The Orrery's Missing Moon: On the Error That Crawled the World

In 1790, the Scottish engineer and polymath James Watt—better known for his steam engines—completed a curious, lesser-known project: an orrery. This was no ordinary model of the solar system. Commissioned for the University of Glasgow, it was a masterpiece of brass and clockwork, a crawling, ticking map of celestial mechanics meant to instruct and inspire. But Watt, in his scrupulous effort to base it on the best astronomical data of his age, embedded a fascinating flaw. He included the four newly discovered moons of Jupiter, but his source, a contemporary astronomical table, listed only three. So, Watt's magnificent machine, this physical sitemap of the heavens, indexed a moon that did not exist.

This is more than a historical footnote. It’s a parable of discovery and the inherent risks of automated collection. Watt’s orrery was, in a very literal sense, a crawler. Its gears were programmed by data, its movements dictated by the authoritative sources of the day. It didn’t question; it executed. It faithfully crawled the available ‘index’—the flawed table—and rendered its contents as tangible truth. The error wasn’t born of Watt’s ignorance, but of his system’s perfect, mechanistic trust. The crawler had no capacity for fact-checking; its job was to map what it was given.

The Ghost in the Celestial Machine

For years, the orrery displayed its phantom moon to students and professors. The mistake was eventually corrected, but the process was telling. It wasn’t an algorithm that found the discrepancy. It was human curiosity—a later astronomer, cross-referencing observations, noticing a celestial body out of place in the real sky compared to the model. The ‘crawl’ had been flawless, but the ‘index’ was wrong. The discovery happened not within the automated system, but in the space between its representation and the reality it sought to model.

We build our own orreries every day. Our crawlers traverse the web’s vastness, guided by sitemaps and links, constructing models of a digital universe from the data they collect. They operate on faith in protocols and structures. But what happens when an authoritative source, a primary ‘sitemap’ of information, contains a ghost? A broken link presented as valid, a page of misinformation elegantly interlinked, a piece of data replicated across domains until it gains the sheen of truth—these are our phantom moons. The crawler, efficient and unbiased, will gladly incorporate them, giving them weight and presence in the indexed model of the world.

Watt’s orrery teaches us that discovery is not merely the act of crawling and collecting. It is, ultimately, an act of skepticism. The map is not the territory. The index is not the truth. The most elegant, mechanically perfect system for gathering information can still propagate an error with impeccable precision. Our task, then, is not just to build better crawlers, but to remain the curious astronomer in the room—to always look up from the ticking model and check the actual sky.

Notes & further reading

A few pages I came back to while writing this: