No published ablation in the 2,824 papers we read varies the sandbox layer, and the budget and governance layers are barely touched. The paper's open-problems section proposes one small controlled study: one harness, one task suite, one bound varied at a few levels.

Wanted. A harness team or a researcher with compute to co-design and run it: pick the harness and suite, pre-register the design on OSF (template: docs/tier3_ablation_protocol.md), run, report. Joint authorship.