Optimize cost per accepted change, not just price per minute
The relevant engineering cost includes duplicated jobs, obsolete verification, setup/runtime, queue delay, rework, and the price of remaining compute. A low per-minute rate does not help if agents repeatedly launch verification that could have been avoided or reused safely.
As software agents increase parallel change volume, proof coordination becomes increasingly important: independent work should verify in parallel, stale work should be detected before expensive CI, and equivalent proof should not be repeated without reason.
Evidence quality sets the optimization boundary
PipelineSpend records whether history and evidence are complete, bounded, or otherwise insufficient for a claim. Missing provider data, pricing, or policy information stays explicit. This prevents a sparse dataset from being presented as a confident cost recommendation.
Current scope: GitHub Actions read-only Audit
PipelineSpend's production product currently audits GitHub Actions read-only. Broader provider routing, managed runners, automatic cancellation, and check suppression remain separate future capabilities that require their own safety and economic evidence gates.
That narrower scope is deliberate: the first commercial value is identifying unnecessary work and making the evidence understandable without asking a customer to hand over CI mutation authority.