The interaction

Most interaction guidance was written against forms, menus and buttons that respond in under a second. Analytics tools break that assumption constantly: a filter can trigger a query that takes forty seconds, a drill-down can silently reshape what a colleague is looking at on a shared screen, and an export can succeed with zero rows and no explanation why. The five sets tonight were not written for this problem specifically, but applied together they cover it completely - general usability, software-dialogue discipline, the cognitive mechanics of a control, fine-grained interaction craft, and a formal international standard.

Session 6 had one list built almost entirely for its job, Few's 13. Interaction quality has no such single list, because the five traditions below were built by people solving different problems: Nielsen and Molich generalised across whatever interface existed in 1990, Shneiderman was writing specifically about software dialogues in 1985, Norman was explaining why physical objects and digital controls succeed or fail at communicating themselves, Tognazzini was compiling decades of hands-on interaction craft at Apple and Sun, and ISO was standardising dialogue principles for international procurement. None of the five is sufficient alone on an analytics UI, which is the reason tonight runs all five against the same two artefacts rather than picking a favourite.

Session 1 introduced these ten by name. Tonight they get applied specifically to filters, long queries, drill-down, export, and empty or error states - the five moments where analytics interaction genuinely differs from a marketing site.

Ben Shneiderman published these in 1985 and has kept them current through six editions of Designing the User Interface. They speak specifically to software dialogues, which makes them a closer fit to a filter or a query panel than a general usability heuristic.

Two of the eight are the most under-applied on analytics tools specifically. Design dialogs to yield closure means a sequence of actions should have a clear beginning, middle and end - a query that just stops updating its spinner, with no completion state, never yields closure. Reduce short-term memory load means the tool should show what is currently filtered rather than expect the reader to hold five selections in their head while comparing two dashboards in separate tabs.

Affordance is what an object's shape allows it to do - a button-shaped rectangle affords pressing. Signifier is the explicit cue that tells you the affordance is there - a chevron icon, an underline, a change in cursor. Mapping is the relationship between a control and its effect - a filter positioned above the chart it changes maps more clearly than one buried in a side panel three scrolls away. Feedback is the signal that an action registered - a chip appearing, a count updating. Constraint is what the control physically prevents - a dropdown cannot hold an invalid value the way a free-text box can. Conceptual model is the reader's mental picture of what the control does, which must match what it actually does or every use becomes a small surprise.

Bruce Tognazzini published these as AskTog's First Principles of Interaction Design, most recently revised in 2014, drawing on his years leading human interface work at Apple and Sun. Nineteen principles by name: aesthetics, anticipation, autonomy, colour, consistency, defaults, discoverability, efficiency of the user, explorable interfaces, Fitts's Law, human-interface objects, latency reduction, learnability, metaphors, protect users' work, readability, simplicity, state, visible navigation.

Three matter enormously for analytics and are almost never applied on purpose:

• Protect users' work - nothing a reader configures, filters or half-writes should vanish because of a refresh, a crash, or a navigation click. Analytics tools violate this constantly because filter state is treated as disposable.
• Latency reduction - perceived speed is often more important than actual speed. A progress indicator, a partial result, or an honest estimate of remaining time all reduce felt latency even when the query itself takes exactly as long.
• State - the system must always be able to communicate what state it is in. A long query with no visible state is not fast or slow, it is simply unknown, and unknown is worse than slow.

This is the 2020 second edition, which superseded the 2006 first edition. It is the closest thing the interaction canon has to a formally standardised, internationally agreed bar, which is precisely why it belongs alongside four traditions that arrived at similar ground independently.

Shneiderman's dialogs-to-yield-closure rule means every query needs a visible, unambiguous end, distinct from its middle. A spinner that simply stops without a completion message does not yield closure - it just stops signalling, and a reader cannot tell "done" from "gave up quietly".

Reduce-short-term-memory-load means the query panel itself should carry the state a reader would otherwise have to hold in their head: which filters are active, how long the current run has taken, what the last successful run returned. None of that should require memory of a click made ninety seconds ago in a different tab.

Tognazzini's three under-applied principles map directly onto the query panel. Protecting users' work means a filter set, in progress or complete, survives a refresh or an accidental navigation. Latency reduction means the forty seconds feels shorter with a progress signal than it does with silence, even when the actual duration is unchanged. And a visible state at every moment means the reader is never staring at a screen that could mean three different things.

Red-green colour vision deficiency, mostly deuteranomaly, affects roughly one man in twelve and a much smaller share of women. On a row that carries meaning through red-versus-green alone, that reader sees two shades of a similar muddy tone and no way to tell which account is healthy. This is not a rare edge case on a team of any size - it is closer to one seat in most rooms of a dozen.

Nothing here required a sixth principle set to name. NN/g's consistency-and-standards heuristic asks whether a signal means the same thing everywhere, which a colour-only signal cannot guarantee across every reader. ISO's conformity-with-user-expectations principle asks whether the dialogue matches what a reasonable user expects a status indicator to give them - a second cue, not just one. The fix five traditions independently point toward is not a new idea, it is a second channel: shape, an icon, or the word itself, so the colour becomes reinforcement rather than the entire message.