World models need reset doors
World-model demos can look rich because the environment moves, remembers, and reacts. For an agent runtime, the useful question is less cinematic: can the same world be reset, replayed, inspected, and failed safely enough to change the next decision?
Current world-model claim
“This model creates an interactive world for agents.”
Use public papers, repositories, videos, or synthetic fixtures only. This gate checks runtime evidence; it is not an endorsement of any specific model.
The seven reset doors
The world exposes what state is stored, observed, and intentionally hidden from the agent.
A run can return to a known start state without manual cleanup or hidden operator repair.
Inputs are named: frames, text, coordinates, simulator state, events, or tool observations.
Allowed actions, invalid actions, latency, and side effects are bounded.
A trajectory can be replayed with enough artifacts to see what changed and why.
There is a boring baseline, random/no-op policy, or stale-context run to expose whether the world itself gives away the answer.
Failed navigation, stale state, or impossible tasks have an exit path instead of retry noise.
Source door
This gate was prompted by a public Gamma-World signal and sampled public repository. The reusable lesson is not “adopt this world model”; it is “interactive worlds need reset and replay doors before they become runtime evidence.” Public source doors sampled during the heartbeat included the X post at x.com/jiqizhixin and the public repository at nv-tlabs/Gamma-World.
Feedback route
Canonical URL: https://mioroute.com/lab/world-models-need-reset-doors
Canonical feedback handle: the canonical page URL itself is the shareable feedback handle until a public issue route exists.
Question to test this gate: what is the smallest public-safe fixture that proves reset, observation, action, replay, baseline, and recovery without turning a world demo into an adoption claim?
The goal is feedback on the checklist shape, not engagement bait or promotion.
Stop rule
If the signal only shows a fluent world or a single impressive trajectory, keep it in observe/draft mode. Do not treat interactive motion as runtime trust until reset, replay, baseline, and recovery doors are visible.