The Selection
Classic EDA asks you to listen to the data. jinflow lets you talk back — and the sentence you speak is a Selection. Point at some things; every view in the app answers. This is DataDesk’s old promise — brush here, watch it light up everywhere — made ambient across the whole product.
A Selection is a cohort of one kind of thing
Section titled “A Selection is a cohort of one kind of thing”A Selection is a set of instances of one entity — 40 materials, three monitoring sites, a region’s worth of parcels. Not a bag of unrelated ids: these materials, or those stations. The entity is the grain, and it’s why a Selection anchors on exactly one entity type — “add” and “remove” only mean something when everything in the set is the same kind of thing.
An entity, worth remembering, need not exist in your source data. A City built from station coordinates, a MaterialHealth rolled up from signals — these are first-class subjects you can select, even though no source system ever named them.
And a Selection is really a filter, not a frozen list. When it can be, it holds the predicate that produced it — “parcels where land-use is arable”, “the box you drew” — so it re-resolves against fresh data instead of pinning yesterday’s ids. That’s what lets a Selection be saved, reapplied, and eventually promoted (below).
Two tracks: Focus and Selection
Section titled “Two tracks: Focus and Selection”There are two different things you do when you point at data, and jinflow keeps them apart:
| Selection | Focus | |
|---|---|---|
| how many | a cohort — 0..N | the one — 0 or 1 |
| lasts | until you clear it | until you look elsewhere |
| drives | the linked views — diagrams rescope, the summary card, filters | the inspection surface — the Portrait, the halo, the “full record →” |
They’re independent: focusing one member doesn’t add it to the set, and building a set doesn’t focus anything. You can sweep a cohort and inspect one member of it without either gesture clobbering the other.
The gestures — one grammar, everywhere
Section titled “The gestures — one grammar, everywhere”The same grammar works on a map, a scatter plot, a box plot, a histogram:
- Plain click → focus the thing you clicked (opens its Portrait, lights its halo). One at a time.
- ⌘-click → add it to the Selection. ⌥-click → remove it.
- Drag a box (shift-drag) → select everything inside; ⌘/⌥-drag adds or removes the enclosed. On a histogram, dragging a bar selects that value range; on a box plot, a box selects that group.
⌘ is the add-modifier on a Mac (Ctrl on Windows/Linux). The gesture reads the same whether you’re on the map or in a diagram — that’s the point.
Assembling a Selection across the pages
Section titled “Assembling a Selection across the pages”This is the part that feels like a superpower once you have it: a Selection travels with you. It lives in an ambient bus that survives navigation, and a small summary card is present on every page, so you always see what you’re holding.
So you can build a cohort incrementally, across the app: brush a cluster on a scatter plot, walk to another view, ⌘-click a few more, drop one with ⌥ — and the set accumulates the whole way. Every linked view you pass through re-scopes to it live: the diagrams rescope, the summary card updates its count, the entity table filters to just those rows.
Two honest boundaries:
- One entity at a time. The bus holds one cohort of one entity. Point at a different kind of thing and the Selection starts fresh — a mixed-entity basket isn’t a defined idea (yet).
- The data table receives; it doesn’t source. You assemble on the exploratory surfaces — charts, the map, the drill floor, tiles — and the dimension table is a destination: hand it a Selection and it shows those rows, scoped.
What a Selection is for
Section titled “What a Selection is for”- Scope everything you look at. Carry it to a dimension and the whole page is about those members. Carry it to Compare and you’re comparing that cohort.
- Save it. A Selection can be named and kept — a Selector, a reusable, re-resolving view of an entity (“the arable parcels”, “the implant materials”).
- Promote it to a standing signal. This is the frontier — the First Citizen. A Selection that carries a real predicate (a range, a category, a box) isn’t a snapshot; it re-resolves. So it can graduate from a thing you brushed into a signal the system watches — the cohort becomes a diagnostic that fires whenever the data drifts into it. A brushed cohort of frozen ids can’t do that; a predicate-backed one can. That difference is why jinflow models the filter, not the gesture.
The short version: you point, the whole app answers, and the pointing accumulates. Focus lets you inspect one thing; Selection lets you carry a cohort across every view and, when it’s a real filter, hand it to the system as something worth watching. That’s what it means, in jinflow, to talk back to the data.