Skip to content

The Projection

The Selection gave you the DataDesk promise — brush here, watch it light up everywhere. But “everywhere” quietly meant everywhere the same rows appear. Brush 40 stations and the stations table lights; the cities those stations sit in stay dark — even though the data knows perfectly well which city each station belongs to.

Projection is the edge that promise was missing. Brush the stations, and the cities and regions that contain a selected station light up too. The linked picture stops ending at one kind of thing and reaches the whole related neighbourhood the model already describes.

It travels a relationship you declared — nothing more

Section titled “It travels a relationship you declared — nothing more”

Projection invents no relationships. It reads the ones already in your Typology. A column typed as a foreign key is a declared edge:

# stations entity
city_id: { kind: identifier, subkind: foreign_key, references: cities.city_id }
region_id: { kind: identifier, subkind: foreign_key, references: regions.region_id }

That is a road from stations to cities and to regions — and the only roads projection may travel. The declaration is the permission. A table your selection has no declared edge to simply stays dark. Nothing is guessed; nothing is inferred from a column name that happens to look like a key.

On the air-quality showcase, selecting five monitoring stations lights the five cities and four regions those stations belong to — while countries stay dark, because a station declares an edge to its city and its region, but none to a country. The silence is correct: no declared edge, no permission.

Lit, but not selected — and it always says why

Section titled “Lit, but not selected — and it always says why”

A projected mark must never pretend you selected those things. You selected stations; the cities are lit because of that, not instead of it. So a projected row reads in a deliberately quieter, derived key — and names its origin out loud:

Selected (your cohort)Projected (reached via the cohort)
in a tablea warm amber tint + a solid guttera cool sky tint + a thin, hollow gutter
on a mapa solid amber haloa sky ring on points, a dashed sky outline on areas
the caption”42 stations selected""5 cities · 4 regions via 5 selected stations — related, not selected”

The rule underneath: you can always tell “lit because I selected it” from “lit because something I selected points here.” No mark ever appears without saying where it came from — the same discipline jinflow applies to every keyboard gesture, carried over to marks.

  • In the dimension table. Open the cities table while a stations selection is live, and the cities your stations belong to are marked in the sky key, with the origin line above the table.
  • On the map. Brush stations on a map that also draws regions or cities, and those region areas and city points light in the same derived key — a dashed sky outline on the region polygons, a sky ring on the city points.

Both recompute live from (your selection, the declared edge) — so a shared link reproduces them, and clearing the selection clears them. The projection is never itself put on the bus; each view that cares re-derives it.

  • Read-only. A projection marks; it is never a second selection. Selecting in the cities view is a separate, city-anchored act. The Selection stays anchored on exactly one kind of thing — projection is a way of reading along an edge, not a way of growing the cohort.
  • “Any”, not “all”. A city lights if any of its stations are in your selection. (Cities where every station is selected is a different, later question.)
  • Children light their parents, first. The foreign key lives on the child (stations.city_id), so the first direction is up it: selected stations → their cities. The inverse — brush a city, light its stations — is the same edge travelled the other way, and comes next.

The short version: the model already knows which rows are related — the edges are right there in the Typology — and projection just lets a Selection read along them. Brush the children, watch the parents light, in their own quiet key so you always know why. The conversation with the data stops being one table at a time and becomes the whole neighbourhood at once.

Written by Miss EDA. One anchor, one new verb, declared edges only. 🌊

jazzisnow jinflow is a jazzisnow product
v0.65.0 · built 2026-11-13 10:20 UTC