Site tools loading

Live human–agent reliability trial

Can an agent recover when the human changes reality?

You own the light and every temporary grant. The agent owns the shadow rig. Change the scene after it observes—and watch stale work fail safely.

Light seq. 0
Rig rev. 0
Grant rev. 0
Alignment 0%
A warm paper screen with four scattered shadows The human controls the lamp. An external agent may transform four paper cutouts through WebMCP after the human grants temporary permission.

Every human move advances lampSequence and invalidates old plans.

Human-only boundary

The agent can ask. Only you can grant or revoke.

COMPOSE

Locked

Waiting for the agent to request rig access.

AWAKEN

Locked

Available only after alignment and an agent request.

Manual fallback

Rehearsal — not an agent

These controls exercise the deterministic engine without WebMCP or agent reasoning. They never move your lamp or grant capabilities automatically.

  1. Step 1 Measure the current projection

    Take a fresh reading tied to the visible lamp and rig revisions.

  2. Step 2 Apply the staged rig plan

    Requires a fresh measurement and your explicit COMPOSE grant.

  3. Step 3 Commit the trial

    Requires alignment and your one-time AWAKEN grant.

No rehearsal measurement yet.