5.7 KiB
Capability status — IMPLEMENTED / WIRED / QUALIFIED
Three different questions get three different answers, and conflating them is what the 2026-09-18 techlead review found over and over: a capability with 12,000 lines and 482 green tests that no running application assembled, and a provider with no wiring that was nonetheless live. "There is code for it" is not "the shipped artifact builds it" is not "it has been proven against a real provider."
| Column | The question | Who decides it |
|---|---|---|
| IMPLEMENTED | Does this repository contain the capability? | src/config/architecture/modules.json — the leaf is registered |
| WIRED | Does the shipped app-bootstrap artifact assemble it when its switch is on? |
the leaf's META-INF/spring/...AutoConfiguration.imports, its edge in app-bootstrap/build.gradle, and a composition test that boots CaSkeletonApplication |
| QUALIFIED | Has it been proven against a real provider, by something CI runs? | a strict test lane against a real provider and a workflow that executes it |
The third column is the one that rots quietly, because a lane can exist and be invoked by nothing. Each row below names the lane and the workflow, so "qualified" can be checked rather than believed.
Status
| Capability | Master switch | IMPLEMENTED | WIRED | QUALIFIED | Evidence |
|---|---|---|---|---|---|
| Relational persistence (JPA/PostgreSQL) | APP_PERSISTENCE_JPA_ENABLED |
yes | yes | yes | six postgresql*IntegrationTest lanes, run by integration-main.yml → jpa-provider-ownership-contracts |
| Document persistence (MongoDB) | APP_PERSISTENCE_MONGO_ENABLED |
yes | yes | yes | six container lanes (mongoReplicaSetTest, mongoFailoverTest, …), run by integration-main.yml → mongo-container-lanes |
| Messaging (broker-neutral core) | APP_MESSAGING_ENABLED |
yes | yes | Kafka only | messaging-kafka:messagingCertificationTest behind verifyMessagingCertificationEvidence, run by messaging-certification.yml. No broker starter ships: a deployment composes messaging-spring-boot-starter-kafka or -rabbit itself, and selecting a broker it did not compose is refused at startup. Rabbit ships validators and no transport. |
| Redis capabilities (cache / rate-limit / lease / idempotency) | APP_REDIS_ENABLED |
yes | yes | yes | cache-redis:redisTopologyTest, run by redis-sdk-topology.yml |
| Notification platform | APP_NOTIFICATION_PLATFORM_ENABLED |
yes | yes | claim-scoped | verifyNotificationEvidence refuses a grade in docs/notification/support-matrix.md that no proven claim backs; run by notification-platform.yml. There is no real-SMTP lane, and the matrix says so rather than implying one. |
| File export publication (R2) | APP_FILESERVER_ENABLED |
yes | yes | yes | A different capability from the platform row below, in its own namespace so the two cannot be switched on together by accident. Atomic publication and durability are proven by adapter:outbound:fileserver's own lanes inside its check, which every PR runs. |
| File server platform | APP_FILESERVER_PLATFORM_ENABLED |
yes | yes | yes | app-bootstrap:integrationTest → FileserverCompositionIntegrationTest (required by the strict lane, so it cannot silently disappear), run by integration-main.yml → bootstrap-integration |
| Outbound HTTP client | APP_HTTPCLIENT_ENABLED |
yes | yes | yes | httpclient:httpClientFailureInjectionTest against a faulted upstream, run by integration-main.yml → httpclient-fault-injection |
| GraphQL transport | APP_GRAPHQL_ENABLED |
yes | yes | compose-level | local-graphql and the three off-* lanes in src/config/runtime/compose-profile-contracts.json, run by integration-main.yml → compose-runtime-smoke. The route's absence when the switch is off is what those lanes prove. |
| Object storage | app.object-storage.enabled (no MasterSwitch entry) |
yes | no | MinIO only | The leaf is not an app-bootstrap dependency — only sample-portfolio consumes it, and app-bootstrap:integrationTest assembles it test-only against digest-pinned MinIO. A fork that wants it adds the edge. |
| gRPC platform | — (build-level) | yes | no | yes | Absent from the main build entirely: optional-platforms is a separate Gradle build, so the switch is the include, not an environment variable. -p optional-platforms ci runs on every PR and integrationCheck certifies the operation ledger against real PostgreSQL. |
| WebSocket transport | — (build-level) | yes | no | — | Qualification-only edge in app-bootstrap; ConditionalTransportBuildContractTest asserts it has no production edge. |
Reading the "no" rows
A no in WIRED is not a defect. It is the review's Build Composition principle working: the repository holds more capabilities than any one derived service should ship, and a capability that is absent from the artifact holds no beans, no connections and no threads whatever its configuration says. The three no rows are the ones where that choice has been made and is enforced by a test.
The rows that say yes are the ones where the choice is still "ship it and switch it off", which
is a weaker guarantee — correct, but it depends on the switch being honest rather than on the code
being absent. Moving more rows to no is what app-bootstrap's composition-ownership work is for.
Keeping this honest
CapabilityStatusDocumentTest derives the switch list from MasterSwitch and fails when a switch
has no row here. That is deliberately the only automated check: a table of facts that a parser
re-derives would be a second source of truth, and the columns above are judgements about evidence,
which is exactly the thing the review says a build must not pretend to compute.