Five Notebooks. No Handbook.

Multi-bot contradiction is usually a missing shared floor — not a broken model.

If you already run more than one helper, you know this busywork.

You taught the refund rule last month. You taught it again last week. Today bot A says fourteen days, bot B says thirty, and bot C never heard of a refund rule at all.

Nothing wild happened. They each have a notebook. Nobody gave them a handbook.

Five notebooks. No handbook.

That is a process failure, not a model failure. Borrow the pattern from @GrokBotRadar (source ()): give the team one shared canon, then teach every seat where each kind of fact lives. Shared floor for the house. Private memory for the specialist. Live sources for anything that can change. Do not merge every bot into one giant brain.

Earlier this week we talked about handoffs — Your Bots Aren’t Broken. Your Handoffs Are. (). Seams fail when work crosses seats with fuzzy owners and fuzzy “done. This piece is the floor under those seams. When each seat invents its own notebook, the handoff was already broken before the baton moved.

Why a “smarter model” doesnt fix this

If you run a small shop, you feel it when the inbox helper and the calendar helper disagree on who may send. If you’re flying solo, you feel it when the research bot and the writer bot cite two different house rules. At home, you feel it when one helper remembers a bedtime and another still uses last year’s version.

You re-explain. They take the note. Next Tuesday the contradiction is back — because the truth never had a home. It lived in whichever chat was open last.

Merging everything into one giant brain sounds neat. In practice it muddies who owns what, burns time re-reading piles nobody needed, and still leaves changing product facts stale. The fix is a map, not a merge.

One shared canon — then seven homes

Write the house rules once. Then put every fact in the right drawer:

1. Hard rules → bot description + house pack

2. Anti-jobs, never-send-without-approve, kill phrases, who may touch money. These live where every seat can read them on day one — not buried in a forgotten chat.

3. Shared facts → a house folder

Hours, addresses, brand voice pointers, “how we name jobs.” One place the whole team looks. On a Grok Bot computer that often looks like a shared house directory everyone can open not five private copies.

3. Role preferences → that bot’s memory

The writer likes short paragraphs. The inbox clerk parks deletes. Specialist taste stays with the specialist. Do not dump every preference into the shared floor.

4. Repeatable method → a Skill

How you always write a refusal log, how you always stage a draft, how you always check a handoff. Method is a recipe, not a vibe.

5. When it runs → a Routine

Clocks, listeners, “every weekday at nine.” Timing is not a memory note. Timing is a schedule.

6. Todays work → this chat + a work folder

The live task, the open file, the draft in front of you. Don’t promote todays mess into permanent canon.

7. Changing product facts → the live source

Prices, plan tiers, inventory, API status. Point at the source of truth. Do not freeze a screenshot in memory and call it forever.

Shared canon for the team. Private memory for the specialist. Live sources for anything that moves. Teach them where to look.

Everyday example: the refund argument

Without a handbook

You tell the inbox helper in chat: “Refunds are fourteen days.” Next month you tell the research helper: “We moved to thirty.” Neither update hits a shared house file. A customer gets two answers in one afternoon. You spend the evening re-teaching both bots and you still dont know which notebook wins tomorrow.

With a handbook + map

Hard rule lives in the house pack and each bot’s description: “Refund window = see house fact; never invent a number. Shared fact lives in the house folder: refund-window-days: 30 with a dated note. Inbox and research both read that file. Role memory only holds tone (be plain with customers”). The method for answering policy questions is a Skill: look up house first, then draft. If the window changes again, you edit the house fact once — you don’t stack a third chat memory on top of two old versions.

Same bots. Different floor.

Sister habits that sit next to this

Handoffs need a clear owner and a defined “done.” They also need a shared floor so the receiving seat isn’t inventing policy mid-pass.

And when a seat parks or refuses — Memory isn’t proof. The refusal log is. () — the receipt still matters. A shared handbook tells them what the fence is. The refusal log proves the fence fired. Canon without receipts still leaves you guessing. Receipts without canon is five notebooks again.

Where Grok Bot fits

Week-one setup is not “add six helpers and hope they sync on their own.”

It is: name the jobs, write the hard rules into descriptions and a house pack, put shared facts in one folder, keep role taste in each seat’s memory, turn methods into Skills, put clocks on Routines, keep today’s pile in chat/work, and point changing facts at live sources. Grok Bot is a solid home for that map — named helpers, clear anti-jobs, shared computer files, skills and routines you can actually open. Soft start when it fits: get on Grok Bot from about $20/mo (confirm live plan pricing before you treat that number as fixed).

Setup over spectacle. One handbook. Seven homes. No giant brain.

Imitate the map. Iterate until contradictions point at a missing home, not a “dumb model.” Innovate only after every seat knows where to look.

One line to keep

Five notebooks. No handbook.

That’s the bug. Shared canon plus a map of homes is the fix — teach them where to look; don’t merge the brains.

Come join me on the journey.

https://sideshopai.com

Your side shop for AI that actually helps the business run.