Module request 

2026-09-26: Request bodies for the video legs, and the decoder-unavailable predicate.

Owner: bench, video. Invariants: none beyond the types.

Constants

ORDER_PROMPT

2026-09-26: The question the ordered-color legs ask. It asks only for the color names in order, so score::order_matches can score the reply.

Functions

data_uri

2026-09-26: bytes as a base64 data: URI with type mime.

image_body

2026-09-26: One request carrying a single image: with text_array_body, the two baselines the pad-arithmetic cell subtracts.

is_decoder_unavailable

2026-09-26: Does this error read as “the server cannot decode that container”? Such a leg is skipped, not failed: ffmpeg decoding is a deployment choice (--video-allow-ffmpeg).

mixed_body

2026-09-26: One request carrying an image, then a video, then prompt. The image part is labelled image/png whatever its bytes; the server picks the image decoder from the bytes (model-layers/src/vision_preprocess.rs).

text_array_body

2026-09-26: Text only, as a content array. See image_body.

text_only_body

2026-09-26: The control: prompt alone, as a plain string content.

video_body

2026-09-26: One streamed request carrying a single video, then prompt.