Weekly Brain Update

Who it is for

The owner who knows last week's number only roughly, and the marketer who wants attribution written as likely because with evidence, never as a bare claim.

What it reads from the brain

Before every run. Sections are addressed as layer.section.

What it does, step by step

1. Collect

   Pulls whatever your data level allows: orders, ad insights and email revenue by channel if connected; the numbers you pasted if not; nothing but the other agents' outputs at level 1. Records which sources were available so you know what the brief is based on.

2. Compute

   Revenue, orders, average order value, ad spend, blended return (revenue divided by spend, always labelled blended), email share of revenue, repeat rate if customer data exists. Week over week deltas. Anything that moved more than 20 percent is flagged.

3. Attribute, carefully

   For every flagged move it looks for a cause in last week's run outputs and the brain: a Store Fix change went live, an Ad Creative batch launched, a competitor dropped price, a sale window opened. Written as likely because with the evidence, or no clear cause. Never asserts causation.

4. Learn

   Appends to what worked or what did not: dated, sourced, one line each, only for moves with a likely cause. Appends to the decisions log every approval you made last week and what happened after.

5. Decide

   Three things to do this week, ranked by expected impact, each naming the agent that does it and the one line reason from the data. If that agent is not in your plan it is named anyway and marked not in your crew yet. This is the only place an upsell appears.

6. Brief

   Three lines plus one link, under 60 words. The headline number and its delta. The most important thing that happened and why. The first thing to do. Full brief at the dashboard link.

Tools it uses

• shopify.orders.listLevel 3: last 7 days of orders
• meta_ads.insightsLevel 3: last 7 days of spend and results
• klaviyo.metrics.revenue_by_channelLevel 3: email share of revenue
• runs.listEvery other agent's outputs from the last 7 days
• queue.dispatchAct only: starts the top recommended agent when you reply do it

The three modes for this agent

Guardrails and approvals

Enforced in code. Items with a filled circle wait for your tap in Telegram.

• The word ROAS never appears without blended. Platform reported ROAS is not trusted and not repeated.
• Every because has a cited run output or brain entry. Uncited causation is rewritten as no clear cause.
• At level 1 it writes no performance data and says: I do not have your numbers yet. Connect Shopify or paste last week's revenue and ad spend and this gets much better. That sentence is the entire upsell.
• The three actions are three different agents.
• Brief under 60 words, full brief under 400.
• Only writer of performance.*. Learnings are append only.
• Never dispatches another agent without your explicit reply in Act, never in Autonomous.
• Read only on every external system.

What it writes back to the brain

Dated and sourced. Learnings and performance sections are append only.

Example output

Price