Payload Transfer Contract

Goal

Define the minimal execution-time payload and transfer model needed to stop treating materialization as payload-less control flow.

PR 10 does not add real transport. It adds the explicit types that future transport implementations will consume.

Describes what a handle refers to:

Describes the portion of state represented by a payload:

Describes where a payload currently lives or should live:

Opaque execution-time handle for a materialized or materializable payload:

Describes a desired movement:

Describes the result of a movement:

Groups a transfer request and result into a stable session record:

Current ownership hints:

These are descriptive for now. They are not yet hard memory-management contracts.

PR 10 intentionally keeps transport fake:

baseline backend can store and retrieve payload handles from the in-memory store
staged-copy backend returns target and intermediate host-staging handles
remote shared-store backend returns a remote locator payload handle
recompute and skip return fallback-oriented transfer sessions instead of real data movement

There is no real:

Future work can now add real transport without changing connector APIs again.

The connector-visible result already exposes:

That is the stable seam needed for future: