Each brandbook was graded independently by Claude (8 parallel grader agents) and Codex (gpt-6-astra, 8 batches) on the same rubric. Scores run 1–10 per dimension, up to 120 in total. Every book was graded against the current positioning: togetherness, fire, sauna, the seasons, local makers, company retreats and scale across Europe. Books 01–12 were briefed earlier on a narrower brief.
Sorted by the mean of both totals. The gap column shows where the graders disagree on the overall verdict.
Hover or focus a cell to read both graders' notes. In the gap view, blue means Claude scored higher and red means Codex scored higher.
The highest mean score on each dimension. It shows which direction to borrow each part from when merging.