Nested Bubble/Gear (NBG-01) v0.4.0: E2-B Hidden-Microstructure Challenge Preregistration and Pre-Run Freeze

Nested Bubble/Gear (NBG-01) v0.4.0 is the preregistration and pre-run freeze release for Experiment E2-B: Hidden-Microstructure Challenge, following the completed NBG-01 E2-A interface-aware challenge published in v0.3.1. E2-B extends the previously frozen interface-aware dynamics by introducing an additional hidden-microstructure coupling through designated internal bits x1 and x6: q_A = (1 - kappa - delta - gamma)m_A + kappa m_B + delta x3 + gamma x1 q_B = (1 - kappa - delta - gamma)m_B + kappa m_A + delta x4 + gamma x6 with p_A = eta + (1 - 2 eta)q_A p_B = eta + (1 - 2 eta)q_B. The frozen parameter grid contains four preregistered values each of eta, kappa, delta, and gamma, for a total of 256 parameter points. The 64 points with gamma = 0 form a complete continuity slice that must reproduce the previously published E2-A v0.3.1 dynamics and scientific behavior before any hidden-microstructure result is interpreted. The remaining 192 points with positive gamma constitute the genuine E2-B challenge. The static observer ladder remains frozen from the prior lineage and includes O0, O1, O_IF, O2, and identity observer O3. The hidden variables introduced through x1 and x6 are intentionally not added manually to these static observers after the experiment begins. E2-B additionally preregisters a governed refinement candidate, RSTAR, designed to test whether predictive closure can be recovered through deterministic partition refinement rather than through post-hoc human feature selection. RSTAR begins from the O0 partition and performs deterministic, split-only, exact-rational stable partition refinement using only transition mass into the currently defined partition blocks. It may refine existing observer classes but may not merge them, inspect future experimental outcomes to invent new features, or receive manually selected hidden variables after canonical output becomes visible. RSTAR is promotable only if it produces a predictive partition with fewer than 64 classes and independently passes the frozen predictive-closure, composition, mixing, and trust-separation gates. Reaching the 64-state identity partition does not count as successful compressed observer recovery. The preregistration preserves the interoperability rules established during the E2-A sequence: canonical MSB-first state serialization, deterministic witness tie handling with equivalence-aware interpretation, explicit separation of exact and near rank deficiency, direct survival-conditioned total-variation contraction for conditional mixing, deterministic artifact generation, semantic and implementation digest rules, and mandatory witness materialization for failure outcomes. All observer definitions, parameter grids, thresholds, cost rules, trust criteria, refinement rules, witness rules, terminal outcomes, baseline checks, serialization conventions, mixing definitions, and anti-rescue policies are frozen before canonical E2-B output is inspected. No E2-B scientific results are included in this release. The purpose of NBG-01 v0.4.0 is to create a public, timestamped experimental contract before execution. Following publication, the frozen E2-B harness is intended to be executed once, followed by an isolated deterministic replay. The resulting canonical receipts, witnesses, refinement traces, continuity checks, and replay comparison will be preserved for a subsequent results release. NBG-01 v0.4.0 does not establish a new physical law and does not make claims about spacetime, quantum mechanics, consciousness, or nature's literal ontology. It defines a finite falsifiable computational experiment testing whether governed observer refinement can recover compressed predictive structure when transition-relevant information is deliberately hidden from the existing observer ladder.