The Refiner
How good is this viz, really?
Drop a workbook or paste a Tableau Public link. The Refiner tries it against the rules Tableau's own authoring AI is given and returns its Viz Assay: a score out of 100, every flaw named, and the fix.
AI is about to write a lot of Tableau. This is how you tell which of it is any good.
// How the Refiner assays
Three fires. Each one burns off a different kind of dross.
Spec lint
reads the workbook XML
Chart choice, sorting, axes that skip zero, unsynced dual axes, rainbow scales, red against green, crowded dashboards, default names and tooltips.
Render and judge
reads the drawn viz
Tableau renders it. The Refiner pulls the palette off the screen, simulates color blindness, checks contrast, then a vision model reviews labels, titles and legibility.
Truth check
reads the data behind it
Recomputes what the viz claims from its source, so a beautiful chart of the wrong number still fails.
// Scoring
Each kind of mistake costs its severity once: major 15, moderate 8, minor 3. Repeats compound, up to three times, so a flaw costs the same in a two-view workbook as in a forty-view one. Every rule cites the guidance it comes from, most of it Tableau's own.
90+
refined
would feature it
75+
clean
a few rough edges
60+
needs work
the message is buried
<60
dross
misleading or unreadable