Introduction

Moonbrush is a behavioral intelligence company. Most marketing systems record what people already did: a purchase, a click, a form. Moonbrush reads the behavior that comes before those actions: what people research, how they shop and why they buy. It turns that behavior into audiences and audience groups a client can act on, and judges each campaign against a result the client chose before it launched.

The platform has three parts that share one consumer graph and one model catalog.

Part	What it is	Who uses it
Gravity	The model catalog. Versioned, dated rankings from 0 to 1: buying windows, habits and shopping manner, traits, reasons, and work.	Everyone, through atlas and stage
atlas	The audience intelligence system. It matches a client’s first-party file to people and households, scores it and returns scores and audiences as packages and files.	Enterprise marketing, data and CRM teams
stage	The behavioral intelligence workspace. Search and size a market, split an audience into groups by what sets them apart, plan creative, and package campaigns.	Agencies, planners and media teams

What these docs cover

These pages explain how each part works, what goes in and comes out, how results are counted, and how data is protected. They are written for the people who review Moonbrush before a client signs: data scientists, engineers, security and privacy teams, and marketing operations.

How the platform worksThe path from the consumer graph to a counted result, in seven steps.

Core conceptsThe vocabulary the product uses, defined once.

What leaves the platformPackages, hashing, exports and suppression.

Evaluation and releaseHow models are tested before anyone can use them.

ResearchThe published science behind each claim, graded, with every source cited.

Claims registerEach scientific claim, its basis and Moonbrush's own evidence.

Conventions

• Words of record. These docs use the words the product uses. A group inside an audience is an audience group. People set aside from every export are a held-back group. The result a client wants is an Outcome. A browser that arrived through an ad link and then completed the Outcome did it from our ads.
• Dated figures. Every count names the date it was measured. Counts move with each monthly rebuild of the consumer graph and each release of the catalog. See Published figures.
• Stated limits. Where a capability is designed but not live, or live with a limit, the page says so. A page never describes a plan as if it were shipped.
• Illustrative examples. Code samples and file excerpts use invented values. Column names in examples show the shape of a file, not an internal schema.

What is not published here

Model internals, feature construction, training data composition, supplier terms and exact configuration values are not in these docs. Qualified reviewers can walk through them with our team under NDA. Request a technical session.