The missing verses, laid side by side
His most repeated argument was that the same verse does not say the same thing in every Bible, and that some verses are not in every Bible at all. He made it by holding editions up to a camera. This does it to the whole corpus at once: every edition in the library, OCR'd, aligned verse by verse, and diffed.
The engine runs on hardware in this office, not on a paid API — which matters, because the point of the exercise is that anyone with the books and a machine can check the claim themselves.
This page is wired and waiting. The Missing Verse engine runs on a local three-node cluster: it OCRs the scans in the library, aligns them verse by verse into a single corpus, embeds each reading so equivalent verses match across translations, and asks a 397-billion-parameter model resident on the cluster to describe what differs. Nothing is published here until the engine emits it. There is no placeholder text below this line and no example diff — an empty page is the honest state of an empty corpus.
How to read a diff
Each verse shows every edition the corpus holds, with the reading it carries. A filled marker means the verse is present, a hollow one means it is not there at all, a half-filled one means it is present but reads differently. Where a passage is absent, the row says so in words rather than leaving a gap that could be mistaken for missing data.
Machine readings are fenced and graded. The cluster's language model writes a short description of what differs. That prose is boxed, labelled with the model that wrote it, and marked Unverified — it is a lead for a human to check against a critical apparatus, not a finding. The underlying verse text is not machine-written: it is OCR of the scans, and OCR errors are possible and are corrected on report.