A panel per customer produces 140 panels on one page - what has the chart stopped doing, and what do you do instead?
answer
- read one panel, sweep the whole set
- area per panel, at the size it is read
- coarser grouping beats smaller panels
- 140 comparisons want one mark each
- layout order is read as meaning
basics
~20 sAt 140 panels each is a few pixels of ink: no panel can be read and the set cannot be scanned. Split on a coarser grouping column, or answer the question with one mark per customer instead.
solid answer
~50 sA panel set works while two things hold: each panel is large enough for its marks to be read, and the whole set sits in one view so the eye can sweep it. At 140 panels both have failed, and shrinking them further only hides the failure. The honest responses are to split on a coarser grouping column so the panels number a handful, to answer the reader's real question with a different comparison - one mark per customer against a common baseline is a chart 140 customers fit into - or to draw a chosen subset and state the criterion on the page. There is a second defect in a page of 140: the order they are laid out in. A reader takes layout order as information, so an order nobody chose is still read as a ranking, and the chart makes a claim by accident.
go deeper
Recall that a panel has to be big enough for its marks to be read, so the number of panels is limited by the area on the page rather than by what the tool will draw.
Explain the two requirements that fail together - reading one panel and scanning the whole set - and derive the ceiling from the smallest mark that must stay legible.
Demonstrate that the fix is a different comparison, not a smaller panel: a coarser grouping column, one mark per group, or a stated subset. Treat panel order as content that needs a reason.
The lead's call is what a report is allowed to ship: a rule that a panel set names its ordering and its selection criterion costs little and stops a class of accidental claims.
## What a panel is for, and when it stops being one A panel set exists so that a reader can do two things: **read any single panel**, and **sweep the whole set**. Both have a size requirement, and they pull against each other. Every panel added takes area from all the others, so the count is not limited by taste - it is limited by arithmetic on the page. At 140 panels, each one is a thumbnail. A line in it is a few pixels long, the axis has no room for ticks, and the label is either unreadable or has been dropped. Nothing can be read, and nothing can be compared. The page is not a chart that has become cluttered; it is a lookup table rendered as pictures, which is the worst form of both. The important move in an interview is to say why, rather than to name a number. ## Working out the ceiling for a given page There is no universal maximum, and quoting one is a weaker answer than deriving it: 1. **Start from the smallest mark that must stay legible.** If the reader has to see a monthly wobble, that wobble needs enough pixels to be a wobble rather than a jitter. That sets a minimum width and height for a panel. 2. **Divide the area available at the size the page will be read at.** Not the size you drew it at - a chart read on a phone, in a printed pack or at the back of a room has far less usable area than the screen it was built on. 3. **Check the whole set fits one view.** A panel set that requires scrolling has lost the scan, which was the main thing panels bought over a legend. That calculation usually lands somewhere in the low tens for a rich mark and higher for a very simple one. The answer is a consequence of the mark and the medium, not a rule of thumb. ## Four honest responses when the count is too high - **Split on something coarser.** Customers become segments, or tiers, or regions. The reader almost never wanted 140 individual stories; they wanted to know how the population behaves and where it splits. - **Change the comparison.** If the question is which customers are unusual, the chart is one mark per customer - a point, a bar, a dot on a distribution - not a panel per customer. A comparison of 140 things against each other belongs on one set of axes at a shared position; a panel set is for studying a handful of shapes in detail. - **Draw a chosen few and say so.** A panel set of eight, with a sentence on the page stating the criterion by which those eight were chosen, is honest. Eight panels chosen silently is not, because the reader has no way to tell a selection from a population. - **Accept it is not a chart.** Sometimes the deliverable really is per-customer detail for 140 customers, and the right artifact is a table somebody can search, with a chart for the population next to it. ## The order the panels are in is information This is the second failure on a page of 140, and it survives even when the count is reasonable. **A reader takes the layout order as meaning.** Left to right, top to bottom, the grid reads like a ranking or a sequence, because that is what grids of things mean in every other context a reader has met. So there are only two situations: - the order carries something you chose - by the value being shown, by geography, by time, by an ordering the domain already has - and the layout is then doing a second job for free, letting the reader find a panel and see a gradient across the set; - the order is whatever the tool encountered first, and the reader **still** reads meaning into it. This is the trap: an accidental order is not a neutral order. The chart asserts a sequence nobody intended, and the assertion is invisible because it is made by position rather than by a mark. The practical consequence is that panel order belongs in the review of a chart in the same way a baseline does. If you cannot say why the panels are in that order, the chart has a claim in it you have not checked. ## What the interviewer is testing Three things, in order of how many candidates reach them. First, that you notice the panels stopped being readable rather than merely crowded. Second, that your remedy changes the comparison rather than the appearance - shrinking, thinning the lines or removing the axes is rearranging the failure. Third, that you treat the order of the panels as part of the chart's content, which is the half most people never mention at all.
- The panels came out in whatever order the groups were encountered. Why is that a defect rather than a detail?Because position in a grid is read as an ordering whether or not one was intended. A reader scanning left to right assumes the sequence means something - a ranking, a time order, an importance - and an order nobody chose therefore asserts a relationship that is not in the data. Either give the order a reason or say there is none.
- How do you decide the ceiling on panel count for a particular page?Work backwards rather than quoting a number. Take the smallest feature a reader must see, work out the panel size that keeps it legible, divide the area available at the size the chart will actually be read at, and check the whole set still fits one view. A panel too small to show its mark is ink, not a chart.
- Your reader insists on seeing all 140 customers. What do you give them?Two artifacts rather than one. A chart with one mark per customer against a common baseline shows the whole population and where the unusual ones sit, and a searchable table carries the per-customer detail. A panel set of 140 serves neither purpose, because it can be neither read nor scanned.
saying these in an interview costs you the question
- Shrinks the panels until they fit and calls the page finished.
- Believes any grouping column can become panels if the page is large enough.
- Leaves whatever order the panels came out in, assuming a reader reads none into it.
- Reaches for a better-looking chart when the comparison itself is the wrong one.
- Shows a chosen few panels without saying anywhere how they were chosen.
- Quotes a fixed maximum panel count without reference to the mark or the medium.