TL;DR
- Check how the seller labels its releases. Minor and patch can mean different things when firms use different rules.
- The Java/Maven study found faults outside major releases. Its older open-source sample does not predict faults in your own product.
- Separate the type of change from proof that an exact pair passed a stated task.
- Show what was tested, what is supported and what is still open. A pair not tested is not a pair known to fail.
A Release Label and a Tested Combination Are Different Evidence
A buyer asks whether a connector still works after an accounting tool gets a minor update. A page says “compatible with all minor updates.” The claim may be wider than the product team’s records. The buyer needs the version pair and task, not just a label.
Semantic Versioning 2.0.0 sets rules for a project’s declared public API: the interface other code can use. After 1.0.0, minor releases add features that work with the old interface. Patch releases fix bugs without breaking it. This is a rule for projects that use the scheme. It is not a test of each buyer’s setup.
For generative engine optimization (GEO), make the supported scope easy to read. A saved AI answer may turn one tested pair into a claim about all future releases. Check that claim against the source record.
Observed Breakage Was Not Limited to Major Updates
Jayasuriya and colleagues’ study, first published in December 2024, examined Java code built with Maven. Table 5 lists 8,118 entries under “Number of Impacting Artifacts.” By release level, the shares are 58.41% major, 33.49% minor and 8.09% patch. The base is that table’s entries, not all updates. Thus 8.09% does not mean that 8.09% of patch updates fail.
The study used builds and code checks to find source-level breaks. It did not test every task at run time. It cannot prove that a named B2B connector works.
Table 5: 4,742 major, 2,719 minor and 657 patch entries. Rounded shares total 99.99%. These are shares within Table 5, not per-update failure rates. Release levels describe version numbers, not proof that each library declared Semantic Versioning. Original results and table basis.
A Separate npm Study Ran Client Install and Test Scripts
Venturini and colleagues’ January 2023 preprint studied npm packages with working tests and code repository links. The data ran through April 2020. It reports 11.7% of sampled packages (45 of 384) and 13.9% of tested releases (450 of 3,230) affected by manifesting breaking changes. The release base covers cases where providers changed, not all releases in the sample.
Here, the study detected a manifesting breaking change through a failed client install or test script. The team restored older releases, resolved provider versions, removed lockfiles and checked failures by hand. Its definition includes some faults beyond changes to an API. This is not a current risk score for an enterprise product.
| Direct study | Actual result and base | Limit of the comparison |
|---|---|---|
| Java/Maven study, 2024 | Major / minor / patch: 58.41 / 33.49 / 8.09 percent of Table 5 entries | Table 5 uses “Number of Impacting Artifacts”; not per-update failure rates. |
| npm study, January 2023 v1 | 11.7% of 384 packages; 13.9% of 3,230 tested releases | Client install/test failures under a broader definition; not the same unit as the Java table. |
Figure 2. Evidence comparison. Separate author teams and ecosystems. One groups Table 5 entries by version level. The other measures affected packages and releases. Do not pool these shares or use them to rank the two ecosystems.
The common lesson is modest: a release label alone is not proof that one buyer’s setup was tested. This does not make versioning useless. Keep what a label is meant to convey distinct from what a test found.
Endor Labs’ vendor explainer Breaking Changes, Breaking Trust explains how changes to code interfaces and changes in behavior can disrupt users in different ways. That helps frame the question. Passing a build check is useful. It is not a test of every task a buyer expects.
A Completed Scope Record for a Connector Page
This example is fictional. The draft page claims support for “all minor updates.” Its approved record covers connector C4 with accounting app 5.2.1. The tested task sends a purchase order one way. A check confirms the right fields arrived. A repeat request did not create a duplicate. There is no result for app 5.3.0 or invoice updates.
The gap is an unsupported range. It is not a known fault in 5.3.0. The numbers below are fictional version labels. They are not research statistics, real products or tests we ran.
| Claim field | Completed fictional record | Public explanation |
|---|---|---|
| Exact pair | Connector C4 with app 5.2.1 | Name the pair covered by the approved result. |
| Task and pass rule | One-way purchase order; expected fields; no duplicate on repeat | State the task checked, including the relevant repeat behavior. |
| Open combination | No recorded result for app 5.3.0 | Mark as not tested in this record; do not call it incompatible. |
| Outside the test | Invoice updates and other app versions | Do not imply these passed from the purchase-order result. |
A useful revision reads: “The approved test record covers connector C4 with app 5.2.1 for one-way purchase orders. It includes the stated field and duplicate checks. This record does not cover app 5.3.0 or invoice updates.” The claim now matches the held evidence.
A support policy may cover more versions than the list of tested pairs. If so, the product owner must explain that policy and its basis. Do not present each untested pair as either a pass or a fault. Link current support terms, release notes and the test record where they are public.
The technical owner checks the pair, setup, task and pass rule. The editor checks that the page keeps those limits. This is a content review. It is not a reason to delay needed updates or a full software test plan.
Our integration ownership guide asks who maintains a link. Our software support-life guide asks how long service lasts. This article asks what proves a version pair works for a stated task.
Minor Updates and B2B Compatibility: Common Questions
Does the same rule apply to version 0.x?
Not by default. Semantic Versioning treats major version zero as early development. Its public API should not be treated as stable. Check the project’s own policy. Three numbers alone do not prove that it follows the scheme.
Does a successful install prove the whole workflow works?
No. Installing the tool, building code and completing the buyer’s task are different checks. Say which one was done. A task may also depend on settings, data shape or another tool outside the test.
Should a page claim support only for individually tested versions?
State the real support policy and its basis. A team may support a range through other checks and commitments. Keep that policy distinct from a claim that each pair was tested on its own.
Should buyers avoid all minor updates?
That does not follow from these studies. Updates can add useful fixes and features. The technical team should follow the right release and risk process. The content should explain the proof, scope and open questions.
Bring the Compatibility Claim and Its Version Records
If AI answers turn a narrow test into a broad claim, bring the page, release policy and approved test records to SemShift’s GEO service discussion. We can discuss how the page should explain supported scope, tested pairs and questions that need technical review.