A linotype shop’s repair sentence must be multiplied by weight matrices inside a language model. What conversion makes that text neural-network ready?

A linotype shop’s repair sentence must be tokenized and embedded as vectors before weight matrices inside a language model can multiply it. Leaving the sentence as a raw PDF byte blob without tokenization is not neural-network ready text. Feeding the printer’s street address as the only feature ignores the repair sentence. Converting the sentence into a spectrogram treats text as if it were audio and skips the language interface. Exam caveat: text readiness is tokens plus embeddings even in multimodal civic shops. Operational check: run the repair sentence through the model’s tokenizer and embedding table; verify tensor shapes match the first language-block input before any matrix multiply. Assert tokenizer vocabulary coverage on linotype repair jargon before training. Linotype repair text becomes matrix-ready only after tokenization and embedding—not as PDF bytes or spectrograms. Gate linotype language training on a tokenizer smoke test that embeds a sample repair sentence and asserts the expected tensor rank and vocabulary hits on shop terms. Close q109 with a municipal evidence note that records why the correct mechanism fits this civic multimodal scenario, why each of the three distractors fails by a distinct concept, and which operational check was observed in logs or dataset cards before promotion. File that note with the run packet so later maintenance cannot silently revive a rejected framing.