The spec called for a CapturingBreachPolicy @Alternative @ApplicationScoped static inner class — observe SlaBreachEvent calls from ExpiryLifecycleService, record the SlaBreachContext received, verify the policy isn’t NoOp.

Reasonable on the surface. But selected-alternatives is a build-time property. Activating CapturingBreachPolicy as an @Alternative displaces SlaBreachPolicyBean for every test in the file. The displacement check — the primary CDI concern — would be untestable in the same class.

The second problem: testing that ExpiryLifecycleService dispatches to SlaBreachPolicy is a casehub-work concern. SlaBreachLifecycleTest already covers it at Checkpoints 3 and

  1. A wiring test that duplicates integration coverage while breaking displacement is worse than what it replaces.

I brought Claude in to settle the alternative. We went with two methods and no alternatives at all:

@Test
void slaBreachPolicyBean_displaces_noOp() {
    assertThat(policy).isInstanceOf(SlaBreachPolicyBean.class);
}

@Test
void slaBreachHandler_onFail_signalsCaseContext() throws Exception {
    UUID caseId = caseHub.startCase(MINIMAL_CTX).toCompletableFuture().get(5, SECONDS);
    assertThat(caseId).isNotNull();
    breachEvents.fire(new SlaBreachEvent(ctx, new BreachDecision.Fail("sla-breach")));
    await().atMost(5, SECONDS).untilAsserted(() ->
        assertThat(instance.getCaseContext().getPath("humanApproval.status"))
            .isEqualTo("sla-breach"));
}

Event<SlaBreachEvent> injected as the test driver — fires directly, bypasses ExpiryLifecycleService. Since @Observes is synchronous CDI, the handler completes before fire() returns. If @Observes is accidentally changed to @ObservesAsync — the Layer 2 gotcha — the second test times out: the event dispatches off-thread and the await windows don’t overlap. Two concerns, two tests, no build-time changes.

The test context needed care. linesChanged=100 below the threshold keeps the human-approval binding from firing after startCase() resolves. Without it, firing the breach event into an already-active binding is a data race. We went through all eight bindings in the pr-review YAML to confirm none would fire with the pre-seeded context.

One postscript. SlaBreachContext’s Javadoc says to construct tests with MapPreferences.empty(). I wrote it in the spec. Claude flagged it during implementation: cannot find symbol: method empty(). MapPreferences has one constructor, no static factories. The Javadoc is wrong; use new MapPreferences(Map.of()).

<
Previous Post
Rendering with and without a Brain
>
Next Post
Two Wrong Build Times