A corpus — Prize Papers, TNA HCA 32
Seized from ships captured at sea and locked in an Admiralty court for two centuries. We read a slice of them: 140 documents, 644 pages of merchant correspondence, bills of lading, and prize-court appeals bound for Calcutta and Bengal, 1740s to 1810s, in four languages.
What it read, and what it didn't
| Material | Documents | Confidence | Status |
|---|---|---|---|
| English prize-court appeals (printed) | 18 | 0.87 | Reliable |
| Portuguese | 1 | 0.84 | Reliable |
| French East India Company papers | 62 | 0.78 | Reliable |
| New Julfa Armenian merchant letters | 59 | 0.29 | Flagged — needs a scholar |
We ran a second, higher-resolution pass over every low-confidence page. It lifted the English and French readings. It barely moved the Armenian, from 0.21 to 0.25, which is the honest answer: the model cannot reliably read the 18th-century Julfa mercantile hand. So we do not pretend it can.
The find
A coherent run of private correspondence from the New Julfa Armenian trade network — Agha Minas son of Ēliaz in Calcutta writing to agents in Isfahan, Basra, Chandannagar, and Chinsurah. It is exactly the kind of source that rewrites economic history, and exactly the hand our pipeline cannot yet read.
So we did the honest thing. We located and imaged the whole corpus, produced rough machine drafts clearly marked unverified, and offered them, with the tooling, to the scholar who owns this world. A machine that knows its own limits is worth more than one that hides them.
Access
Machine-generated, not human-proofread. The source images remain with The National Archives and are not reproduced here.