Analyzing revision lists with an AI agent

Anyone who has worked with revision lists (the poll-tax censuses of the Russian Empire in the 18th–19th centuries) knows they’re a treasure trove. They show not just who lived in our ancestors’ households, but a cross-section of the whole village’s history. It’s especially exciting when you have not just one revision but every one you could get your hands on.

So how do you make them easier to process and mine for insights? I use AI agents: Claude Code or Codex.

Here’s how.

Step one: I transcribe all the revision lists into a Google Sheet. For now I do this part myself, because AI transcription quality is still poor. Here’s a sample page.

Next, I launch Claude Code and ask it something like this:

Your task: identify potential marriages within the village.

To do this:
1. Study the spreadsheet with the revision list data (each revision is on its own tab) at this link: https://docs.google.com/spreadsheets/d/1u19CyiLlNFWfytkvClnLYuP8XqsqHHybXPTDgOYlJHE/edit?gid=0#gid=0
2. For each revision, work out the genealogy of the village for yourself: who is related to whom, and how
3. Then, as you move on to the next year, whenever you find a new marriage, try to figure out whether the bride came from this same village, and if so, from which household
Important:
- names may be spelled differently
- ages and years may be off by ±2–3 years

After my buddy Claude had buzzed away for a while, I got its list of candidates: which household’s son married which bride, and where each bride, listed with no surname, most likely came from.

But we don’t stop there. Next I ask:

Also, think deeply about what other insights into the village's history you can get from the revision lists.
Epidemics? The years when the most people were born?

I could have been more specific here, but I was feeling lazy :)

Claude buzzed even longer and gave me its take on how the village was affected by the Patriotic War of 1812 (Napoleon’s invasion of Russia) and by a cholera epidemic, and on who was the most well-off in the village.

I could have done all this myself, of course, but it’s my least favorite part of the work, and I’m too lazy to do it.

What have you learned from revision lists? What other unexpected things could you ask an AI agent to dig out of this kind of statistical data?