Context
The research record preserves failed approaches when they expose a model boundary, an invalid transfer, or an unsupported strengthening.
Proof or evidence
RPR-D12 has a verified common-DAG receipt and substantially lowers additions, scalar gates, fan-out, and copy count. RPR-D11 remains preferable in raw dependency depth, one scalar of declared storage, and rounded-path length.
Verification notes
The failure status and preserved lesson are recorded in the cumulative failure ledger.
Limitations
- retain the Pareto set; do not silently select one realization or merge their best coordinates.