Skip to content

Definitions you can argue with

Why every metric has an owner, a written meaning and a change history.

The failure this prevents

Analytics rarely fails because data is missing. It fails because two people quote "active users" and get different answers, the meeting becomes an argument about whose dashboard is right, and within a quarter somebody is keeping the real figures in a spreadsheet on their own machine. Once that spreadsheet exists, the tool has lost.

Surveys of product teams put measuring success at the top of the unresolved-problem list, ahead of prioritisation and roadmapping. The gap is almost never the query. It is that nobody wrote down what the number counts.

You can diagnose this without asking anyone. If a shadow spreadsheet exists, or if people export rather than link, the definitions are not agreed.

What a definition contains

Open Settings, then Definitions. Every metric shows what it counts in one plain sentence, who owns it, and which version of the definition you are looking at. The sentence is generated from the actual calculation rather than typed by hand, so it cannot drift away from what the query does.

FieldWhat it is for
MeaningOne sentence describing exactly what is counted, derived from the calculation itself.
OwnerThe person who answers when someone disputes the number.
VersionIncrements whenever the meaning changes. Cosmetic edits do not bump it.
ChangedWhen the meaning last moved, and who moved it.

Why the history matters

Every change to what a metric counts is recorded with the full definition as it stood, the author and the timestamp. Nothing overwrites history and nothing can be edited after the fact, including by an admin.

The reason is specific. If activation quietly changes from "created a project" to "created a project and returned", every chart drawn before that moment now means something different from every chart drawn after it, and the step change looks like a product regression. With a version stamp, the same graph is readable: the line moved because the definition did.

Renaming a metric or rewriting its description does not create a new version. Only changes to what is counted do.

Two names for one action

Signed Up, signup_completed and SignupSuccess are one event with three names, usually because three people instrumented three surfaces. Nothing errors. The funnel simply counts a third of the signups it should, and every conversion rate built on it is understated until somebody notices.

Termind normalises event names — case, separators and common tense endings — and shows you the collisions under Settings, then Definitions. Merging is offered, never automatic: sometimes one name is the server and one is the browser, and the difference is the whole point.

Merging sets an alias. Nothing is deleted and no history is rewritten: the aliased name keeps its own counts and its first-seen date, so "when did we start double-naming this" stays answerable, and the merge can be undone.

What is flaggedWhy it matters
Two names, one actionYour funnel is splitting one step across two counters.
Stopped arrivingIt fired regularly for weeks and then stopped. Usually a release removed it.
Never caught onAlmost no events under this name. Usually a typo nobody cleaned up.
NextAI and agent trafficWho read your pages, who they sent, and how much of your traffic is not a person.