What the v10.4.0 examples gate measured
At tag v9.3.0, the five Java example pom.xml files pin the SDK to older releases: four at 9.0.0 and authzen at 9.1.0. A user who copies an example gets an SDK several minors behind the one the tag ships, and a registry-installed test of the examples silently exercises the old artifact unless the harness overrides the pin (the gate run built scratch poms for exactly that reason).
Two adjacent facts from the same run, recorded so they are not rediscovered:
- Two Go example modules ship
replace => ../.., which makes a registry test read the working tree instead of the published module.
- Java ships 5 examples where the other four SDKs ship 8 (no connectors, planning, proxy-mode or indonesia_compliance coverage). A coverage asymmetry to close or to state deliberately.
Why it is worth fixing rather than noting
The examples are the first thing a user runs, and an example pinned to an old SDK is a silent downgrade recommendation. The durable form is a check: the release-prep version sweep already walks version sites, and example build files are a site it does not reach. Pin the examples in lockstep with the SDK version and add the example build files to whatever the sweep asserts, so this cannot drift again.
Evidence: axonflow-internal-docs/engineering/testing-logs/E2E_SDK_EXAMPLES_V10_4_0_PUBLISHED_SDKS_SEP07_2026.md.
What the v10.4.0 examples gate measured
At tag
v9.3.0, the five Java examplepom.xmlfiles pin the SDK to older releases: four at9.0.0andauthzenat9.1.0. A user who copies an example gets an SDK several minors behind the one the tag ships, and a registry-installed test of the examples silently exercises the old artifact unless the harness overrides the pin (the gate run built scratch poms for exactly that reason).Two adjacent facts from the same run, recorded so they are not rediscovered:
replace => ../.., which makes a registry test read the working tree instead of the published module.Why it is worth fixing rather than noting
The examples are the first thing a user runs, and an example pinned to an old SDK is a silent downgrade recommendation. The durable form is a check: the release-prep version sweep already walks version sites, and example build files are a site it does not reach. Pin the examples in lockstep with the SDK version and add the example build files to whatever the sweep asserts, so this cannot drift again.
Evidence:
axonflow-internal-docs/engineering/testing-logs/E2E_SDK_EXAMPLES_V10_4_0_PUBLISHED_SDKS_SEP07_2026.md.