Multi-agent systems spin. Most of that spinning is motion without progress. TFB Hamster is the lean layer that watches the wheel and refuses to call motion progress unless it actually closes distance to the goal.
Motion is not progress unless it closes distance to the goal.
Press Play. An agent works on a real-shaped task. Distance-to-goal opens, narrows, then quietly stops narrowing. The conscience notices before the agent does, and closes the loop on the agent's behalf.
Three plain primitives. The internals stay private; the shape is enough to see why the wheel cannot lie to itself.
The job, the success conditions, the constraints, and the places where a human still has to judge. Sealed at the start of work so the finish line cannot drift.
What the agent saw, did, and changed. Hash-chained so the past cannot be quietly rewritten and motion can be checked against actual progress, not narration.
When a loop is caught, the lesson becomes a named guardrail bound to its source. The next agent in the same shape of trouble meets the lesson on the way in.
TFB Hamster is not a planner, a router, or a smarter model. It is the layer that watches the wheel and refuses to let motion masquerade as progress. It rides under whichever agents are doing the work and stays out of the way until it has to speak.
When it speaks, it speaks once, quietly, with a receipt.
Live inside Trust Fund Baby on its own build cycles. A monitored lane is up on the Withyou Operator OS. Pilot review runs weekly; the gate that opens it for outside work is a calendar gate, not a software one.
Reach out only if you have a specific multi-agent loop you want a second pair of eyes on. We are not collecting waitlist signups.