By Every bookmark library I have run grows by the same mechanism. Links arrive through the extension, through imports, through shared folders. Review never arrives. The library grows by saving, never by review, and the damage stays invisible until a needed link no longer resolves.
The measured number that anchored the design: 10 of 36 links dead in our Demo workspace over 12 weeks. A library that grows by saving needs a board that reviews it back. That board is Insights, under Navigate in the app. This article walks it in the order I read it: Needs attention first, then the filing panels, then clutter, then the rest of the board and its chart-or-table switch.
Needs attention, before anything else
Needs attention is the only action list on the board. Four rows, compared on the same axes:
| Row | What it counts | What a click does |
|---|---|---|
| Dead links | Saved URLs that fail their health check | Filters the board to those links |
| Never checked | Links whose health has never been verified | Filters the board to those links |
| Never opened | Saved links with zero opens | Filters the board to those links |
| Going cold | Links that had opens once and none recently | Filters the board to those links |
Clicking a row filters the whole board to the links behind it. In the Demo workspace over 12 weeks the rows that carry numbers read: 10 dead links, 28 never opened, 10 going cold.
I put the action list at the top of the board. The alternative, a dashboard that opens with charts, is pleasant to look at and shows nothing to do. It lost because a number without a next action is decoration. In fact, the four rows also set the vocabulary for the rest of the board: dead, unchecked, unopened, cold.
Tags across folders
Where it's stored draws folder × health bars: one bar per folder, split by link health, and clicking a bar scopes the board to that folder. Tags across folders draws a matrix instead. Tags sit on one axis, folders on the other, cells counted in links. Clicking a cell filters both dimensions at once: that tag inside that folder.
I built the matrix two-dimensional from the start. The alternative, a tag list with counts, is compact and familiar. It lost because a count cannot say whether a tag lives in one folder or in nine, and that difference is the filing signal. A tag scattered across many folders is a concept the library never agreed on.
Where each concept lives
The Concepts tab carries the concept flow. Concepts and folders are nodes, and saved links draw the edges between them. The shape of the flow diagnoses the filing. A concept fanning into many folders is a filing problem: the library keeps the same idea in too many places. A folder fed by many concepts is a junk drawer: one place holding everything.
Unfiled links count in the flow instead of staying outside it. In the Demo workspace that number is 29 links, the largest single group in the diagram. I kept them visible; the alternative, a diagram of filed links only, would read cleaner. However, 29 unfiled links are the story, and a diagram that hides its own subject is marketing, not measurement.
Kept, or clutter?
Kept, or clutter? plots age against opens and sorts every link into one of four shapes:
- Hot: links opened recently.
- Evergreen: old links that still get opens.
- Fresh, unread: recently saved and not opened yet.
- Dead weight: old links with no opens to show.
In the Demo workspace over 12 weeks the plot reads: 26 fresh unread, 10 dead weight. The shapes are worth more than the counts. Fresh unread is a queue, not a problem. Dead weight is a deletion candidate list with a timestamp.
Instead of one clutter score, I kept four shapes. A score is easier to average and faster to scan. However, a number cannot say whether the fix is to read, to file, or to delete, and the four shapes can.
The rest of the board
The remaining panels, in one pass:
- What it's been about: the concept mix per week, on the Concepts tab.
- Top sources: which sites the workspace collects from.
- Saved per week: separates own folders from folders shared in from another tenant.
- Who saved what: resolves member names only with the
rbac:assignments:readpermission; without it, names stay hidden.
I kept that permission gate strict. The alternative, resolving names wherever the data exists, is friendlier to read. It lost because a board that leaks member names against permission would cost more trust than the column adds. As a result, the count line in the header reports links visible to the member for the same reason: shared folders make raw totals lie.
Chart or table, every panel
Every panel switches between its visual view, Chart, Matrix, Flow, or Plot, and a plain Table. The chart answers shape questions: where, how much, in which direction. The table answers which-link questions. As a result, the reader who spots a suspicious bar never leaves the panel to name the link behind it.
I made the table a first-class toggle, not a fallback. Charts alone look better in a tour. They lost because the board exists to act on, and action starts with names, dates, and counts in rows.
What is not in Insights
Scope, stated plainly. Insights has no alerts, no exports, no automatic fixes, and no plan limits. Instead, the first version keeps one promise: open Navigate, Insights, and the whole library reviews itself at a glance.
The full walkthrough of every panel lives in the Insights section of the user manual. The release notes are in the changelog entry for Insights. Finally, a workspace with history gives the board something to measure: start free and open Insights under Navigate.