The first case. I was eight. London, a family holiday, and somewhere very close to Baker Street. I knew exactly who lived a few streets away, and I asked to go. My parents, reasonable people with a day already planned, said no.

A reasonable answer. Not, to an eight-year-old, an acceptable one.

So I did what Holmes would have done. I kept my own counsel, and I went. Nobody saw me go. One street, then another, then the number itself: 221B. And a small boy standing in front of it for as long as an eight-year-old can stand still, which is not long, but long enough.

Then came the problem every runaway discovers too late: the way back. I worked it out in my head, a cunning detour that would, by pure deduction, deliver me exactly to wherever my parents would be. It had no business working. London is very large, and I was very small. And yet, at the turn of one corner, there they were.

I will not describe their faces. Furious, and frightened, and entitled to be both. I have rarely been happier. I had seen 221B.

Some walks never quite end. That one went on through a lifetime of London, of Holmes, of computers and, later, AI, and this map is where it arrives. The idea, and the thinking behind how it is built, are mine. The machines did the tracing. I just followed the clues, the way I did at eight.

Only what Doyle wrote. Every place name was pulled from the complete canon and looked up within London. Editors’ notes and essays in the source edition were filtered out, so a place only counts if Doyle himself put it in a story. The streets are those that demonstrably existed in 1889 (OpenHistoricalMap), not today’s. The buildings in 3D are traced one by one from the Ordnance Survey 1:2500 sheets of 1869–1880, and Charles Booth’s poverty map of 1889 is warped to fit the same street grid; most streets line up within about five metres.

The vault. Behind the map sits my vault: a personal knowledge base of plain-text notes, kept in Obsidian, that works as a second brain. The stories, the place lists, the sources, every correction and every decision about this map live there as linked notes. The map is, in a sense, one view of that vault.

Searching by meaning. To find the right passage for each place, the canon is cut into short passages and each one is turned into a vector: a list of numbers that captures what the text is about. Those vectors are stored in Qdrant, an open-source vector database, running on a small server at home. Ask it for “a cab rattling through the fog” and it returns the passages that mean that, even when they use none of those words. The same technique searches the vault itself.

The other walk. Whitechapel, autumn 1888, on the same maps, is not a Holmes story: the Whitechapel murders in 3D. It tells where five women were found, and nothing more.

A challenge for the Baker Street Irregulars: some of the 71 names we could not place may be real streets we simply failed to find. Spot one, or spot a place on the wrong corner, and say so in the comments — every correction makes the map better.