Why Type Scales Are Geometric, Not Additive
The modular scale generator on this site builds every size by multiplying, not adding — size(step) = base × ratiostep. That's not an arbitrary implementation detail; it reflects a real fact about human perception that has nothing to do with the golden ratio specifically, and everything to do with why "add the same number of pixels each time" produces a noticeably worse type scale than "multiply by the same ratio each time," at any ratio. That single design decision, not the ratio value plugged into it, is where the real justification lies.
The naive alternative, and why it goes wrong
Picking type sizes by simple addition — 16px, 20px, 24px, 28px, 32px, each one 4px larger than the last — feels like the obvious approach, and it produces a sequence that looks perfectly reasonable written down as a list of numbers. The problem shows up once those sizes are actually compared to their neighbors. Going from 16px to 20px is a 25% increase. Going from 28px to 32px — the identical +4px step — is only about a 14% increase. The step size in absolute pixels stayed constant, but the step size in proportion shrank as the base size grew, which means the visual "jump" between adjacent sizes gets progressively less noticeable the further up the scale you go, even though every step added the exact same number of pixels.
Why proportion is what perception actually tracks
This isn't a matter of taste; it's a well-established finding in psychophysics called the Weber–Fechner relationship, developed independently by Ernst Weber and Gustav Fechner in the 19th century (the same Fechner whose later, specific claim about people preferring golden rectangles has not held up well under replication, discussed separately on this site — the two findings are unrelated beyond sharing an author). Weber's law states that the smallest change in a stimulus a person can reliably notice — the "just noticeable difference" — scales in proportion to the stimulus's existing size, not as a fixed absolute amount: a 1kg weight added to a 2kg load is easy to notice; the identical 1kg added to a 50kg load isn't. Fechner's law builds on that observation to describe perceived intensity as roughly proportional to the logarithm of the physical stimulus. Applied to type sizes, the practical upshot is direct: if you want each step on a scale to feel like a consistent, evenly-spaced jump in perceived size, the steps need to be constant ratios of each other, not constant absolute differences — which is exactly what a modular scale's multiplicative formula produces and an additive "+4px every time" scheme does not.
Seeing the difference in real numbers
Running an actual geometric scale alongside the naive additive one makes the contrast concrete. The modular scale generator, base 16 and ratio 1.25, produces 16, 20, 25, 31.25px for steps 0 through 3 — every single step is exactly a 25% increase over the one before it, with no exceptions, checkable directly from the generator's own output. An additive scale trying to match those same first two numbers (16, 20) and then continuing by the same +4px logic gives 16, 20, 24, 28 — and the percentage jump between each pair quietly drifts: 25%, 20%, 16.7%, shrinking every time even though the absolute step never changes. By the time an additive scale reaches its higher sizes, consecutive steps can end up feeling almost indistinguishable, which is a common, specific reason a hand-tuned type system starts to feel muddy at the large end of the range without anyone being able to say exactly why.
The signature that gives it away: equal spacing on a log scale
There's a precise, checkable way to see why a geometric scale satisfies the perceptual requirement above and an additive one doesn't: take the base-10 logarithm of every size, and see whether the gaps between consecutive log-values stay constant. For the geometric scale (16, 20, 25, 31.25), the log values are 1.20412, 1.30103, 1.39794, 1.49485 — and the gap between each consecutive pair is exactly 0.09691, every single time, with no drift at all. For the additive scale (16, 20, 24, 28), the log values are 1.20412, 1.30103, 1.38021, 1.44716 — and the gaps are 0.09691, then 0.07918, then 0.06695, shrinking at every step. Equal spacing in log-space is precisely what Fechner's law says corresponds to equal spacing in perceived size, which is why the geometric scale's gaps hold steady in log-space while the additive scale's visibly don't — the log-domain calculation isn't just a mathematical curiosity, it's the exact quantity the perceptual law is about.
Any ratio, always the same fixed percentage jump
One more way to see the constant-relative-step property directly: the percentage increase from any step to the next on a modular scale is always exactly (ratio − 1) × 100%, regardless of which step you're looking at or how large the base size has grown by that point. A ratio of 1.25 always means a 25.00% jump, whether that's the jump from step 0 to step 1 or from step 6 to step 7. A ratio of 1.333 always means a 33.30% jump. φ always means a 61.8034% jump — a considerably bigger relative step than either of the more conservative ratios, which is the real, checkable content behind describing φ as sitting toward the "aggressive" end of common type-scale choices, discussed in more practical depth elsewhere on this site. None of that changes with scale size, base value, or how far out on the scale you've gone — the percentage jump is a fixed property of the ratio alone, which is exactly the constant-relative-step behavior an additive scale can never provide at any point beyond its very first interval.
Why this argument doesn't pick a specific ratio
It's important to be precise about what this actually justifies and what it doesn't. The Weber–Fechner reasoning is a case for using some consistent ratio rather than a consistent absolute difference; it says nothing at all about which ratio is correct, and there's no psychophysical result establishing that φ (≈1.618) produces more perceptually even steps than 1.25, 1.333, or 1.5 do — all of them are equally "geometric" in the sense that matters here, and all of them equally satisfy the constant-relative-step property that a naive additive scale fails to provide. The choice between them remains the aesthetic decision described elsewhere on this site: how large a visual jump feels appropriate between adjacent sizes, not a question with a psychophysically correct answer.
A related, separate justification: spacing needs the same logic
The same reasoning extends to spacing values — margins, padding, gaps — for an identical reason. A 4px gap growing to an 8px gap (double) reads as a clearly deliberate, noticeable increase in breathing room. The identical +4px change applied near the top of a spacing scale, say 60px to 64px, is barely perceptible by comparison, for exactly the same proportional reason type sizes behave this way. Building a spacing scale from the same multiplicative formula as a type scale — base 8, ratio φ, giving 8, 12.94, 20.94, 33.89, 54.83px for steps 0 through 4 — keeps every spacing jump feeling like a consistently sized decision, rather than large jumps up top and imperceptible ones at the bottom, for reasons that trace to the same underlying perceptual fact as the type scale itself.
Type scales aren't a special case of this principle
Plenty of other measurement systems people interact with daily are built geometrically for exactly the same underlying perceptual reason, which makes the type-scale argument easier to trust rather than a one-off design trick invented for typography specifically. Camera f-stops step by a factor of roughly √2 ≈ 1.414 (f/1.4, f/2, f/2.8, f/4, f/5.6…), because each step is built to correspond to a constant, easily-reasoned-about halving or doubling of the light reaching the sensor, not a constant absolute change in the f-number itself. Sound intensity is measured in decibels, a logarithmic unit built directly on the same Fechner relationship, precisely because human hearing perceives loudness roughly logarithmically rather than linearly — a jump from 10 to 20 decibels represents a much smaller absolute change in sound energy than a jump from 80 to 90 decibels, but both are perceived as roughly comparable increases in loudness. The Richter and modern moment-magnitude scales for earthquakes work the same way, each whole-number step representing roughly a thirty-fold increase in released energy rather than a fixed additive amount. None of these systems has anything to do with the golden ratio, or with each other beyond sharing the same underlying insight: when a human perceptual response scales with the logarithm of a physical quantity, the measurement system built to match that response should step geometrically too. A modular type scale is simply that same well-worn idea applied to font sizes.
Try the comparison directly
The modular scale generator builds the geometric scale directly from any base and ratio, making it straightforward to generate the numbers above yourself and compare the percentage jump between any two adjacent steps — a quick way to confirm the constant-ratio property holds at every step, not just the first couple, regardless of which specific ratio gets chosen.