DAG Journey with Typed Gates

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Onboarding includes approval, remote login, local setup and local build. Teaching dependencies are A approval→B login and C setup→D build. Only the approver decides A; agent execution of C/D is authorized. Pending approval need not block independent setup.

Mechanism

Model prerequisites as a DAG, separately recording executor, inputs and exit evidence for nodes. Execute authorized ready nodes only; human decisions stay pending while independent work proceeds. Verify current account/environment and artifacts on resume rather than trusting old completion marks. Resolve cycles before execution.

Bad example

Wait for A before any local work, or answer A yourself as approved and log in remotely.

Good example

Follow the teaching graph. A awaits actual approval; meanwhile complete authorized C and verify it before D. B needs valid A evidence plus current access checks. Persist pending/running/verified and evidence for each node without turning silence or old accounts into completion. Prerequisites and execution ownership are separate fields.

Why the change matters

Edges express prerequisite reasons; ownership identifies who can satisfy them. Separation enables independent progress without simulated human decisions.

Observable expectation

The fixture permits A pending with C/D verified while B waits. Actual approval then permits current-access checks before B.

Require evidence for verified states, no agent-invented approval and no stale credentials or other-machine build results treated as current.

Limits

A graph grants no tool permission. Approval/environment can expire; project authorization/evidence rules govern nodes. Parallel work also requires compatible resources. Provenance remains unconfirmed.