Draw a pelican on a bicycle. With code. No looking.
Measured on Idapt. Every score below has a receipt: click any verdict for the model's actual answer.
Each item asks a TEXT model to produce an SVG drawing. The model never sees what it drew: there is no rendering step in its loop, so the output is a read on spatial reasoning expressed through coordinates.
Nothing is scored. There is no defensible automatic metric for "is this a good pelican", and inventing one would be dressing an opinion up as a number. The SVGs are rendered side by side and you decide.
Contamination: not applicable — nothing is scored, so there is nothing to leak.
These items are famous, so any model trained after they circulated may know the answers directly. Read scores as a timeline of what still bites, not a measure of intelligence.