Prerequisites
A collection has grown too large for one view, such as an archive, a backlog, a library, or a record of runs, and the actor's question is not yet specific enough to write as a query. No pattern here settles this.
An actor is working with a collection too large to take in through one view. They do not yet know exactly what they are looking for, so they cannot simply search or filter for it.
Choosing how to show the collection
Choose how detail and the overview share the screen
Choose for each surface; the options can be combined. Each keeps the overview available at a different cost: a second view takes space, a scale control takes time, a graded view loses detail at the edges, and a generated summary has to be checked. In every case, open on the overview and show detail on request. Skip this step when the work is discrete items moving through a defined process, where the item and the queue already give enough structure.
Combine any of:
Split the view — Overview and detail
Show the overview and the detail as two linked views. The splitting steps of composing views continue from here.
Change the representation with scale — Semantic zoom
Show one level at a time under a scale control, with a few named levels such as year, month, and day. Decide where the thresholds between levels sit and how the change between them looks.
Show full detail only where the actor is looking — Focus and context
Show one continuous view with full detail at the focus and a reduced outline around it. Decide what produces the outline, and label it if it is generated.
Generate summaries at several levels — Text lens
For prose, generate the levels, such as heading, abstract, and full text, instead of writing them in advance. For a whole collection, generate an overview that helps the actor gather items rather than judge them, and link every claim in it to its source.
Show the difference directly — Comparing · optional
Give the comparison its own representation, side by side, overlaid, or with the difference drawn, so the actor does not have to keep one view in memory while looking at the other. Build it when the actor asks for it, within whichever option is in use. Skip this step when the questions are about items rather than differences.
Marking items while reading
Applies when
The actor reads, scans, and sorts the collection. Each pass produces judgements, and they are lost when the session ends unless the actor can record them. Marking has to take almost no effort, because a mark the actor postpones is usually never made.
Let the actor mark items where they are looking
Both kinds of mark, in either order, follow the same rules: made in one action where the actor is looking, private by default, and drawn from a flat vocabulary that grows with use. Do not ask for structure up front.
In any order:
Annotate in place — Annotation
Let the actor highlight and add notes to content they cannot edit. Keep the marks from obscuring the content as they accumulate.
Label recurring themes — Tag
Let the actor apply reusable labels, several per item if needed, and accept near-synonyms while the vocabulary settles. Each label becomes a facet to filter by.
Offer structure once the marks show it — Incremental formalisation
When marks have accumulated, offer to add structure: merge near-synonyms, confirm the vocabulary, turn a recurring label into a field, or turn list-like prose into a list. Offer these changes rather than applying them automatically, and make them reversible.
Share selected marks with others — Commenting · optional
For the few marks that need a reply, rewrite the note and its anchor so that someone else can find and understand them; making private marks visible as they are is not enough. A shared mark becomes a thread others can reply to. Skip this step when the actor is the only reader.
Saving useful views
Applies when
When a view answers a question that comes up again, the actor rebuilds it by hand each time, because it was not saved. Composing views builds the view itself, and the labels provide its facets. The steps here decide what is saved, for whom, and how much of the judgement the system takes over.
Save the view that answered the question — Data view, its saved-views dimension
Save the view under a name, as a saved search, a smart folder, or a label opened as a view, for one person, for a role, or as the default. Remove saved views when they stop being used.
Give each recurring purpose its own view — Purpose-keyed view
Give each recurring purpose behind the saved views, such as a quick check, a review, or a follow-up, its own view, and keep the general view reachable from each. When the system offers a view on its own schedule, treat it as an interruption and show where its contents come from.
Write the judgement down as rules — Problem-curated view · optional
Write the recurring judgement down as detection rules, and show the result as a checklist that says why each item was flagged. Keep false flags rare, let the actor report them, and keep the unfiltered collection one step away. Skip this step when the domain cannot define its problems in advance.
Leaves: Recurring questions have saved answers, which need maintenance: stale views removed, rules corrected, and the general view kept one step away. The marks and saved views are now part of the collection, and the next actor starts from them.
Connections
Hands off to Composing views
At Choosing how to show the collection, step 1, via Overview and detail
the split-view option continues in that sequence's splitting steps, which cover how the two views are paired and kept in sync
Calls Calibrating reliance
At Choosing how to show the collection, step 1, via Citation
a generated summary, outline, or overview is an AI surface whose claims link back to their sources, which that sequence sets up for each surface
Hands off to Composing views
At Marking items while reading, step 1, via Filtering
an applied label is a facet that sequence's framing steps filter by
Hands off to Working on the same thing
At Marking items while reading, step 3, via Status feedback
threads need someone to close them, and that sequence covers the review conversation itself
Hands off to Working on the same thing
At Saving useful views, step 1
a named view shared with the team passes on how the collection is organised, which often helps the next actor more than the content itself
Hands off to Attention and tempo
At Saving useful views, step 2, via Notification
a view the system offers on its own, such as a weekly view or a digest of flags, arrives as an event, and that sequence decides how prominently to show it