Skip to content

Calibration

Every score here is checked against 100 companies rated by hand on all seven dimensions by a frontier model, Claude Opus 5.5, with a written reason for every rating. The names were picked so that each dimension has good, middling and poor examples. No single person's taste decides the answers.

The gate

A scorer may run over the corpus only when it agrees with the reference on at least 80% of ratings and is never a full colour away from it. Brevity and Greppability are computed by rule, so they are checked here every time the page loads. The five judged dimensions are checked by the model pass before each corpus run, with every name left out of the examples the model sees while it is being rated.

89.5%

the computed rules agree on 179 of 200 ratings · 21 one step off · 0 a full colour off

Pass
Brevity
99/100
Greppability
80/100

The reference set

An outlined tile is where a computed rule says otherwise

Evocativity
30 39 31 green, yellow and red counts
Brevity
77 15 8 green, yellow and red counts
Greppability
70 22 8 green, yellow and red counts
Googlability
42 22 36 green, yellow and red counts
Pronounceability
74 21 5 green, yellow and red counts
Spellability
41 36 23 green, yellow and red counts
Verbability
7 8 85 green, yellow and red counts

These ratings were made by hand, one name at a time, by the same model family that judges the corpus. Reeves' own ratings for 19 names, which this set replaces, are kept on Methodology.