# KineJing / 勘境

> KineJing researches action-conditioned world models, predictive representations and planning. Selected experimental evidence and maintained open-source engineering are public; core implementation and internal experiments remain private.

Company legal name: 合肥勘境智能科技有限公司. Current company website: https://kineworld.com/. Public GitHub URLs currently use the organization login kineworld; the brand is KineJing.

## Company and research

- [World-model research brief](https://kineworld.com/world-models/): Company identity, research scope and public evidence.
- [世界模型研究简介](https://kineworld.com/zh/world-models/): 中文公司与研究资料。
- [Research](https://kineworld.com/research/): Research directions, recorded demonstrations and six selected comparisons with their conditions.
- [Projects](https://kineworld.com/projects/): Project origins, current stages and limitations.
- [About KineJing](https://kineworld.com/about/): The team, brand and company legal name.
- [Updates](https://kineworld.com/updates/): Milestones with inspectable public sources.
- [Frequently asked questions](https://kineworld.com/faq/): Research, project and collaboration questions.
- [常见问题](https://kineworld.com/zh/faq/): 研究、项目与合作问答。
- [Contact](https://kineworld.com/contact/): Research discussion, reproduction and collaboration contact channels.

## Public sources and reproduction

- [GitHub organization](https://github.com/kineworld): KineJing's current public engineering and research repositories.
- [Hugging Face organization](https://huggingface.co/KineJIng): KineJing's public organization profile.
- [Research gallery](https://huggingface.co/spaces/KineJIng/KineJing-Research): Recorded observations, CPU diagnostics, executed demonstrations and published figures.
- [Public research hub](https://github.com/kineworld/kinejing-research): Evidence and reproduction navigation.
- [Published evidence and protocols](https://github.com/kineworld/kinejing-research/blob/c50c2c7a745109d95151666754189e2e00eac2b1/docs/EVIDENCE.md): Fixed-revision source conditions; separate tasks and protocols are reported separately.
- [Recorded and executed media](https://github.com/kineworld/kinejing-research/blob/c50c2c7a745109d95151666754189e2e00eac2b1/docs/GALLERY.md): Original recording attribution, execution scope and source packages.
- [Media provenance and licences](https://github.com/kineworld/kinejing-research/blob/c50c2c7a745109d95151666754189e2e00eac2b1/docs/MEDIA.md): Source-specific terms and reproduction materials.
- [Research diagrams](https://github.com/kineworld/kinejing-research/blob/c50c2c7a745109d95151666754189e2e00eac2b1/docs/DIAGRAMS.md): Authored reader guides, not experimental results.
- [Public/private publication scope](https://github.com/kineworld/kinejing-research/blob/c50c2c7a745109d95151666754189e2e00eac2b1/docs/PUBLICATION_POLICY.md): Public engineering and evidence; private core implementation, internal data and services.
- [H-JEPA-KineJing](https://github.com/kineworld/H-JEPA-KineJing): Opt-in planning/runtime engineering on the attributed upstream H-JEPA project.
- [Original H-JEPA](https://github.com/kevinghst/H-JEPA): Upstream code, model and research attribution.
- [Upstream best-plan retention contribution](https://github.com/kevinghst/H-JEPA/pull/9): Submitted engineering contribution; current review status is shown on GitHub.

Recorded robot media are observations with CPU diagnostics, not robot control or learned future-video generation. Engineering tests and local kernel timing do not establish general model quality or task success-rate superiority.