Testing and verification

Run the narrowest relevant test while developing, then the complete component gate before handing off the change. PostgreSQL is mandatory whenever a claim depends on schema, locking, persistence or recovery.

Commands

Engine

cd engine
./mvnw test
./mvnw clean package

Studio

cd studio
npm ci
npm audit --audit-level=low
npm run lint
npm test
npm run build

Worker SDK

engine/mvnw -f sdk/java/pom.xml verify

Documentation

cd documentation
npm ci
npm audit --audit-level=low
npm run check
npm run build

Evidence matrix

Change	Required evidence
Core logic or REST behavior	Relevant tests and full engine test suite
BPMN behavior	Parser/validation test, executable BPMN fixture, success and rejection semantics
APL behavior	AplParser rejection matrix, executable APL fixture, AplRuntimeTest; keep the kitchen-sink APL twin 1:1 with the BPMN original and run Studio verify:kitchen-sink
Agent worker or external-task behavior	Durable lease/recovery evidence (AgentWorkerResilienceTest, WorkerHealthServiceTest, FirstPartyWorkerCapabilityTest) plus positive/negative global-fetch authorization
Insight behavior	Fact-commit/rollback, analyzer window recovery, proposal lifecycle and policy tests (PostgreSQL), stale-target SUPERSEDED checks
Migration or repository lock	PostgreSQL fresh/upgrade, rollback, restart and concurrency tests
Public API	OpenAPI compatibility test, typed errors and documentation
Authentication or authorization	Positive and negative JWT/proxy/role/resource/lane tests
Studio	Lint, tests, production build and APL round-trip checks
Documentation	Dependency audit, Astro check, build and internal-link validation
Release metadata	Backend/workers/frontends/SDK/docs, JAR hash, image and Compose validation

Failure-injection expectations

For atomicity and cluster changes, test crashes or exceptions before and after commit. Verify that failed commands leave no partial state/history/outbox and that committed work is recoverable without a duplicate workflow transition.

Exactly-once has a narrow meaning

Tests may prove exactly one committed internal workflow transition. They must still describe uncontrolled remote effects as at-least-once.

Reporting results

Record the exact commands, test counts, failures/skips and non-blocking warnings. Do not call a release ready when Docker/PostgreSQL evidence was skipped or a critical correctness defect remains.