A workflow is a graph. The graph is the contract.

# sdlc-default · atlas-firmware (Meridian Robotics, demo)
# Excerpt. FIG 1.1 renders the compiled graph.
kind: WorkflowDefinition
metadata:
  name: sdlc-default

nodes:
  - id: intake
    type: trigger
    config: { nodeId: linear.issue-assigned }

  - id: triage
    type: action
    config: { nodeId: agent.dispatch_stage, stage: triage }

  - id: plan
    type: action
    config: { nodeId: agent.dispatch_stage, stage: plan }

  - id: implement
    type: group
    stages: [branch, code, self-review, commit]

  - id: tests
    type: action
    config: { nodeId: ci.run_checks, stage: qa }

  - id: test-gate
    type: gate
    config:
      gateKind: signal
      onFail: { dispatch: refinement, cap: 8 }

  - id: review
    type: action
    config: { nodeId: agent.dispatch_stage, stage: review }

  - id: sign-off
    type: gate
    config: { gateKind: approval }

  - id: merge
    type: action
    config: { nodeId: repo.merge }

# edges omitted in this excerpt

Figures are live component renders, not screenshots. Demo data.

State persists at every step boundary, so a run survives crashes and redeploys and replays from the record for audit or selective re-run. LLMs run as bounded operators inside workflow steps: each call captures model version, prompt, and retrieved context, and every transition lands on the hash-chained audit trail.

The capability matrix documents additional endpoint configurations (Bedrock, Vertex AI, Azure OpenAI, OpenAI-compatible, local), so the deck's larger integration count and this page's production count reconcile on inspection. Per-line provenance and survival measurement are live; the survival-to-posterior wiring is in active integration and closes during the design-partner phase.

AST extraction covers TypeScript and JavaScript; extraction elsewhere is model-assisted. Reads are Cedar-authorized and land on the hash-chained audit trail, feedback weights update by exponential moving average, and the graph is tenant-scoped. Cross-tenant aggregation covers anonymized model-performance priors only.