CIPatch
Our own numbers

What 186 public repositories actually save

Every repository the public scanner has priced, in one table. We publish it because the result is not flattering: most repositories have nothing worth fixing, and a tool that tells you otherwise is selling something.

186
repositories scanned
$0.00
median saving, per month
$89.23
largest saving we have found
6
would clear the $29/month price

The distribution

Saving per monthRepositories
Under $1124the scanner found no change worth making
$1 to $532real, and far below the monthly price
$5 to $2020still below the $29 monthly price
$20 or more10the band where a plan can pay for itself
$100 or more0none, so far

0 of these were priced from real run and job history rather than from the shape of their workflows, which is the difference between a measurement and an estimate. A repository's own file cannot tell you how often it runs, which is why the estimates are ranges and why the measured ones are the ones to trust.

What this means if you are reading it

If your CI bill is small, you are most of the way there and CIPatch will tell you so rather than invent a reason to charge you. That is why the price is a flat one and the guarantee is on the month: a saving of ten cents cannot justify a $29 plan, and the dashboard will say so next to the fee. A saving appears when a workflow does something specific and expensive: a job asking for macOS at $0.062 a minute that needs nothing macOS can do, or a full run starting for every commit on a pull request when the previous one is still going.

The public scanner reads public repositories only. Most of the money is in private repositories, where CI volume actually is — which is what the app measures and the public scan cannot see.

Check your own repository instead of reading about other people's.

It takes a few seconds, needs no account, and changes nothing.

Scan a public repository

Method: workflow files are read from the default branch, every job is priced from GitHub's published runner rates, and a change is only counted when a supported rule can make it without touching application code. Figures for the same repository change when GitHub changes its rates; the price table is versioned and dated. How the scanner works.