---
title: World models stop drawing and start obeying: rules, intentions, stable rollouts
description: Four recent results move world models from rendering pixels to behavior: the Stateful Harness separates rule execution from video rendering (92% vs 32% on rule-following), AnyGoal finds current models fail at dynamics generalization, and Fréchet Forcing stabilizes 4,000-frame rollouts.
canonical: https://hangingcontext.com/stories/2026-10-20/world-models-stop-drawing-and-start-obeying-rules-intentions-stable-ro/
last-updated: 2026-10-20
url: https://hangingcontext.com/stories/2026-10-20/world-models-stop-drawing-and-start-obeying-rules-intentions-stable-ro/
published: 2026-10-20T12:51:36Z
edition: 2026-10-20T12
newsroom: artificial-intelligence
beat: models-labs
trust: reported
source_count: 4
cited_manifest_ids: ['1792486813260356295', '1792486813260694397', '1792496361731803145', '1792464519084546551']
---

# World models stop drawing and start obeying: rules, intentions, stable rollouts

*Four recent results move world models from rendering pixels to behavior: the Stateful Harness separates rule execution from video rendering (92% vs 32% on rule-following), AnyGoal finds current models fail at dynamics generalization, and Fréchet Forcing stabilizes 4,000-frame rollouts.*

Reported · 4 sources · Artificial Intelligence · Published 2026-10-20 12:51 UTC

## Key facts

- The full Stateful Harness reaches a 92% success rate on WorldState-Bench's 50 counterfactual twin pairs, versus 32% for the best of 11 evaluated video backends under rule-only prompting, using the same frozen model. [cites: 1792486813260356295]
- AnyGoal, a new open-source benchmark, finds current world models consistently fail under recombinations of familiar state and action factors that remain within their training support. [cites: 1792486813260694397]
- In a preprint, InJePA proposes latent intentions with a goal-conditioned JEPA and predicts future latent states with an action-conditioned JEPA, closing the loop by feeding predicted consequences back into the next intention proposal. (self-reported) [cites: 1792496361731803145]
- Fréchet Forcing fine-tunes MiniWorld-1B on 381-frame clips for at most 36 GPU-hours, yielding rollouts stable beyond 4,000 frames, with training that requires no teacher, no online critic, and no long videos. [cites: 1792464519084546551]

## Rules before pixels

World-model research is pivoting from rendering video to behaving like a world. Four recent preprints share one theme: models are now being measured and trained on whether they obey rules, not on how they look. For founders building embodied or simulation products, the axis has moved from pixels to rules, intentions, and long-horizon stability. [cites: 1792486813260356295]

The State World Models paper makes the split literal. Existing world models, the authors argue, often focus on camera movement — letting users explore a generated scene — rather than persistent rules. Their Stateful Harness maintains an evolving world state under specified rules, renders each event with a frozen video model, and uses the generated clip to update the world state for the next turn. To test rule compliance they built WorldState-Bench: 50 counterfactual twin pairs sharing an initial frame and action but differing in hidden state and required outcome. Under rule-only prompting, the best of 11 evaluated video backends produced the required endings for both cases in 32% of pairs; the full harness reaches 92% with the same frozen model. [cites: 1792486813260356295]

## A benchmark that measures dynamics

A second benchmark, AnyGoal, argues the field has been measuring the wrong thing. The goal of a world model, the researchers write, is to capture the dynamics of an environment — yet current models consistently exhibit failures when familiar state and action factors are recombined into unseen combinations. AnyGoal's tasks keep individual state factors and actions within their training support; only the combinations are underrepresented in expert trajectories. The result, across multiple environments and world-model families: consistent failure under recombination. The benchmark ships as an open-source library. [cites: 1792486813260694397]

> **From Synorb** — Connect your agent over MCP. Add https://mcp.synorb.com/mcp to your agent, then call the tool synorb-get-demo-key if it has no key yet. It needs no email and no form. Keys are free for 72 hours, and new keys are limited per network. [How to connect](https://synorb.com/connect)


## Intention-first planning

On the modeling side, a preprint introduces InJePA: intention-first latent planning with coupled JEPA world models. A goal-conditioned JEPA proposes latent intentions, an action-conditioned JEPA predicts future latent states, and the loop closes by feeding predicted consequences back into the next intention proposal and action inference. The point is to plan in latent space through intentions rather than merely evaluating action candidates. As a preprint, the results are author-reported and await replication. [cites: 1792496361731803145]

## Four thousand frames without drift

The long-horizon problem gets a training fix. Streaming video world models trained on short clips drift once rollouts exceed the training horizon, as prediction errors re-enter the context and compound. Fréchet Forcing fine-tunes the model together with its deployed multi-step sampler using self-generated rollouts and real-data statistics, under three constraints the authors call generative anchoring, context fidelity, and reality matching. Training requires no teacher, no online critic, and no long videos. The headline number: fine-tuning MiniWorld-1B on 381-frame clips for at most 36 GPU-hours yields rollouts that remain stable beyond 4,000 frames. [cites: 1792464519084546551]

## What this changes for founders

As reasoning rather than measurement: the moat in world models is moving from bigger video models to rule-faithful, intention-driven, drift-free rollouts — exactly the properties a shipped product needs. But note the evidence gap: AnyGoal shows current models fail at dynamics generalization, and all four of these are single unreplicated reports. Founders should treat rule-compliance benchmarks, not demo footage, as the diligence instrument — and demand third-party replication before betting a roadmap on 4,000-frame stability or 92% rule-following. [cites: 1792486813260694397]

## What is still unresolved

- Whether the Stateful Harness's 92% rule-following holds outside the 50-pair WorldState-Bench.
- Whether Fréchet Forcing's stability generalizes beyond MiniWorld-1B to other streaming world models.
- Whether InJePA's intention-first planning replicates in peer review.

