nx_pathway.nx

nx_pathway.nx -- cell-graph manifest. Biological analogue: a signaling pathway is a sequence of molecular events where signal transduction passes through ordered cells via vesicles. Nishi pathway is the substrate-level declarative composition format: a typed directed graph of cells with vesicle edges between them. Same .nx pathway runs local-only, split across laptop+NAS+peer GPU, or all-cloud -- the operator changes the per- cell tropism, not the graph. Per [[feedback-pathway-tropism-block-composition-location-agnostic]]: the structural EXCEED-axis vs ComfyUI / Langflow / Airflow DAG / K8s Argo / Ray Serve / Step Functions ASL / Temporal: those bind the location of work at graph-definition time. nx_pathway lets the SAME manifest run differently per cell's nx_tropism resolution. Composes: nx_budget -- each cell has a budget pointer; pathway aggregates for whole-pathway forecasting nx_attention_class -- each cell declares its class nx_tropism -- each cell declares preferred location nx_ribosome -- each cell declares ribosome target nx_metabolism -- pathway-level hot-site list informs replan nx_homeostasis -- migration signals at the cell level may trigger pathway-wide topology re-resolve V1 ships a fixed-capacity graph with cells + edges. Sub-graph composition (pathway-of-pathways) is queued; parallel sub-graphs are representable but the scheduler that exploits them is V2. Gap list (V1 honest perf verdict): - no typed payload validation between cells (caller's job) - no cycle detection (caller must build DAG-correct) - no per-edge bandwidth budget (queued for distributed scheduler) - no live re-routing (pathway is rebuilt, not patched, on migration signal) - no NxCellSpec field count is bumping the parser's 16-field ceiling -- stay strict to that contract per [[feedback-nishilang-16-arg-function-limit]]