AI read 400 years of old archives overnight and found a meteorite fall missing from every catalog

After reading Breen's dodo post, Waites thought, "I can do this." The first step was asking a Deep Research assistant which historical questions could likely be answered with material already online. It came back with 13 candidates. Next, Waites built a pipeline with Claude Code. Index the documents with embeddings, search by meaning, filter with a small judging model called Jev, have Claude Haiku translate and extract dates and places, then have a Claude Code agent check each find against the original scans and match it with scholarly catalogs of meteorites and volcanoes. jessewaites

Indexing 4.35 million pages of Dutch East India Company records, 5.7 million passages in all, took 12 hours overnight on a single GPU. By Waites's math, a person reading 8 hours a day, 5 days a week, at 2 minutes a page would need about 70 years to get through it. Filtering was cheap too. One search for animal records pulled up 59,000 passages that mention elephants, and having Jev read and sort them cost about $3. For meteorites, the searches used sentences like 'a heavy stone fell from the sky with a loud report,' according to JTPCK's case study page. jessewaites +1

Before the real search began, Waites had the system re-find events that were already known: Breen's dodo record, the 1783 Laki eruption, and the 1815 Tambora eruption. If those came up, the search was working. jessewaites

Three results came out of it. The first is the 1812 meteorite fall on the page shown at the top. That record came from an old newspaper, not the East India Company archives. Waites searched old newspaper articles along with the company records, and the meteorite story ran in one of those papers, the Java Government Gazette. The original page can be viewed on Delpher, the historical newspaper service of the Royal Library of the Netherlands. It is missing from all six meteorite catalogs. If confirmed, it comes 26 years before the earliest meteorite fall on record so far in Maharashtra, India, where Cokurrgaum is located. jessewaites +1

The second is three Javan rhinos sent as gifts to the king of Kandy in Sri Lanka in 1738–1740. One was shipped in 1738 and two went aboard the Loverendaal in 1740, and all of them died before arriving. This shipment did not appear in any of the rhino record lists Waites checked. The third is volcanic eruptions. After filtering 769 eruption reports, 3 eruptions remained that are not in the Smithsonian's volcano catalog: Gamkonora in February 1722, Ciremai in early 1712, and Slamet in February 1780. The original rhino and volcano records are East India Company documents held by the National Archives of the Netherlands. jessewaites

All three results are still candidates that no expert has reviewed. Waites has reached out to a meteorite curator, rhino researcher Kees Rookmaaker, and volcanologists, and is waiting to hear back. Waites also tried to pin down the great eruption of 1808–09, whose source volcano is still unknown. But even after the system read all 633 articles in Batavia's (now Jakarta) first newspaper, nothing relevant turned up. jessewaites

While preparing the dodo chapter of a book on animal extinction, Breen had an Opus 5.5 agent search the East India Company records in GLOBALISE. GLOBALISE is a project in which Dutch research institutions use AI to transcribe the company's handwritten documents and publish them for free. Breen also used GPT-6. As Breen explains it, agents like Opus 5.5 spin up several copies of themselves to divide up sources written in multiple languages, and given an API key, they run embedding searches on their own. On this project the agents multiplied to about a dozen. globalise.huygens.knaw.nl +1

The passage the agent found was in the April 1615 journal of the ship Wapen van Amsterdam. The crew wrote that on Mauritius they "caught many tortoises, dodos [dodeersen], and some geese and parrots". The original is among the East India Company documents at the National Archives of the Netherlands. The timeline of dodo sightings had been blank for 1611–16. Breen sees this passage as a new record for that gap and wrote, "I do think that this is a publishable result." In the comments, Breen added that, as far as they knew, the April 1615 passage had never been reported before. resobscura.substack +1

The agent also reread known records in the original. In a 1638 document, 'velthoenderen' (a bird that appears to be the red rail) had been rendered as 'perdrix' (partridge) in an 1890 translation by a French scholar. Opus 5.5 went back to the original manuscript and corrected that translation. Neither the 1615 passage nor this correction has been formally reviewed by specialists in the field yet. resobscura.substack

This case is not about searching a pile of documents. It is about reading one document that no one could read. Church gave GPT-6 Astra a single scan of a cipher letter addressed to General Marmont, published in a French military journal in 1969, and ran it for about six hours. Working from that one scan alone, the agent handled everything from transcribing the symbols to breaking the cipher. Church wrote, "I spent six hours of model time and a few evenings of my own". carter.church

The letter is made up of 1,300 cipher units using 155 distinct symbols. Only 33 symbol meanings had been published before, and those covered 435 of the 1,300 units. The agent worked out the remaining two-thirds or so from the letter itself. It produced the full text of the letter in French and English, and corrected the date, which had been misprinted in the 1969 reprint, to late March 1809. carter.church

Church also published a solution package with the key (key.tsv), the transcribed ciphertext, and verification scripts (verify.py and reproduce.py), so anyone can rerun it and check. The Israeli outlet Calcalist reported that Satoshi Tomokiyo, who runs the cipher research site Cryptiana, has marked the cipher as "solved." carter.church +1