Pillar: P7 Steering (the owner: "we need to be sure that what we are spending tokens on is valuable… I want to get to a point where we are capturing token ROI").
The forge records what each crew run cost and which pulls merged, but nothing puts the two together. In the last 7 days the crew ran 723 times for about $258 at list price and merged 88 pulls; about a third of the dollars went to runs that failed. Nobody could see that without a database query. Work done by outside sessions (a build team of other agents, a person's own assistant) is invisible to the ledger.
[ ] one function computes, for a repository and a window: dollars and runs; changes landed, by pillar; dollars per landed change (median and total); dollars on runs that landed nothing, grouped by why (failed with its top reasons, cancelled, parked, superseded, issue closed as not doing); and the issues that cost the most with what came of each
[ ] the board shows it as ONE line with one tap to the breakdown ("7 days: $258 → 88 landed · $85 landed nothing: 61 timeouts, 40 conflicts…"); owner facts only; nothing to a reader or a stranger
[ ] the same numbers over the API and an MCP tool (read-only), so an agent can ask before it plans
[ ] outside spend can be reported: an API and MCP spend report (who, which pull or issue, tokens or dollars, model), writers and crew only, shown separately from the forge's own runs and never trusted for limits
[ ] a milestone page shows what its limb has cost and landed, so a limb can be trimmed on evidence
[ ] tests: the numbers add up on a fixture with a landed change, a failed run, a cancelled run and an issue closed as not doing; a reader and an anonymous visitor get none of it
Pillar: P7 Steering (the owner: "we need to be sure that what we are spending tokens on is valuable… I want to get to a point where we are capturing token ROI").
The forge records what each crew run cost and which pulls merged, but nothing puts the two together. In the last 7 days the crew ran 723 times for about $258 at list price and merged 88 pulls; about a third of the dollars went to runs that failed. Nobody could see that without a database query. Work done by outside sessions (a build team of other agents, a person's own assistant) is invisible to the ledger.
spendreport (who, which pull or issue, tokens or dollars, model), writers and crew only, shown separately from the forge's own runs and never trusted for limits