A type scale is a base size and a ratio. Multiply the base by the ratio to get the next size up, repeat, and stop at four levels. Here are six candidate ratios from a base of 18 point. Each is named for the musical interval it borrows, and each is computed from the exact ratio, not from its decimal:
| Ratio | Level 1 | Level 2 | Level 3 | Level 4 |
|---|---|---|---|---|
| 1.125 (major second) | 18.00 | 20.25 | 22.78 | 25.63 |
| 1.200 (minor third) | 18.00 | 21.60 | 25.92 | 31.10 |
| 1.250 (major third) | 18.00 | 22.50 | 28.13 | 35.16 |
| 4/3 (perfect fourth) | 18.00 | 24.00 | 32.00 | 42.67 |
| sqrt(2) (augmented fourth) | 18.00 | 25.46 | 36.00 | 50.91 |
| 1.500 (perfect fifth) | 18.00 | 27.00 | 40.50 | 60.75 |
Which type scale ratio should you use?
The ratio we ship is 4/3. It gives whole numbers from an 18 point base, and its steps are big enough to be unmistakable at projection distance. Work downward as well as upward from 18 and you get an exact caption size of 13.5 point. That rounds to 14, a deviation of 3.70%, and the four shipped levels are:
| Role | Exact (pt) | Shipped (pt) | Deviation |
|---|---|---|---|
| Caption, source, axis label | 13.50 | 14 | +3.70% |
| Body, bullet, table cell | 18.00 | 18 | 0.00% |
| Lead-in, sub-head, KPI value | 24.00 | 24 | 0.00% |
| Title | 32.00 | 32 | 0.00% |
Adjacent ratios on the shipped scale are 1.2857, 1.3333 and 1.3333. The Linter's hierarchy rule asks two things of a type scale. It must have five or fewer distinct sizes, and every adjacent pair must differ by at least 1.15. The shipped scale's smallest ratio is 1.2857, comfortably clear of the floor. Title to caption is 2.2857, and that's the total range you have for expressing rank by size alone.
Advertisement
Why stop at four sizes?
Now the argument for stopping at four. Somebody always wants a fifth level between body and lead-in, "just for the sub-heads". A fifth size has to clear the 1.15 floor against both neighbours. So it must be at least 18 × 1.15 = 20.70 point and at most 24/1.15 = 20.87 point. The best candidate is the geometric midpoint, sqrt(18 × 24) = 20.7846 point. It's 1.1547 times each neighbour, above the floor by under half a percent, and no whole or half point size falls in that 0.17 point window. Try 21 point and you get 24/21 = 1.1429 against the lead-in. Try 20.5 point and you get 20.5/18 = 1.1389 against body. Both are below the floor. Two sizes that differ by about 15% or less don't read as two levels. They read as one level with an inconsistency, which is worse than either. So the answer to the sub-head request is no, and now you can show your working.
The count of sizes isn't the count of type roles, and that's where decks get away from people. Four sizes and two weights give eight roles, and eight is plenty. Eight sizes and three weights give 24, and no deck has 24 distinct things to say about rank. Nobody picks eight fonts on purpose. The eight-font deck comes from eight separate afternoons, each of which added the one size that particular slide needed.
Worked example: cutting eight sizes to four
The hierarchy score makes the difference concrete. Before the rebuild, the opening slide carries sizes of 11, 14, 16, 18, 20, 24, 28 and 40 point. That's eight distinct sizes, with adjacent ratios of 1.2727, 1.1429, 1.1250, 1.1111, 1.2000, 1.1667 and 1.4286. The smallest ratio is 1.1111, so both halves of the rule fail and the score is 0.0. Rebuilt on 14, 18, 24, 32, both halves pass and the score is 1.0. Four numbers replaced eight and no information was lost. Those four extra sizes hadn't been carrying any distinction a reader could name.