Multi-agent review/refine loop. Use after a change or milestone, or when asked to review work: dispatch a fresh-context full-spectrum reviewer plus a dissimilar correctness reviewer, consolidate and fix, add a regression test for every defect, and repeat until a round is clean.
Assess the QUALITY (tightness/optimality, not correctness) of a generated Holoso schedule and write a crisp defect report: wasted cycles such as an op/select/install/branch scheduled later than its operands or condition allow, or a cascade of such single-cycle delays. The analysis is done strictly from first principles over the schedule's final state, NEVER by calling the compiler's own timing helpers (which would make the audit blind to bugs in that very timing). Use during the review/refine loop.
Cross-check that the emitted Verilog RTL and the HTML schedule report describe the SAME schedule for each bundled example: every operator issue and landing present in one artifact must be present in the other, at a consistent clock and on the same registers and operands. Two independent max-thinking agents verify in opposite directions (RTL against HTML, and HTML against RTL). The check reads ONLY the two emitted text artifacts and NEVER uses any compiler internals, so a divergence between the two backends cannot hide behind shared code. Use during the review/refine loop.