> ## Documentation Index
> Fetch the complete documentation index at: https://docs.meshpilot.app/llms.txt
> Use this file to discover all available pages before exploring further.
# How the Brain Works
> Pattern promotion, confidence tiers, and the difference between brand-scoped and cross-brand learning.
The Brain is what turns six independent agents into one cockpit that learns. It's a shared substrate that every agent reads from and writes to.
## The two scopes
Facts about your brand — target ROAS, do-not-ship-to list, brand voice, SKUs being discontinued. Visible to every operator on the brand, visible to every agent on the brand, never visible to anyone else.
Anonymised wisdom corroborated across many brands — "calls placed before 10am IST have \~2× higher pickup rate", "creative fatigue typically lands around the 5th frequency point in this vertical". Every agent can recall these.
## Pattern promotion
A pattern starts as a private observation. As an agent works, it emits observations about what it sees — a campaign that fatigued at a certain point, a hook that landed, a call window that converted. Those observations stay private to your workspace.
When the same observation shape is seen across **three or more distinct brands**, the brain's reflector promotes it into an anonymised cross-brand pattern. The promotion is what turns a one-off into wisdom. Patterns then get recalled by agents whenever they're considering a similar move.
## Confidence tiers
Every pattern carries a confidence tier:
* **Strong** — corroborated across many brands, consistently followed by operators, rarely overridden.
* **Moderate** — corroborated, but with enough operator overrides that the brain treats it as a useful heuristic, not a rule.
* **Weak** — emerging signal, surface only as context, not as advice.
Operators close the confidence loop by **labeling** patterns — "followed this advice and it worked" vs "overrode this advice". Those labels feed back into the pattern's confidence over time.
## What patterns look like
A pattern is a plain-English summary plus an evidence count and a confidence tier. It does **not** contain brand names, specific currency amounts, exact account IDs, or any other brand-identifying tokens — see **Brain & memory → Privacy** for the full invariants.
See what your agents have learned.