Steps / Historical AI

The reading room

The actual pages. Read them yourself.

These are the machine transcriptions, page by page, exactly as produced — with their confidence, the source scan, and every equation the checker flagged. Nothing is smoothed over. Where the machine could not read, it says so.

Leonhard Euler · 1748 · public domain

Introductio in analysin infinitorum

Read + verified

A 250-year-old error, caught at the 113th digit. Euler prints π to 127 digits. The validator flagged the 113th — de Lagny's 1719 hand-computation mistake, reproduced in print and unnoticed for decades.

320 leaves·4,397 expressions·Latin + LaTeX
Read the transcription →
Leonhard Euler · 1770 (1828 English ed.) · public domain

Elements of Algebra

Read + verified

Euler's own imaginary-number mistake. He writes √(−2)·√(−3) = √6 (it is −√6). A real author error, famous in the history of complex numbers, flagged by the numeric checker with no prompting.

627 leaves·20,431 expressions·English + LaTeX
Read the transcription →
Leonhard Euler · 1744 · public domain

Methodus inveniendi lineas curvas

Honest limit

Where the checker honestly finds nothing. The founding text of the calculus of variations transcribes cleanly, but its equalities are differential relations, not universal identities. The validator can't check those and says so, instead of inventing an error.

336 leaves·5,600+ expressions·Latin + LaTeX
Read the transcription →
The National Archives (UK) · HCA 32 · 1740s–1810s

Prize Papers — Calcutta & Bengal

Read + one refusal

Three languages, and an honest refusal. English prize-court appeals and French East India Company letters read in the original at high confidence. The New Julfa Armenian mercantile hand is not machine-readable, so those letters get a name/date index, not a fabricated transcription.

140 documents·644 pages·English · French · Armenian
Read the transcriptions →
On rights & sources

The Euler volumes are public-domain scans from the Internet Archive, linked page by page. The Prize Papers text is transcribed from documents held by The National Archives (UK); the source images remain theirs and are not reproduced here — each document links to the Prize Papers portal. The verification tool that produced the math findings is open source: github.com/steps-re/reading-the-archive.

Know a corpus that should be read this way? mike@stepsventures.com · How the reading is verified →