Each case's id, title, and input.
Everything the model is told: system text, prompt
templates, schema and tool descriptions (instructionParts in
@crowandraven/evalkeeper/ai-sdk gives these).
Other lengths than IN_INSTRUCTIONS_MIN_CHARS and
IN_INSTRUCTIONS_MIN_WHOLE.
The cases found, each with its longest matching words.
The cases whose input text appears in the instructions: all of one of its texts (when at least
minWholecharacters), or a run of whole words at leastminCharslong, after evening out case and punctuation. Use it to warn, not to fail: an example may belong in the instructions, and then the case should be rewritten or retired.