One definition of 'active user'

Data, ML and researchProduct analytics team after a metric argument

Marketing counted logins as activity, product didn't. After an hour-long meeting we agreed: active means at least one non-login event in 28 days. Two weeks later an agent built a dashboard with the old 7-day definition.

How do we make AI agents use our agreed definition of active user in every query?

Save the definition once in ArcRoom with arcroom decide "active user = 1+ non-login event in 28 days" and a --why reason. Recall then puts it in front of any agent writing a metric query or dashboard. If the definition changes, the new decision replaces the old one. ArcRoom tells the agent the definition, but it can't stop a query that ignores it.

Also asked as: An agent built a dashboard with our old metric definition. How do we prevent that? · How do I give AI agents one agreed metric definition? · Can an AI agent remember a metric decision our team made in a meeting?

The work

The definition is saved once with arcroom decide, with the reason and who agreed.

ArcRoom keeps
  • Decision
  • Recall
  • Wrong-memory report
  • Dashboard
The next agent gets

$ arcroom decide "active user = 1+ non-login event in 28 days" --why "logins inflate DAU by 30%; agreed by product + marketing, Oct 20"

Step by step

  1. The definition is saved once with arcroom decide, with the reason and who agreed.
  2. Recall puts it in front of any agent writing a metric query or dashboard.
  3. If the definition changes, the new decision replaces the old one and the old memory is reported with arcroom wrong.

What the agent sees

$ arcroom decide "active user = 1+ non-login event in 28 days" --why "logins inflate DAU by 30%; agreed by product + marketing, Oct 20"
Decision saved. Agents see it from the next prompt.

Compared with today's tools

Semantic layer (dbt metrics or BI tool metrics)

Define active_users once in the semantic layer so every dashboard uses the same calculation.

Setup
2–5 days
Ongoing
low once set up; new metrics go through the layer

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
2 min (npx arcroom init)
Ongoing
None

A semantic layer is the stronger fix because it enforces the definition; ArcRoom helps when agents write raw SQL that doesn't go through it.

Where it stops

ArcRoom tells the agent the definition; it can't stop a query that ignores it. A semantic layer enforces it in the warehouse.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in data, ml and research