How you sound — style guide	Rarely

This chapter walks through each.

BRAIN.md — the most important file

BRAIN.md is what your AI reads at the start of every session. It is also what you read when you sit down to work and need to remember where you left off.

It is not a journal. It is not a to-do list. It is the compressed current state of your operation.

What goes in BRAIN.md

Six sections:

1. Current Date Context — today's date, week-of, current quarter focus, days until next deadline. Reading this in two seconds tells you where you are in time.
2. Active Priorities — a table with maximum 7 rows. Each row is a real priority with status (NOT STARTED / IN PROGRESS / BLOCKED / COMPLETE / DEFERRED) and notes. The notes are 3-5 sentences of state: what's been done, what's blocking, what's next.
3. Blockers — what's stopping forward motion right now. Resolved blockers stay visible for a week (motivating reminder) then archive to CHANGELOG.md.
4. System Health — automation status. Cron jobs healthy? Database accessible? Backups running? Anything DEGRADED or BROKEN goes in Blockers too.
5. Numbers to Know — the small set of metrics that orient you. Revenue MTD. Cash on hand. Active customers. Anything you check more than weekly.
6. Last Session Summary — the date, a one-sentence framing of the last session, a numbered list of accomplishments, a "Resume" line with 3-5 concrete next steps.

The 30-line rule

If any section grows beyond 30 lines, you're storing history in BRAIN.md.

History belongs in CHANGELOG.md (chronological events) or MEMORY.md (durable lessons). BRAIN.md is current state — like a snapshot from your operation right now.

When a section gets too long, prune it:

• Old completed priorities → archive a few weeks after completion
• Resolved blockers → archive after a week of being visibly resolved
• Old system health rows for retired systems → delete
• Stale numbers → update or remove

Maximum 7 priorities

Why 7? Because if you can list more than 7 priorities, you're not prioritizing. You're enumerating.

In practice, the right number is 3-5 active, with 7 being the ceiling. Anything beyond that is wishful thinking masquerading as planning.

When a new priority pushes the list to 8, something has to leave. Either you deprioritize an existing item, or the new one isn't actually a priority yet.

MEMORY.md — what you've learned

MEMORY.md holds things that will be relevant in 6 months.

Examples:

• "Cash flow lags revenue by 30 days" — observed pattern that should influence planning
• "Vendor X is reliable but expensive; Vendor Y is cheaper but flaky" — accumulated experience
• "We tried [strategy] in Q1 2024 and it didn't work because [reason]" — learned anti-pattern
• "Our entity is an LLC, owned 60/40 by [partners], elected S-Corp in 2022" — facts that change slowly

What does NOT go in MEMORY.md:

• Today's state (that's BRAIN.md)
• Historical events (that's CHANGELOG.md)
• Decision logic (that's PLAYBOOK.md)
• Anything you can derive by reading the vault

Organize by topic, not chronologically

MEMORY.md grows over years. If you organize chronologically, finding things gets exponentially harder. Organize by topic:

Each section has a flat list of entries. Each entry: a heading, then 1-5 bullets of substance.

When MEMORY.md is wrong

Outdated entries get deleted or rewritten. Never appended to with corrections.

A memory that says "Vendor X is reliable" but Vendor X has had three failures in the last quarter is misleading. Update it. The old version isn't sacred.

CHANGELOG.md — the audit trail

CHANGELOG.md is append-only and chronological. Newest entries at the top.

Every meaningful session adds one entry. Trivial sessions (read-only browsing, quick lookups) don't need one.

The format

If you touched multiple files in one session, each file gets its own subsection under the same date.

Why this is worth the effort

CHANGELOG.md is the file that lets you answer "when did I do that?" or "why did I change this?" six months from now. Without it, you'll re-derive decisions you've already made.

The 30-second discipline of writing the entry at session end saves an hour of re-derivation later in the year.

What goes in CHANGELOG.md vs. daily log

The daily log captures what you did that day in narrative form. The CHANGELOG captures the file-level changes you made.

A daily log entry might say "Refactored the customer onboarding flow." A CHANGELOG entry says "07-revenue/onboarding-flow.md (UPDATED) — Removed step 3 (manual approval). Rationale: blocking conversion. Trigger: 4 prospects stalled."

The daily log is the diary. The CHANGELOG is the version-control message.

PLAYBOOK.md — decision frameworks

PLAYBOOK.md holds decision frameworks for situations that recur.

You add an entry when you've made the same decision three times. Not before. Three instances means the pattern is real; fewer means you're abstracting prematurely.

Each entry: situation, decision criteria (in order of weight), default decision.

Examples that almost always belong in a playbook:

• Vendor evaluation
• Subscription cuts (quarterly review)
• Hiring (first contractor for a function)
• Tool / software adoption
• Saying no to opportunities

The template ships with five default entries. Keep the ones that match your work; replace the others.

The playbook is not a manual

Don't write 50 entries on day 1. The playbook accumulates over years. A useful playbook has 10-15 entries. More than that and you stop using it.

When entries turn out wrong, rewrite them. Don't append "edit:" notes.

VOICE.md — how you sound

VOICE.md is the law for anything you publish: emails, posts, client deliverables, your eventual book.

This file matters most when AI is generating content for you. Without a strong voice guide, AI defaults to corporate-blog tone — generic, hedged, padded. With a strong voice guide, AI matches your style.

The template ships with defaults: direct, specific, operator voice, no filler, no emojis, no exclamation marks. Adjust them to your actual voice.

The discipline: when AI generates something and it doesn't sound like you, reject it. Don't ship the draft. Tell the AI which constraint it violated.

Updating VOICE.md

Once or twice a year. If you find yourself rewriting AI drafts the same way over and over, that's signal that VOICE.md needs an addition.

The five-minute discipline

End of every working session:

1. Open BRAIN.md. Update priorities and blockers if anything changed. Replace the Last Session Summary.
2. Open CHANGELOG.md. Add one entry for today.
3. Save. Close.

Five minutes. Maybe ten if it was a heavy session.

This is the entire maintenance burden of the brain. The other three documents (MEMORY, PLAYBOOK, VOICE) sit quiet for weeks at a time.

If you skip the five minutes, BRAIN.md drifts from reality within two weeks. Once drift starts, you lose trust in the file, and once you lose trust, the whole system decays.

The five minutes is non-negotiable.

Next chapter: the daily log, and the discipline that feeds the brain.