Auditing policy dependent benchmarks in allocation simulations

When an allocation policy changes which agents remain in a queue, a benchmark built from the best agent present at each decision changes with the policy. A ratio against that benchmark can then rise while total value falls, or fall while it rises. In a two-period housing example, the moving-benchmark ratio rises from 0.80 to 1.00 while realised value falls from 16 to 11. In a deadline queue with an exact offline optimum, value rises from 19 to 27 while the ratio falls from 1.00 to 0.90. I provide a routine that reports numerator and denominator changes against a fixed reference. This note has been merged into the paper When the Benchmark Moves with the Policy (version 7.0 of the Execution-Certainty Wedge record), which supersedes it. Version 1.0.3 adds, in policyspace2_audit/, the instrumentation patch, analysis script and per-seed results of the PolicySpace2 audit reported in When the Benchmark Moves with the Policy; the earlier files are unchanged. Revision details are in revision_notes.md.