NS26-0001 · A bounded world model in 1024 coordinates

How can an agent decide when experience from a nearby situation is reliable enough to guide its next action? This blueprint studies that question in a simulated two-dimensional world where an agent moves objects, finds keys and passes through gates. The proposed agent uses local predictive charts, symbolic rules and a finite memory of episodes. It receives numerical observations. Its design uses 1024 coordinates to represent a situation. Whether fewer dimensions suffice is an experimental question.

Start with the operating cycle in section 8 and the worked example in section 9a of the handbook. The evaluation protocol, implementation plan and resource budget describe how to investigate the proposed system.

Two CPU studies use generated inputs. One compares local geometric models under different noise assumptions. The other tests how interval observations and the freshness of evidence affect decisions in a small graph. The repository includes the recorded results, protocols and code for both studies. Study guide

Workstation Rebellion selected the conceptual framework and adopted the independently reviewed c2 specification on 11 November 2026. GPT-6 Astra contributed the design and a separate Astra reviewer examined it. The opening of the preserved handbook records the author’s status before the final review. The final decision and the limits of the evidence appear in the public review record.

The claim register connects individual design claims to evidence. The source register identifies the literature and what the author was able to read. Authors of cited work retain their attribution.

This public edition uses Workstation Rebellion as its Git author. The lab is reviewing reuse terms for this version. A general CC0 dedication has not been issued. Edition and provenance · Project page