The collection
Each is a body of archival material we have transcribed and verified, or are working on with the people who own it. Every reading carries its confidence. Nothing here pretends to be finished when it isn't.
A full 1748 volume: Latin prose and thousands of mathematical expressions, each checked against mathematical truth. The validator flagged Euler's printed π at the 113th digit — de Lagny's 1719 error, reproduced and long unnoticed.
The complete algebra text, 20,431 expressions checked with zero truncation. In the chapter on imaginary quantities the checker flags Euler's own mistake, √(−2)·√(−3) = √6, with no prompting.
The founding text of the calculus of variations. It transcribes cleanly, but its equalities are differential relations, not identities — so the checker reports an honest limit rather than inventing errors. We publish it to show that boundary.
140 documents, 644 pages seized from ships captured at sea. English prize-court appeals and French East India Company papers read reliably; the New Julfa Armenian merchant letters get a name/date index, not a fabricated reading.
The judicial record of the Calcutta Supreme Court, transcribed and reunited as a modern digital edition, alongside a reconstruction of Hicky's Bengal Gazette, the first newspaper printed in the subcontinent.
In collaboration with the historian Andrew Otis
Enciphered diplomatic manuscripts. Accurate capture of the cipher glyphs is the step that unblocks decipherment, and the step ordinary OCR fails. Early, honest, unfinished.
Two centuries of Company correspondence from Asia to Amsterdam. GLOBALISE transcribed the handwriting and released it under CC0; nobody had translated it. Every inventory is now in English with each passage anchored to the scan it came from, published with its gaps listed rather than quietly patched.
Afghanistan's first Persian-language newspapers, lithographed in nasta'liq — a dense script ordinary OCR reads as a picture, not text. Machine-read and scored per page, with historians supplying human transcriptions so accuracy can be measured before anything is called finished.
Hundreds of thousands of imaged but un-transcribed pages — the correspondence of Newton, Boyle, Halley and hundreds more. English scientific prose, the founding record of modern science.
Math in manuscript. Reading the hand is the open problem, but the render-and-validate stack transfers unchanged once the LaTeX exists.
An 18th-century apparatus drawing becomes a structured, editable graph of components and connections, not just a caption.
Keep an archive, study one, or want a corpus read? mike@stepsventures.com · How the reading is verified →