As data magnitude shifts, a once-fitting chart stops working

What it is

A chart's fit is never a one-time verdict: it rules on the data profile of its moment (row count, category count, value range, density, distribution). Business data keeps growing and drifting — categories climb from 8 to 40, values from two digits to millions, scatter points from a thousand to a million — and yesterday's qualified chart crosses its own capacity line: pies overshoot their slice limit, bars cannot fit their labels, a linear axis flattens small values to zero, scatters smear into ink. Fit is a review item that tracks the data, not a freeze-frame at delivery.

Why it happens

Each chart's capacity comes from the physical constraints of its encoding: angular differences fall below the discrimination threshold past a few slices (the pie's ceiling); distinguishable bar-height differences shrink below label height as categories multiply (the bar's ceiling); relative differences under a linear scale get crushed by cross-magnitude data (calling for logs or indices); ink density past a threshold smears structure shut (calling for binning or aggregation). Data growth trips these constraints one by one, and the failure is usually gradual, not a crash: the chart slowly becomes hard to read, nobody can name the day it broke, so it runs on sick. Distribution drift is just as fatal: zeros or negatives disable log axes; a heavy tail disables linear ones.

Where it stops holding

The review trigger is the data profile changing, not the calendar: category cardinality, value-range width (max/min ratio), point count, appearance of zeros or negatives, missingness. Static reports are judged by the profile at publication; live dashboards should automate profile monitoring (cardinality over threshold prompts a chart change). A chart that passes review still faces the question check — switching charts does not license "anything renderable"; the reading-cost constraint stands.

Applying it

• Keep a data-profile record per live chart (cardinality, range, point count at publication) with threshold alerts: categories > 15, value ratio > 1000, points > 10k, or any negative value triggers review.
• Review checklist: does the encoding channel still suffice (slice count, label count, ink density), is the scale still right (linear vs log), should the representation level up (facets, aggregation, heatmap)?
• Verification: quarterly, sample live charts against current data profiles; "last year's chart on several-times-bigger data" means review debt — repay it.

Related

• Same group: U2.17.1 A mismatched chart forces readers to compute the answer mentally · U2.17.2 One question, several viable charts — they differ in reading cost · U2.17.3 Automatic recommendation reads data types, not analytical intent · U2.17.4 Choosing the chart first manufactures questions to fit it
• Nearby: U2.15.4 Too many facets shrink panels below readability · U2.05.2 Many points overplot; use transparency or binning
• Search terms: chart scaling · data growth · visual scalability · representation review