Reference
How to cite
The metric is free to use, adopt, and reference. The original definition appears in The Delivery Gap; this site is the canonical online definition.
In writing
For papers, posts, and reports, the simplest reference is:
Hill, B. (2026). The Delivery Gap. See aifinops.dev for the canonical definition.
BibTeX
@book{hill2026deliverygap,
author = {Hill, Brenn},
title = {The Delivery Gap},
year = {2026},
}
@misc{cpac2026,
author = {Hill, Brenn},
title = {Cost per accepted change: canonical definition},
year = {2026},
howpublished = {\url{https://aifinops.dev}},
}
Plain text
Cost per accepted change, defined in Hill (2026), The Delivery Gap.
Canonical reference: https://aifinops.dev
Inline first mention
The recommended phrasing for the first mention of the metric in any document is:
cost per accepted change — the fully-loaded cost of producing software that reaches production and stays there, divided by the number of changes that do (Hill, 2026).
In dashboards and tooling
When the metric is exposed in a dashboard, report it alongside the formula and the window length. The minimum recommended label is:
Cost per accepted change — last 4 weeks
$ per change that reached production and stayed there
Source: https://aifinops.dev
Naming convention
The metric is written as cost per accepted change — lowercase, no hyphenation — in formal writing. The acronym CPAC sees informal use across this site and in practitioner discussion; readers should note that it collides with an unrelated US political acronym in web search, so the full phrase is preferred in document titles, headlines, and SEO-relevant text.
The Delivery Scorecard and machine catch rate
The Delivery Scorecard is the AI FinOps presentation of complementary cost, detection, stability, recovery, and flow measures. Use the full names cost per accepted change and machine catch rate on first mention; CPAC and MCR are short labels. There is no combined score or claimed industry standard.
Machine catch rate — the percentage of all known, confirmed issues first discovered by a machine before production. A human may implement the fix. Canonical definition: https://aifinops.dev/#machine-catch-rate.
When reporting MCR, include the numerator, denominator, change cohort, production boundary, and observation cutoff. Credit discovery, deduplicate defects, and report N/A when there are no known issues.
Credit DORA for change fail rate, deployment rework rate, and failed deployment recovery time. Credit the Kanban Guide for throughput, cycle time, work in progress, and work item age. The scorecard does not rename these measures or claim their invention.
Version and governance
| Version | v0.1.0 (reference release, May 2026) |
|---|---|
| Editor | Brenn Hill |
| Definition stability | The formula and the size-normalization rule are stable for v0.x. Refinements to recipes, worked examples, and supporting documentation are welcome. |
| Refinements | Describe the measurement window, counting boundaries, and empirical evidence behind a proposed refinement. |
This is a reference definition, not a self-declared standard. Adoption builds the standard; the site provides the shared vocabulary.
License
The definition is free to use. The calculator library is MIT-licensed.