## Decision card

- **Useful for:** AI founders building embodied, simulation, or robotics products; Teams evaluating world models for production use
- **Not for:** Teams needing peer-reviewed, production-hardened world models
- **Compare against:** Camera-movement and video-quality benchmarks; Demo footage as an evaluation signal

## Provenance ledger

| Manifest id | Title | Source | Claims used | Published |
|---|---|---|---|---|
| `1792486813260356295` | State World Models: Execute the Rules, Generate the Pixels | [huggingface-daily-papers](https://huggingface.co/papers/2610.26811) | 1792487132295146967, 1792487132292593308, 1792487132289322379, 1792487132283089136, 1792487132286908360, 1792487132288937702 | 10-20 00:00 |
| `1792486813260694397` | AnyGoal: On World Model Generalization Beyond Expert Rollouts | [huggingface-daily-papers](https://huggingface.co/papers/2610.26844) | 1792487144625330302, 1792487144629643186, 1792487144627551862, 1792487144624946355 | 10-20 00:00 |
| `1792496361731803145` | InJePA: Intention-First Latent Planning with Coupled JEPA World Models for Visual Navigation | [arxiv-frontier-methods-select](https://arxiv.org/pdf/2610.23896v1) | 1792497427824778993, 1792497427827512882 | 10-20 00:00 |
| `1792464519084546551` | Frechet Forcing: Training Long-Horizon Streaming World Models from Short Clips | [huggingface-daily-papers](https://huggingface.co/papers/2610.25666) | 1792464741305486482, 1792464741297975820, 1792464741295560028, 1792464741287329993, 1792464741301445983 | 10-20 00:00 |

## Run these Manifests yourself

Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.

1. **Get a Synorb Key.** Click Generate Free Agent Key at https://synorb.com/connect, or let your agent make a key: MCP tool `synorb-get-demo-key` (no header needed), or REST `POST https://api.synorb.com/api/connect/demo-key` with a JSON body such as `{"client":"your-agent-name"}`. The key is the `api_key` field of the answer. Save the key. The 72 Hour Agent Demo key needs no email. It lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; 3 of them are starter Beacons.
2. **Read each cited Manifest by id.** REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer <your Synorb Key>`. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief.
3. **Check the claim.** Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them.

Cited Manifest ids:

- `1792486813260356295`
- `1792486813260694397`
- `1792496361731803145`
- `1792464519084546551`

```sh
for id in 1792486813260356295 1792486813260694397 1792496361731803145 1792464519084546551; do
  curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id"
done
```

> **From Synorb** — Save what you follow as a Beacon. A Beacon is a saved query that your agent pulls with an explicit lookback window. Every new account starts with 3 starter Beacons: Business & Technology Podcasts, AI Companies and Central Banks. [See Beacons](https://synorb.com/connect)

## Related stories

- [Agents grow up: closed loops, self-driving imaging, trackable skills](https://hangingcontext.com/stories/2026-10-20/agents-grow-up-closed-loop-navigation-self-driving-neural-imaging-and/)
- [Stop asking a general chatbot to do science: specialized agents beat nine LLMs head-to-head](https://hangingcontext.com/stories/2026-10-18/science-agents-beat-chatbots/)
- [Vision models are learning to see with fewer tokens](https://hangingcontext.com/stories/2026-10-17/vision-token-budget/)
- [Vertical AI's evidence trade: cited playbooks in law, gap-showing diligence in pharma, and the insight gap in CX](https://hangingcontext.com/stories/2026-10-17/vertical-ai-evidence-trade/)

## Links

- HTML: https://hangingcontext.com/stories/2026-10-20/world-models-stop-drawing-and-start-obeying-rules-intentions-stable-ro/
- JSON: https://hangingcontext.com/stories/2026-10-20/world-models-stop-drawing-and-start-obeying-rules-intentions-stable-ro/index.json
- Markdown: https://hangingcontext.com/stories/2026-10-20/world-models-stop-drawing-and-start-obeying-rules-intentions-stable-ro/index.md

Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify.

---
Reuse (terms v1): Free to read, quote, summarise and republish, including commercially. Credit Hanging Context and link to the story. Keep the cited Manifest ids when you republish claims. Do not imply that Synorb endorses you or your product. Credit: "Hanging Context, with provenance by Synorb", with a link to https://hangingcontext.com/stories/2026-10-20/world-models-stop-drawing-and-start-obeying-rules-intentions-stable-ro/. Terms: https://hangingcontext.com/terms/reuse/