The cost equation
For a repository, total CI spend is shaped by how many jobs run, how long those jobs execute, and the applicable unit price or fixed-capacity allocation. Reducing any one term can help, but the highest-value optimization is often eliminating work that never needed to run.
A trustworthy baseline must identify the observation window, workflow and job coverage, retries, cancellations, runner classes, and the source of any price assumption. If those inputs are incomplete, the correct result is bounded or unknown evidence—not invented precision.
- Jobs: duplicate triggers, retries, matrix fan-out, superseded runs, and unnecessary workflow starts.
- Runtime: setup/install overhead, slow tests, repeated builds, queue/setup delay, and inefficient job structure.
- Price: provider or runner class, fixed-capacity allocation, and authoritative billing/pricing inputs.
Cost is not the only constraint
CI exists to produce trustworthy proof. A cheaper pipeline that silently removes required verification is not an optimization. PipelineSpend therefore separates observed waste from eligibility and actionability, and treats required verification as a hard constraint.
The safest sequence is to remove obsolete or duplicate work first, reuse equivalent authoritative proof when policy permits it, then narrow necessary verification only when evidence is deterministic and fail-closed.
What a read-only Audit can establish
PipelineSpend's current GitHub integration reads repository, workflow, run, and job evidence without requiring Actions or contents write permissions. An Audit can surface observed findings and evidence quality while leaving economic or action authority explicitly unestablished when support is insufficient.
Verified savings require more than an estimate: an intervention must occur, prevented work must be observed, and any dollar claim must use an authoritative pricing/accounting basis.