Emergence Without Scale: Dynamics and Organisation Tested Against the Substrate Bet in a Spiking Architecture

What this is. Why intelligence emerges is usually asked as a question about scale. This paper tests the opposite bet --- that dynamics (global state variables that bias learning) and organisation (the arrangement of rules and modules) can carry a system further than substrate size would suggest --- and reports what an instrument built to test it measured. Three pre-registered readings. First, two local rules sharing one weight: a fixed 50/50 mixture is measurable (Fisher ratio 0.651 against 0.472 and 0.331), while choosing the mixing weight is not --- no adaptive scheme beats the fixed mixture, and the static optimum is the default value. Second, a cognitive layer of working memory (0.9269/0.9276), episodic memory (0.805; 0.922 at K=4), metacognitive gating across several OOD families, a world model with a termination head (AUROC 1.0 in receding-horizon closed loop), and a state-level end-to-end gate at 18/18. Third, orchestration by a replaceable language model: real dialogue passes every criterion, while the same spiking trigger that passes on a synthetic salience stream fails on a real one (AUC 0.732 against 0.7475 for a plain novelty gate, one trigger per conversation). What it implies. From those failures the paper derives an architectural rule: writes belong to the dynamics, the language model may only propose, and the bus is the only door. A released implementation is described --- a read-only, per-call-audited MCP server with four read tools and five proposal tools, a gate-mediated write path specified against six pre-registered criteria (specified, not yet built), and a host that enforces read-only at the storage engine. Where the numbers live. Every number traces to a reproduction record in the linked repository; negative results are reported with the same care as positive ones. v1.1 (2026-10-31). The author line now carries the Chinese name (沈锦垚) as an embedded vector graphic; no results, numbers or wording changed. v1.2 (2026-10-31). Typesetting and pointer fixes only; no result, number or claim changed. (1) Table 5's first column -- the Chinese section names of the pre-registration record -- was blank in v1.1 because no CJK font was loaded; it now renders. (2) Tables 1 and 2 ran past the right margin and their result column was clipped; they are re-typeset with fixed-width columns. (3) The availability section cited the superseded companion and self records; it now gives the concept DOIs of Paper A (10.5281/zenodo.23817506) and of this manuscript (10.5281/zenodo.23817510).