Compare three or more

Paired bars take any number of parties. The first party is always the one being judged; versus names the one it is compared with, so a third party can stand there for scale without being in the tags.

Write the data

This one is JSON, and lower is better: it measures latency.

{
  "title": "REQUEST LATENCY",
  "unit": "MILLISECONDS AT THE 99TH PERCENTILE",
  "better": "lower",
  "subtitle": "SAMPLE DATA, ONE SERVER, 200 CLIENTS",
  "date": "2026-10-11",
  "suffix": " MS",
  "versus": "CACHE V1",
  "parties": [
    { "name": "CACHE V2" },
    { "name": "CACHE V1" },
    { "name": "NO CACHE" }
  ],
  "rows": [
    { "label": "READ ONE KEY", "values": [0.8, 2.1, 14.0] },
    { "label": "READ 100 KEYS", "values": [6.5, 9.7, 88.0] },
    { "label": "WRITE ONE KEY", "values": [1.9, 1.7, 3.2] },
    { "label": "SCAN A PREFIX", "values": [41.0, 39.5, 120.0] }
  ],
  "notes": [
    "SAMPLE DATA FOR THIS README. EACH ROW COMPARES CACHE V2 WITH CACHE V1; NO CACHE IS THERE FOR SCALE."
  ]
}

Every row has one value per party, in the parties’ order. Without versus, the first party is compared with the second.

Draw it

$ pcmp latency.json
latency-paired.png
latency-ratio.png

Paired bars with three parties, lower is better: request latency of cache v2, cache v1 and no cache, each row tagged against cache v1

With better = "lower", the unit line reads LOWER IS BETTER and a smaller value counts as faster: READ ONE KEY is 2.1 / 0.8, 2.6× faster.

The ratio chart compares the same two, the first party and versus:

The balanced ratio chart of the latency data: cache v2 against cache v1, lower is better

Values that aren’t speeds

The tags say FASTER, SLOWER and SAME SPEED by default. For sizes, scores or anything else, words changes them:

words = { better = "SMALLER", worse = "LARGER", same = "SAME SIZE" }

Next: pick the colours when three or more parties need telling apart.