Spreadsheet speed, without losing track of where the state came from

Filters live in the layer they came from: the dashboard, the person, or the single widget. Nothing is labelled for that — the section is the label.

Two of them carry a marker. An arrow leaving means the filter is published and something else is reading it. An arrow arriving means it came from a subscription. Hovering either one names the widget on the other end and the fields that travel with it. Everything else sits unmarked, because a filter that stays where you put it holds no surprise.

The complaint was «too many filters». The real problem was that the same value could arrive from three places and the screen looked identical in all three. Once state became reusable, origin stopped being a nicety: you are working inside other people's decisions.

Publishing and subscribing were given separate semantics and separate visual treatment, and a filter inherited through a chain of subscriptions is surfaced rather than hidden. When widget C subscribes to B and B subscribes to A, C shows that its state originates at A.

A date filter behaves the same way. T-1 to T-5 replace a calendar for the choices people actually make, a period can have days taken out of it, and the label on the dashboard states what survived: two dates, a range with its length, or a custom selection with the number of days it contains. Two dates and a fortnight used to look identical on the dashboard, and the difference was left for the reader to work out.

SKINS — saved filter sets. Open a new widget and apply a saved filter set instead of rebuilding it. Frequently used SKINS pin to the top and the widget reopens with them next session.

IDEAS — pre-wired widget templates. A step up in scale: whole layouts where the widgets, the connections between them and their filters are already wired. Open one instead of assembling five widgets and connecting them by hand.

Both came from the same request, made repeatedly: stop making me rebuild my morning. The reasoning inside a workspace was being thrown away every session while the layout survived.

Dashboards and analysis templates can be shared, and shared objects have owners. When someone wants to change one they do not own, the change goes out as a request: the owner sees what is being asked, by whom, on which object, and approves or rejects it. Access itself is layered — Reader, Co-Owner, Owner — and a file inherits the permissions of the folder it lives in.

The moment saved state became reusable, it stopped being personal. A dashboard three teams read every morning cannot be edited by whoever opens it first, and it cannot be locked either — the person who spots the error is rarely the person who owns the object.

Unresolved and resolved are separate lists. A request stays in the first one until somebody acts on it, then moves to the second carrying the decision with it, so nobody has to ask whether they were rejected or simply forgotten.

Session History records what you changed on top of a shared dashboard: applied presets, excluded desks, completed drills. The list is yours alone — you inspect a state before returning to it, and nobody else sees it. Ctrl+Z lives at this level and is scoped to the widget you are working in, so a familiar shortcut keeps a small, honest meaning instead of a global one.

Published versions are the second level, and undo never reaches them. Publishing to a shared dashboard is a separate right; what arrives for everyone else is a named version with an author and a date, not somebody’s working session. The two levels never touch, which is the point: no shortcut can quietly rewrite what a room full of people is reading.

Three status palettes, switchable per user: Classic red/green, Signal orange/green, and Accessible amber/blue. The colour is not decoration here — in market risk it is the sign of the movement. An analyst who cannot separate red from green cannot read the direction of a delta across forty rows, and the number stops carrying its most important attribute.

The orange mode came first, and it came from users: red on dark and elevated surfaces was reported as hard to read, and pure red gives 4:1 against white — below the 4.5:1 that WCAG 2.2 AA requires for normal text. Fixing that exposed the larger problem. Red-green is the one pair that fails for the most common forms of colour vision deficiency, and the platform had no alternative to it. Amber against blue survives those forms, and it is the convention professional market terminals already use — so the mode is unfamiliar to the eye but not to the domain.

Colour is never the only carrier. In Accessible mode direction is also encoded as shape — an upward and a downward glyph — and the sign stays on the number. That is the actual requirement: information must not depend on colour alone. Designed against contrast and non-colour-dependence requirements; no formal audit was run, so no compliance is claimed.

A second decision, inside the grid: the colour indicator moved out of the cell. Colouring the cell background put text on top of a saturated fill, where contrast collapses at exactly the moment the number matters most. The value now sits on the normal surface and the status colour becomes a chip immediately to its left — same scanability down a column, no text-on-colour anywhere.

Conditional formatting lets a user set a cell background of their own, and some of those pairings fail contrast regardless of which mode is active. What the product should do about a colour the user chose is an open question, not a solved one.

Every mode was rendered through simulations of colour vision deficiency. Three of these are audiences; the fourth is a stress test rather than a group of people.

Everything in the platform — widgets, dashboards, filters — is a query underneath. It used to be written by hand, by admins and a small number of power users. The redesign moved the frequent combinations into the interface and left the query editor in place for everything else, one tab away rather than in another tool.

Between them sits a filter builder: when the value you need is not in the list, you compose the condition yourself without dropping to the query. Three rungs, and the interface never pretends the third one is unnecessary.

Where a hand-written query expresses something the panel cannot render, the interface says so — applied, but some filters cannot be presented on the edit panel — instead of silently showing an incomplete picture. A panel that quietly omits what it cannot draw is worse than no panel.

Users bookmark filter combinations, then drill through them: each saved filter becomes a step in the path. A drill-in stops being something you reconstruct by hand and becomes something you can name, repeat and hand to a colleague.

The Advanced Table is where the work actually happens. Pivot and tabular modes, drill in by any dimension, dense rows sized so a full book fits on one screen, and explicit progress states for queries that take real time to come back — a risk analyst should never wonder whether the screen is thinking or broken.

Drilling in rewrites the table, so every drill leaves a breadcrumb: Original State / Drill in by Scenario Title. One click back, always.

The part that mattered most sits in the right-click menu. «Why did this number move since yesterday» is asked constantly, and the honest answer used to live in a different report. Now Drill in by Insights → Active & Passive Drivers splits the change into what the desk did and what the market did — one click from the cell, without losing the table you were reading.