How Moonbrush models behavior, builds audiences and counts what came through.

1. 01Consumer graph

   U.S. adults and households, rebuilt monthly; every person keeps one permanent ID.

2. 02Signals

   Behavior as named, defined signals and intent topics with recency and strength.

3. 03Gravity models

   Versioned, dated scores from 0 to 1, released only after passing gates.

4. 04atlas and stage

   Match and score a client file, or search, size and split an audience into groups.

5. 05Delivery

   Campaign packages of SHA-256 hashed identifiers, never raw emails or phones.

6. 06Results return

   Inquiries come back by file, tag or push and are matched to the campaign that reached the person.

The loopA client's results are used only for that client. They never change the shared models and never cross to another client.

268.7M

U.S. adults in the consumer graph, resolved to people and householdsAs of October 19, 2026

62.6B

resolved behavioral events, read on a daily feedAs of July 31, 2026

10,312

intent topics people research, in one topic treeAs of August 3, 2026

26

categories in the intent topic treeAs of August 3, 2026

How each figure is defined and measured

Products

• For enterprise teams. Match a first-party file to people and households, score it with Gravity, and take scores and audiences away as packages and files.Read the atlas docs
• stageFor agencies and planners. Search and size a market, split an audience into groups by what sets them apart, plan creative and package campaigns.Read the stage docs
• GravityThe model catalog behind both. Versioned, dated scores from 0 to 1, each with a stated purpose, its evidence, measured size and refresh cadence.Read the Gravity docs

Reading paths

Pick the path that matches the review you are doing. Each is four pages, in order.

Data science

How signals, models and evaluation work, and what a score does and does not claim.

1. The consumer graph
2. How models work
3. Evaluation and release
4. Held-back groups

Security and privacy

What enters, what leaves, how clients are separated and who can see what.

1. What leaves the platform
2. Architecture and isolation
3. Access and roles
4. Privacy and permitted use

Marketing operations

Files in, packages out, and how results come back and get counted.

1. Matching
2. Package and file formats
3. Campaigns and packages
4. How a result is counted

Ground rules for these pages

• Dated, not timeless. Figures carry the date they were measured. When a behavior changes, the changelog records it.
• No overclaiming. A result is called response, never lift, and no page says a campaign caused a result. A model is described by what it was evaluated on, a held-out test set, and by the evidence behind it.
• What we leave out. Model internals, feature recipes, training data composition and exact configuration values are not published. We walk qualified reviewers through them under NDA. Ask for a technical session.