Level 2: Certified ArcFlow Systems Engineer

Title: Certified ArcFlow Systems Engineer

Prerequisite: Level 1: Workflow Developer

What certified means at this level

You design graph workflows with conditional routing and joins, operate vector RAG ingest and retrieval, apply retry timeout and fallback patterns, implement HITL approval flows, integrate external callbacks with HMAC verification, configure SDK streaming, and read complex traces including retry events, MemoryRetrieved, and HITL states.

Competencies added over Level 1

CompetencyDemonstration
Graph DAGBranch selection and join nodes with trace proof
Vector RAGIngest plus query with MemoryRetrieved
ReliabilityRetry, timeout, or step fallback on at least one step
HITLInterrupt and approve via API
External callbacksOutcome posted with verified auth
Streamingrun_stream or runStream consumer
Complex tracesInterpret retry and interrupt metadata

Continue with

Practical project

Build a graph-routed customer support workflow with RAG, HITL escalation, retry configuration, and external webhook notification.

Architecture sketch

text
classify (graph entry)
 ├─ billing branch
 └─ technical branch
 └─ join → answer with RAG
escalation branch → HITL manager approve → external notify webhook

Requirements

RequirementDetail
Graph routingAt least two conditional branches plus one join
RAGKnowledge base ingested; answers cite retrieved context in output behavior
HITLEscalation path interrupts; approve via POST.../approve/{key}
RetryConfigure retry on a flaky tool or stub failure injection
External webhookPost outcome via report_outcome or signed HTTP
Server runtimeUse arcflow-server for HITL and external paths
Trace packExport trace showing MemoryRetrieved, graph nodes, interrupt, completion

Suggested examples to compose

PatternExample
Graph joinparallel_search.py
RAGdocument_qa.py
HITLexpense_approval.py
Externalplaywright_stub_callback.py
Streaming (optional UX)chat_stream.py

Verification commands

bash
docker compose -f docker/docker-compose.server.yml up -d
python your_support_workflow/run.py
bash examples/hitl/approve_cli.sh RUN_ID # when escalated
python examples/external/playwright_stub_callback.py --run-id RUN_ID --status success
curl -s "http://localhost:8080/v1/runs/RUN_ID/trace" -H "Authorization: Bearer dev-secret"

Pass criteria:

CheckPass
Wrong branch not executedGraph trace proves branch taken
MemoryRetrieved on FAQ pathyes
HITL Interrupted then Completed after approveyes
External callback acceptedrun advances
Retry event in trace when failure injectedyes
Trace policyno prompts in exported trace

Known gaps to acknowledge

IDImpact on project
Graph recovery resumeGraph checkpoint resume partial; design linear recovery fallback
Server SSE streamingServer SSE streaming deferred; use SDK streaming or poll

Note these in your README.

Self-assessment checklist

QuestionAnswer must be yes
Can you explain join preconditions?
Can you rotate HITL approval keys conceptually?
Can you verify webhook HMAC without disabling checks?
Can you read MemoryRetrieved without seeing chunk text?

Next level

Proceed to Level 3: Platform Engineer for deployment, static product, and trace data policy operations.