Public use-case reports with replay evidence, artifact packages, and MCP handoff context. These are the release-facing examples people can inspect before running their own workflow.
The main group-difference result changed completely with the connectivity estimator. A human audit also caught a sign-scoring error that automated review had missed.
Across 36 defensible analysis choices, none of five prespecified connectivity-behavior findings survived correction. Brain Researcher stopped confirmatory promotion and produced a replication plan.
The analysis suggested an mPFC pattern, but the literature was too small and imbalanced for a settled cross-cultural claim. Brain Researcher stopped at an exploratory, paradigm-matched follow-up.
This PDF is an archived snapshot; the manuscript-current Figure 4F result is summarized above.
The frozen workflow had a directional advantage in 47 of 50 same-cohort repeat-level comparisons. Across 116 candidate workflow proposals, 104 parent-run evaluations returned scores and 12 ended in controller transport failure; there was no external replication.
The four-new-source endpoint was inconclusive after Holm correction: 3/4 directions were concordant and Holm p=.39636. The recurring-source panels were directionally concordant in 11/12 cells.
A separate curated report on hubness, decoding, and generative fidelity in high-dimensional connectome matrices. It is not one of the five manuscript cases.