The stricter label wins
A version number is a claim by the vendor. Most of the time it holds. Sometimes a vendor raises the major version on every release, even when little changes. And sometimes a patch is really a break.
So pinup measures where it can. And it has an opinion about what counts then: the measurement may relax only where it is explicitly trusted. Otherwise the stricter label wins.
01 — The opinion in one sentence
An update carries two labels. An analysis may relax a decision only where the rule explicitly trusts it. Otherwise the stricter, declared label applies.
The declared label comes from the version numbers: major, minor, patch and so on. The effective label comes from an analysis of what really changed. It is unknown as long as no rule asks for the analysis.
In the .pinup.yaml it looks like this:
packageRules:
- matchDatasources: [docker]
matchPackageNames: ["registry-1.docker.io/bitnamicharts/**"]
versioning: semver
analyze: true
- matchDatasources: [docker]
matchUpdateTypes: [major]
matchEffective: [patch, minor]
trustEffective: true
automerge: true
dependencyDashboardApproval: false
Without trustEffective such an automerge stays off. The merge request then names the rule that switched it on and that trustEffective is missing.
02 — The alternative: Merge Confidence
Renovate answers a similar question with Merge Confidence, a service by Mend. It evaluates how old a version is, how many projects already use it and how often updates to it pass. For widely used public packages that is a strong signal. The crowd often knows more than any analysis.
For my estate it did not fit. Many dependencies are internal packages or charts for which there is no crowd. And I did not want to make a decision depend on an external service whose basis I cannot check myself. So pinup deliberately did not rebuild Merge Confidence and measures on its own instead.
03 — The price
An analysis downloads both versions. That costs time and network. So it is off by default and runs only for dependencies a rule marks with analyze.
There is only one analyzer so far. It reads Helm charts. For everything else the effective label stays unknown, and the declared one applies.
Trust has to be spoken out. An analysis alone relaxes nothing. That is intended, but it means one more rule in the configuration.
04 — The occasion
The occasion was a chart vendor that raises the major version on every release. Under the declared label every one of those updates is a major and waits for approval. The application inside and the values I set are often the same.
The Helm analyzer compares appVersion, the keys in values.yaml, kubeVersion and the dependencies of both versions. Measured on Bitnami's Redis chart from 20.13.4 to 28.1.0: the application jumps from 7.4.3 to 8.10.1, and two keys are missing from the values. That gives breaking-values, the strictest label. No automerge.
In a test with a constructed 21.0.0 where application and values stay the same, the result is patch. Only that case may go through automatically with trustEffective.
What the analysis shows along the way
A chart brings container images of its own. They are in no file of the repository, so no manager reads them. The analyzer names them as evidence in the merge request, with every reference that moved between the versions.
05 — The enforcement
- Relaxing only with trust. If a rule with
matchEffectiveswitches automerge on because the analysis reports less than the version number, it stays off withouttrustEffective. Then the stricter, declared label applies. The analysis never makes an update stricter than the rules on its own. - Absence does not count. The value
unknownis not one a rule withmatchEffectivereacts to. An analyzer that could read nothing can never fire a rule. - The evidence is in the merge request. Below the update table pinup shows what was compared and with which result. If a rule switched automerge on without trust, it also says which rule that was and what is missing.
Frequently asked questions
Can the effective label also make an update stricter?+
No. The analysis can only relax a decision, and only with trustEffective. If it reports breaking-values for a minor, what the rules say stands. The finding appears as evidence in the merge request, and a rule can react to it with matchEffective: [breaking-values].
Are there analyzers for container images or npm packages?+
Not yet. The Helm analyzer is the first. Planned are a comparison of the rendered manifests with your own values and an image analyzer over OCI labels and attestations. An analyzer is a package of its own behind a fixed interface.
Does this also work with a renovate.json?+
Yes. pinup reads the rules from a .pinup.yaml just as from a renovate.json. analyze, matchEffective and trustEffective exist only in pinup, though. Renovate does not know them.
Conclusion
A version number is a claim; an analysis is a measurement. An analysis alone may not wave an update through. It may relax only where the configuration explicitly trusts it. Otherwise the stricter label stays.
Next week: the plan that comes before every change. The overview of all parts and the terms that recur are in the opening of the series.
Are your charts waiting for approvals they do not need? I will measure.
An analysis of your Helm charts with pinup: which major updates are really patches, which values disappear and where an automerge is worth it with a clear conscience.
About the author

Kai Ole Hartwig
Programming since 2002 – self-taught, set up my own business with KO-Web in 2012. Over 100 projects, with a focus on security, performance, automation and quality. Today freelance: DevSecOps consulting, training and software development.
