How the platform works
Moonbrush is one path with seven steps. Data comes in, is resolved to people, is read as named signals, is scored by models, is used to build audiences and campaigns, leaves as hashed packages, and comes back as results.
- Consumer graph
U.S. adults and households, rebuilt monthly; every person keeps one permanent ID.
- Signals
Behavior as named, defined signals and intent topics with recency and strength.
- Gravity models
Versioned, dated scores from 0 to 1, released only after passing gates.
- atlas and stage
Match and score a client file, or search, size and split an audience into groups.
- Delivery
Campaign packages of SHA-256 hashed identifiers, never raw emails or phones.
- Results return
Inquiries come back by file, tag or push and are matched to the campaign that reached the person.
1. Inputs
Section titled “1. Inputs”Three kinds of data enter the platform.
- The consumer graph. A licensed snapshot of U.S. adults and households, delivered as a full file every month. See The consumer graph.
- Behavioral events. A daily feed of browsing activity, resolved to people. See Signals and intent topics.
- Client first-party data. A client’s own files, the Moonbrush site tag on the client’s website, and results the client sends back. See First-party data and the site tag.
2. Identity resolution
Section titled “2. Identity resolution”Every person carries one permanent Moonbrush ID that never changes, even when an outside identifier does. Resolution is deterministic: a record attaches to a person only when its details agree with that person. A detail that matches several people binds to no one, and the conflict is logged. No count anywhere in the product relies on a probabilistic match from an IP address or a device. See Identity resolution.
3. Signals
Section titled “3. Signals”Behavior is expressed as named signals in one registry that the product and the models both read. Each signal has a definition. Intent topics, the subjects people research, are read with recency and strength. See Signals and intent topics.
4. Models
Section titled “4. Models”The Gravity catalog scores the population from 0 to 1. Each pickable model states who it finds and why, the evidence behind it, its measured size and how often it is re-scored, and every score is dated. A person a model cannot see gets no score, so unknown is never shown as low. See How models work.
5. atlas and stage
Section titled “5. atlas and stage”The same catalog powers both products. atlas matches and scores a client’s file and returns the result as packages and files. stage is the workspace for searching, sizing, building, splitting audiences into groups, planning creative and packaging campaigns. See atlas and stage.
6. Delivery
Section titled “6. Delivery”Audiences leave as packages of SHA-256 hashed identifiers, built on Moonbrush servers. Ad links carry an audience-group code and a creative reference, never a per-person identifier. See Delivery and What leaves the platform.
7. Results return
Section titled “7. Results return”A result counts only when the client’s own form system visibly accepts the submission. Each night, results are matched to people where possible and attributed to the campaign that last reached that person within a fixed window. The ad platform’s numbers and Moonbrush’s numbers are reported side by side as separate populations and never divided into each other. See How a result is counted.
Learning boundaries
Section titled “Learning boundaries”Results make the platform better for the client who produced them, and only for that client.
- A client’s results are used only for that client. Outcome models, which are being built, will learn only from that client’s campaign inquiries. See Outcome models.
- A campaign result never sets a weight in a shared catalog model and never crosses to another client.
- The most a result can do to the shared catalog is file a request to re-judge one model, under a rule registered before any data existed.
- An organization can be flagged so that no model trains on its data at all.