boxplot

Use to compare distributions across categories from raw observations: latency per service, scores per class. One row = one observation: x category, y number. Repeated categories are the point. Optional series puts one box per series inside each category. Keep category order as first seen. Quartiles use linear interpolation (type 7, as numpy, R, and Excel QUARTILE.INC). Whiskers reach the furthest observations inside 1.5 × IQR; the rest are hollow outlier dots. A group with fewer than five values shows its points and a median tick, not a box. An empty y is skipped. Negatives are fine.

Do not use for one number per category (bar) or for a single distribution's shape in detail (hist).

x category · y number (one observation per row) · series optional (one box per series)

Rules. Repeated categories are expected; keep first-seen order. Type-7 quartiles; whiskers at the furthest points inside 1.5 × IQR; points beyond are outliers. A group under five values shows its points and a median tick, not a box. An empty y is skipped.

Example

markvis: 2
type: boxplot
title: Search answers fastest but has the longest tail
unit: ms
x: service
y: latency

service,latency
Search,42
Search,38
Search,45
Search,40
Search,39
Search,44
Search,120
Checkout,88
Checkout,92
Checkout,85
Checkout,95
Checkout,90
Checkout,99

Search answers fastest but has the longest tail

service	latency
Search	42
Search	38
Search	45
Search	40
Search	39
Search	44
Search	120
Checkout	88
Checkout	92
Checkout	85
Checkout	95
Checkout	90
Checkout	99

When the block is wrong

The rows stay, with one line that names the error.

markvis: 2
type: boxplot
title: Text in a measure
x: service
y: latency

service,latency
Search,fast

service	latency
Search	fast

E_BAD_NUMBER: row 1, column latency: "fast" is not a number

Generated from packages/types/boxplot/ and the type's row in SPEC.md.