Methodology
How the graph is built.
The Signal Graph is one deduped map of behaviour, resolved onto people and companies and re-measured every day. Every signal in it is a plain-English definition with three numbers attached, and every number is measured rather than asserted. This page is the whole method.
- people
- 287.9M
- businesses
- 64.9M
Four steps
In the order the data moves. The rail on /explore is this list, shortened.
01collect
Public exhaust, licensed panels, your own feeds
Filings, permits, job postings, app and web panels land as raw events, each carrying its source and its timestamp. Nothing enters the graph anonymous about where it came from, because a signal you cannot trace is a signal you cannot defend.
02resolve
Events resolve to entities
One person, one company, one row, deduped across every source. Each entity carries an identity Watt will defend, and the joins that built it stay inspectable rather than collapsing into a score.
03define
A signal is a definition, not a keyword list
Every signal is written in plain English first, then scored. A definition can be argued with; a keyword list can only be tuned. A signal nobody has measured says so rather than defaulting to zero.
04re-measure
Every score, every day
Accuracy is checked against holdouts rather than asserted, and freshness decays on its own between measurements. When a source retires, its contribution leaves the history too instead of standing in old numbers.
Recently published
Making the world’s signals consumable.
· 2 min
Signal Drop 01
What landed in the first Signal Drop: education signals on 118M+ people, geo as a composable signal, better identity resolution, and 1,700+ intent topics.
· 5 min
What is a Signal?
Stop thinking in fields. Start thinking in signals. What a signal actually is, what makes it good, and why one signal is almost never the answer.
· 5 min
Where our signals come from
Two vendors, thousands of upstream sources. The answer to where Watt's signals come from, and why no one else has built this.
· 5 min
What is Signal Engineering?
The operating model that replaces the legacy 5-role data chain with one operator composing against a reasoning-time substrate.