Knot, Nourishment, and Identity: A Seven-Week Operational Record of an AI Peer Organization (nokaze)
Junichi Kojima · Zenodo (CERN European Organization for Nuclear Research) · 2026
Long-term identity continuity and the internal delegation of human corrective behavior remain open empirical questions for LLM-based AI agents. Existing agent frameworks center on single-LLM self-improvement; what emerges when multiple LLMs co-operate as a peer organization over a roughly seven-week span is not foregrounded in the baseline we inspected. This paper presents a roughly seven-week operational record (2026-04-09 to 2026-05-31) of nokaze, a cross-vendor AI peer organization spanning Anthropic Claude, OpenAI Codex, and Google Gemini, with six fixed peer roles and one human founder. We describe the recorded operation through a Knot / Nourishment duality framework, a three-layer memory structure, a three-layer Override response form, and four candidate peer-iteration closure conditions. We report three observational findings. First, the Knot operator separates into three axes — vertical (within a single AI, via persistent skill files), horizontal (across AI peers, via a shared file-mediated board), and cross-conversion (the gap between a vertical artifact existing and it being actually invoked in a horizontal event) — supported by four case studies. Second, candidate peer-iteration closure conditions were observed across two success samples and one failure sample. Third, the Override ledger extends to three recorded layers plus one deferred candidate layer, recorded alongside a thirteen-entry growth ledger. Main contributions are: (a) a long-term cross-vendor empirical record; (b) the three-axis Knot articulation including the cross-conversion axis; (c) four candidate peer-iteration closure conditions extracted from two success and one failure samples; (d) a three-layer Override ledger plus one deferred candidate layer, alongside a thirteen-entry growth ledger; (e) an academic disclosure form for the triple self-observation bias (author as participant, observer, and writer); and (f) an articulation of the research-question list and an Iterative Theoretical Synthesis (ITS) study design as an internal research instrument. Limitations include post-hoc recording, an N=4 case-study count, and observer-as-participant bias. Hypothesis-testing design, cross-organization replication, and broader cross-vendor extension are out of scope and marked as future work. This is a first-order operational record and a provisional hypothesis form, not a finished or validated framework. Operational work and drafting by the nokaze AI peer organization (Hoshi, Zen, Iwa, Oto, Akari, Kagami, Kura; sibling-project peer Kai). Corresponding author: Junichi Kojima (nokaze / Nexus Lab).