FAQ
Why are there no raw images or raw headlines here?
Because this artifact is evaluation-first and respects source-asset terms. The benchmark runs on frozen derived arrays and rebuild metadata instead of mirroring restricted raw assets.
Why are there two upload bundles?
Splitting data and code keeps the dataset artifact clean and reduces ambiguity about what counts as the benchmark state versus what counts as execution logic.
What should a reviewer run first?
The code bundle's figure/table reproduction helpers. They operate on frozen derived arrays and are the shortest path to the paper outputs.