Nested Bubble/Gear (NBG-01) v0.4.1: E2-B Canonical Execution and Deterministic Replay

Nested Bubble/Gear (NBG-01) v0.4.1 reports the canonical execution and deterministic replay of Experiment E2-B: Hidden-Microstructure Challenge, following the publicly timestamped NBG-01 v0.4.0 preregistration and pre-run freeze. E2-B extends the E2-A transition system with preregistered 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 experiment contains 256 parameter points. The 64 points with gamma = 0 form a complete continuity slice against the previously published E2-A experiment, while 192 positive-gamma points constitute the hidden-microstructure challenge. The canonical execution was performed from the publicly frozen v0.4.0 harness without changing the observer definitions, parameter grid, thresholds, refinement rules, witness rules, serialization convention, mixing definition, trust criteria, cost rules, or anti-rescue policy. All 64 gamma = 0 continuity points matched the corresponding E2-A parent kernels under both the frozen implementation-digest and semantic-digest checks. The identity observer O3 passed all 256 of 256 parameter points. On all 192 positive-gamma hidden-microstructure challenge points, the static observers O0, O1, O_IF, and O2 fail predictive closure. Thus, the previously successful E2-A observer hierarchy does not retain sufficient information once the hidden-bit coupling is activated. The preregistered governed refinement procedure RSTAR was then evaluated without human-added post-hoc features. RSTAR begins from the O0 partition and performs deterministic, exact-rational, split-only stable partition refinement using transition mass into the current partition blocks. RSTAR recovered exact one-step and two-step predictive closure and composition at all 192 positive-gamma challenge points without reaching the 64-state identity partition. The resulting refined partitions contained: 16 classes at 96 challenge points, corresponding to a 2.0-bit compression gain relative to identity; 36 classes at 96 challenge points, corresponding to approximately 0.830075 bits of compression gain. RSTAR converged in one split round across the complete positive-gamma challenge set. Predictive closure alone did not guarantee sufficient metastable persistence. Under the preregistered trust-separation gate, 53 of 192 challenge points receive PROMOTE_RSTAR, while 139 of 192 receive FAIL-TRUST-SEPARATION. No positive-gamma challenge point required promotion to the 64-state identity observer. Following the canonical execution, a clean same-host isolated deterministic replay was performed from the identical frozen source. All declared deterministic scientific artifacts reproduced byte-for-byte, yielding: REPLAY_EXACT The overall canonical execution status is: E2B_CANONICAL_EXECUTION_COMPLETE During the first execution attempt, the surrounding execution environment terminated the process at its wall-clock limit before the harness completed its final export stage. No partial scientific output from that interrupted attempt was inspected or used. The canonical experiment was then restarted from a fresh copy of the identical frozen source, without modification, and allowed to complete. This execution-environment interruption did not alter the frozen scientific protocol. The replay is a same-host deterministic reproduction and is not presented as an external independent-laboratory replication. NBG-01 v0.4.1 does not establish a new physical law and does not make claims about spacetime, quantum mechanics, consciousness, or nature's literal ontology. It reports the outcome of a finite preregistered computational challenge testing whether governed partition refinement can recover compressed predictive structure when transition-relevant information is deliberately hidden from the static observer hierarchy. This release closes the E2-B experiment. Subsequent external-scale or cross-domain challenges should be preregistered separately rather than modifying or rescuing the frozen E2-B result.