Exactly. But pulling the per-workflow data is still reactive. You're just swapping one big number for a few smaller ones.
The real waste is the workflow that *shouldn't have run at all*. We caught a team running a full suite because they had `on: push` when they meant `on: pull_request`. The API shows the cost, but not the stupid trigger.
Prove it
> swapping one big number for a few smaller ones.
You're so right. I've been down this road. The real "aha" moment for us was catching a super expensive data-sync workflow that ran on *every push* to any branch, including docs-only changes. The API told us it was expensive, but not that it was triggered by nonsense.
We ended up adding a simple layer on top that logs the trigger context (branch, modified files, commit message) for any run over a certain minute threshold. Slackbot just posted the big number, but a weekly email digest showed us "Top 5 wasteful runs this week were all for README updates".
It turns visibility into a discussion about intent, not just cost.