How do you build a type scale that stays consistent without feeling mechanical?
On this page
Start from a ratio, then tune by eye. A type scale stays consistent because the sizes follow a deliberate mathematical relationship instead of being chosen at random, and it stays human because you adjust individual steps where real reading comfort asks for it rather than obeying the math blindly. The method is two-handed: the ratio gives you a coherent skeleton, and your judgment gives it life. Treat the formula as a starting point that earns the right to be overruled, not as a law that must be honored at every step.
The ratio matters first because consistency is what makes a system read as designed. When each size relates to the next by the same multiplier, the jumps feel intentional and the whole interface hangs together, the way a musical scale sounds coherent because its intervals are related. Pick a multiplier that suits the product’s voice, a gentle one for calm dense interfaces, a wider one for expressive editorial layouts, and generate your sizes from it. That single decision buys you a hierarchy whose steps already feel like members of the same family rather than a pile of arbitrary numbers someone typed in.
The trouble is that pure math does not know about reading. A clean ratio will often hand you a body size that is a fraction too small to be comfortable, a caption that the formula pushes below the legible floor, or two adjacent heading sizes so close that the difference looks like a mistake rather than a level. So you override. Suppose a ratio gives you a body of fifteen-and-a-half pixels; you round it to a size that actually reads well and keep the rest of the scale relating to that anchor. Suppose the formula produces a display size that dwarfs everything and breaks the page rhythm; you pull it back to where it commands attention without shouting. Each of these is a deliberate, named exception in service of the eye, not a drift away from the system. The scale stays recognizably ratio-based; it just bends at the points where the math and the reading disagree, and reading wins.
The risk to watch is overcorrection. If you adjust so many steps that no relationship survives, you have not humanized the scale, you have abandoned it, and you are back to arbitrary sizes that happen to look fine today and will fall apart the moment someone adds a new role. The discipline is to keep the ratio visible as the backbone and treat every manual tweak as a justified deviation you could explain out loud, not a habit of nudging numbers until they feel nice. A scale with two or three reasoned exceptions is still a scale. A scale with twelve is a list. A simple gut check is whether you could regenerate most of the scale from the ratio and your anchor size if you lost the file; if you could, the system is intact, and if you could not, you have drifted too far.
When you build your next scale, generate the sizes from a chosen ratio first so the system has a coherent spine, then read real text at each step and adjust only where comfort genuinely demands it, anchoring the rest to your body size. Keep a short note of which steps you overrode and why, so the next person can tell the deliberate exceptions from accidents. Math to make it consistent, judgment to keep it human, and a clear record of where you chose one over the other.