How this works

Locus takes one point on a map and shows you the historical records closest to it. Here is what that actually means.

What you give it

A pin. Click the map, or drag the pin you already placed. That is the whole input — no name, no date, no account, nothing to fill in. Locus does not ask your browser for your location and cannot: the page ships a policy that refuses the browser location API outright.

What it does with it

The coordinate goes to GeoHistory, the open dataset and read-only API underneath Locus, as a single request asking one narrow question: which events in the record are physically nearest to this point. GeoHistory owns the history and the distance maths. Locus owns the map, the search policy, and everything you read on the page.

The coordinate travels in the body of that request rather than in the address, which is why you will never see it in the URL bar, in your history, or in a shared link. That is a deliberate design decision and the Privacy page explains what it buys you.

The radius ladder

Locus asks for a tight circle first and widens only if it has to, stopping at the first radius that can fill a list of twelve.

Rung What it is for
5 km A dense city centre. Where the record is thick, this is the whole search.
15 km A metro area and its immediate surroundings.
50 km A small town's region. Most towns need this rung, and that is normal.
150 km Genuinely rural country. The last rung; Locus does not widen past it.

The radius actually used is always stated above the list, and when Locus had to widen it says so in plain words rather than quietly pretending the first circle worked. If the last rung still finds fewer than twelve records, you get the few that exist and a sentence saying how few. If it finds none, you get told where it looked.

The ladder lives in Locus, not in GeoHistory. The API takes one radius per call and answers honestly; walking the rungs is a product decision, and it belongs on this side of the line so a second client can make a different one.

Chosen by distance, read by date

This is the part that is easy to get backwards. The twelve records are selected by how close they are, and then displayed oldest first. So the list you read is a small chronology of one place, not a ranking — record five may well sit closer to your pin than record one.

The numbers on the map markers follow the reading order for exactly that reason. They are positions in a list, not a relevance rank, because proximity order would imply the engine computed an importance it never computed. Click a marker to jump to its card.

Ties never wobble

Two calls with the same pin return the same records in the same order, for a given build of the dataset. Equal distance is broken by date, and an equal date by the record's own identifier. Nothing is sampled and nothing is random.

What is left out, and why

Births and deaths

By default Locus asks GeoHistory to exclude them. Measured against the dataset, 85–94% of the records nearest a dense city centre are births and deaths of minor figures; a list of the twelve closest records without that filter answers "who was born or died here?" instead of "what happened here?" The people are in the record and a later view may surface them deliberately. They are not allowed to crowd out the answer you came for.

Events that were not about a place

A handful of records in GeoHistory are marked as reaching everyone alive rather than anywhere in particular. Circa shows those on purpose, because a timeline of a life should carry them. In a distance-first list they are noise — an event with no meaningful location cannot be near you — so Locus leaves them out.

Records pinned to a country's centre

Some events are placed by a fallback claim rather than their own coordinate, which can put a national treaty three kilometres from your pin because that is where the country's centre happens to sit. Locus asks for directly-placed records only. On the current dataset build that filter turns out to change very little, which is worth saying plainly rather than implying it is doing heavy lifting.

Why a card sometimes carries only a year

Locus keeps whatever precision the record actually holds instead of filling in the rest. A card reading 1859 means the year is known and the day is not. It is a smaller claim, made honestly, and it is a better answer than a confident date nobody can support.

Every card links to its source

The event data is reused under CC BY-SA, so attribution is not decoration. Locus fails closed on it: if a response arrives containing a record with no usable source link, the whole response is rejected rather than rendered without credit. You will never see an unattributed record here, because the code will not draw one.

What it is not

Locus is not a local-history archive and it is not a search engine. It reports what one open dataset happens to hold near a point. The record is broad rather than complete, and it thins out the further back you go and the further you get from large cities.

Two honest failure modes worth knowing about. Out at the wide rungs, results can come back almost entirely as town and municipality foundings, because that is what the record holds at that scale. And open ocean returns nothing at all — correctly, which is why the empty state says where it looked instead of claiming something broke.

More questions · Where the data comes from