The actor is attending to one point in something large — one event in a long record, one part of a deep structure — and needs its full detail without losing the whole. Focus and context grades detail with distance from attention inside one continuous view: the focused item told in full, its near context summarised, the far context condensed to an outline. Where overview and detail spends a second view to keep the whole present and semantic zoom spends time, showing one level after another, this pattern spends fidelity — everything stays on screen, told unevenly.
What counts as distance varies: position along a sequence, steps through a structure — or, where a model can judge relatedness, how certain a connection is.
Examples
Timeline
On a timeline, detail follows the reader’s attention: the event being read is shown in full, the stretch around it as a summary, and the rest of the record at increasingly coarse levels.
Contextual navigation
When working with a data structure of varying degrees of completeness, it can be useful to see relationships with varying degrees of confidence:
- Explicit connections are authored in the model — parents, siblings, related entities.
- Inferred connections are based on existing data and a formal set of inference rules.
- Deduced connections are proposed by a language model — components that aren’t modelled at all but likely belong, given what is.
Where the timeline puts AI generation to work on the representation side — its coarse rungs are generated tellings — here the model computes nearness where no edge exists, and extends the context past what is modelled.
Resources & references
- Furnas (1986) Generalized fisheye views — the degree-of-interest formalism (an item’s display priority as a function of its importance and its distance from the actor’s focus) that this pattern descends from
- Cockburn, Karlson & Bederson (2009) A review of overview+detail, zooming, and focus+context interfaces
- Wattenberger (2024) Fish-eye — the immediate ancestor of both examples: coarse rungs as generated tellings, the lens following attention
Consequences
- everything stays on screen, told unevenly: the focus in full, its surroundings coarser the further out they stand, and nothing moves as attention shifts
- the actor steers the lens but the system decides what counts as near, so the distance metric — position along a sequence, steps through a structure, a model's confidence — is doing editorial work under the appearance of geometry
- the coarse rungs are summaries something has to produce, and at the rim they may be generated; generated text then sits in a place the actor reads as structure
Related patterns
Enacts
- Agency — the actor steers the lens, but the system decides what counts as close
- Adaptability — the focused frame and its surrounding context recalibrate as the actor's task shifts
- Density — grades density by distance from attention — full detail at the focus, an outline at the rim
Complements
- Progressive disclosure — staged exposure pairs naturally with focus-and-context, revealing detail as the actor moves toward it
- Workspace — focus and context manages detail and overview within a single context; workspaces manage switching between distinct contexts
Alternatives
- Semantic zoom — keeps focus and context simultaneous — detail follows the actor's attention, non-uniformly, inside one continuous view; semantic zoom shows one uniform level at a time under a global scale control
Preceded by
- Embedded intelligence — infers what contextual information is most relevant to the actor's current focus