Why this page exists. A fair objection to any hand scored dataset is that the scorer saw what they wanted to see. The defence is reproducibility: if a stranger with no stake, using the same written rubric, lands on or next to the same numbers, the scoring is doing its job. Score the photograph below on all six features, then reveal how the project scored the same image.
The scale, as written in the project catalogue: 0 = absent, 1 = minor / partial / unclear, 2 = present, 3 = prominent. Judge only what is visible in the photograph.
The project's scores were assigned by one scorer, image by image, against the same rubric shown above; the full matrix and methodology are documented in the project knowledgebase. This page is a reproducibility demonstration, not an inter rater study: the site is static and stores nothing, so your scores stay on your device and are discarded when you leave. Exact agreement on every feature is not the expectation; the rubric's own middle grade is "minor / partial / unclear", and landing within one point is how two careful scorers of the same photograph typically relate.