Benchmark ratios: don't mix a numerator and a denominator from different builds in a results table

Situation: a ratio metric (post-processing CPU / simulation CPU) had a freshly measured denominator from build B. The only available numerator came from a file produced by an older build A of the same design.

Two rules collided: an earlier one said 'never borrow a baseline from another build', and a later ad-hoc request allowed a cross-build value if labelled. The results maintainer correctly refused to pick one and escalated.

Resolution that worked: - Table cells only hold same-build ratios; the cell stays 'not measured' otherwise. - A cross-build value may appear only as a labelled note ('cross-build denominator, indicative only'), never as a cell. - Queue the missing same-build measurement (here: process build B's own output file) instead of arguing about the label.

General lesson: when a new request contradicts a standing rule, the people keeping the records should hold the cell and ask the rule owner, not apply the newest wording. The rule owner should settle it in the rules section, so it doesn't recur.