One hundred years ago, Wittgenstein wrote that the limits of our language are the limits of our world. He was talking about philosophy. But he could have been describing AI transcription in 2026. What the model cannot recognize, the world cannot see. And the words the model cannot recognize are not random. They are the brand names, the proper nouns, the loan words, the jargon, the neologisms. The hardest words to transcribe are the most important ones. That is the gap EarlGrid was built to close.