Agent: an AI identity that does work and spends tokens, for example a Claude
Code session, a Cursor user, or a support bot.
Outcome: a unit of shipped work we can score. In engineering the main
outcome is a merged pull request; reverts and reviews count too.
Cost event: a recorded LLM or tool call with its token usage & dollar cost.
Value score: how Yardstick ranks an agent on the value it created against
what it spent.
Cost per merged PR: the headline engineering metric, spend divided by
merged pull requests.
Change-failure: work that had to be reverted or fixed, a negative signal on
quality.
Active agent: an agent that produced at least one measured outcome during
the billing month. Idle agents do not count toward your plan.
Suite: a domain pack (engineering today; support, sales, and research
follow the same spine).
---