Explicit Interoceptive Modulation versus Direct State Conditioning: Amended Stage 1 Registered Report Protocol in a Closed-Loop Cognitive Architecture (BICCA R4.2/A2)

This record archives protocol revision R4.2/A2 of BICCA, a pre-confirmatory Stage 1 Registered Report proposal comparing explicit bounded interoceptive modulation with expressive direct nonlinear state conditioning in matched closed-loop recurrent architectures. The two model classes receive identical observed variables and are matched for recurrent-state width, trainable-parameter count, training data, update count, and dense matrix multiplications. Remaining differences in elementwise operations and cached values are explicitly reported. Amendment A2 corrects a mismatch confined to the prospective S2 power analysis: exact two-test Holm adjustment is the primary multiplicity rule, while the former requirement that both unadjusted p-values be below .025 is retained as a named conservative sensitivity analysis. The amendment does not change the primary estimand, confirmatory sample size (N = 80), smallest effect size of interest, or ordered decision matrix. The archived package contains the manuscript PDF and editable source, executable code, tests, locked confirmatory configuration, prospective power-analysis outputs, amendment and deviation records, internal quality-assurance reports, environment records, and SHA-256 manifests. Scientific status: PRE-CONFIRMATORY. Confirmatory seeds remain locked, no confirmatory results are included, and the documented internal G3 checks do not constitute independent replication, peer review, journal registration, or in-principle acceptance. External clean-room verification remains open. This work tests a finite functional distinction between computational architectures. It does not provide evidence of phenomenal consciousness, qualia, semantic understanding, biological implementation, or the universal necessity of either architecture.