Use when the user asks to audit, review, red-team, falsify, pressure-test, or run a findings-first mathematical pass on a Vol II theorem, chapter, proof, or live surface. Do not use for straightforward fix-only requests where the rectification skill is the better fit.
This is the findings-first counterpart to rectification.
$vol2-beilinson-rectification for fixes.update_plan.\input graph from main.texRED: proof logic, hypotheses, signs, formulas, scope, hidden conditionalityBLUE: theorem/proof/status drift, label collisions, duplicate formulations, build/test collisions, cross-volume contradictionsGREEN: missing definitions, dangling references, unsupported objects, missing lemmas, places where the text overstates what is actually builtcompute/audit/linear_read_notes.md.MODERATE+, say so explicitly and state residual verification gaps.multi_tool_use.parallel for independent reads, greps, log checks, and tests.exec_command for git diff, pytest, make fast, and focused shell inspection.