Three trigger families.
One missing documentation entry.
| Scope | HTTP | Kafka | Scheduled | Total |
|---|---|---|---|---|
| Service A | 11 | 1 | 3 | 15 |
| Service B | 8 | 3 | 5 | 16 |
| Combined | 19 | 4 | 8 | 31 |
The independent committed-source audits and the catalogue agreed on 31 of 31 roots, with no extras within those scopes. Each annotation anchor fell within its corresponding declaration range. The combined catalogue returned two pages (30 + 1) with one stable catalogue digest, no duplicates and a final null cursor. Comparing Service A with its existing documentation exposed one omitted scheduled job. No speed or token-saving claim was measured in this pilot.
Bind the scope.
Compare the complete inventory.
- Use the packaged candidate. Run the v0.3.1 core release candidate in
RELEASEmode against two related private services. Both use Kotlin 2.3.0 / Java 21, Maven and Spring. Kotlin analysis uses the packaged K2 engine and retains the native-project version separately. - Select committed production source. Use explicit
--committedanalysis. The developer checkout contained local edits; those edits stayed in place and were excluded from the snapshot. - Enumerate the selected compilation. Open the admitted analysis context and read the Spring catalogue with its callable identities, source anchors and coverage boundaries. Root enumeration is independent of which search terms started navigation.
- Audit the same committed bytes independently. Inventory HTTP handlers, Kafka listeners and scheduled jobs from source, then compare membership in both directions. The 31/31 result includes the check for unexpected extra roots. Five outbound Feign client methods were excluded from the inbound HTTP inventory after the pilot exposed and corrected their misclassification.
- Compare with the document. Use the audited inventory to identify missing documentation. The observed gap was one scheduled job.
Inspect the send.
Keep the handoff conditional.
In both services, an exact Kafka callable identity from the catalogue led to a supported source result. Inspection of retained outbox source distinguished successful, failed and unresolved sends, giving the agent a concrete basis for describing delivery behavior.
A possible handoff between the services remained unproven because the committed topic declarations differed. The document therefore retains a candidate relationship and its unresolved boundary; root enumeration alone does not establish an integration edge.
A declared trigger
is a place to start investigating.
Conditional analysis
The project's Kotlin compiler and the packaged analyzer differ. The catalogue retains that boundary; this pilot does not qualify mutation or prove native-compiler equivalence.
Runtime remains unproven
Bean activation, profiles, resolved configuration and programmatic registrations need additional evidence. Annotation inventory does not establish that every listed handler is active in a deployed environment.
Thread documentation needs source
A root identifies a callable and its trigger. Describing subsequent calls, side effects and inter-service handoffs still requires inspecting source and resolving the relevant boundaries.
Only anonymous aggregates are published here. Private repository names, paths, source, symbols, topics and content digests are omitted. The private audit is not a public reproducibility dataset.
Read the implementation.
Keep distinct studies distinct.
This report describes the verified packaged v0.3.1 core RELEASE candidate built from public Codeclew source e1d4e6b. Public regressions exercise the catalogue contract, annotation resolution and cross-repository pagination; they complement the private-service audits without reproducing their confidential inputs.
The earlier 16/16 Kotlin engineering study covers a separate test ledger. Its historical paired-task measurements do not measure this pilot or promise performance for the current release.