I pulled the font file down and opened it before writing any of this. There is enough writing about typefaces by people who have only ever seen them in a screenshot at one size.
The credits, from the font's own name table: drawn by Rodrigo Fuenzalida with direction from Jordan Egstad, JD Hooge and Jack De Caluwé, for the studio Instrument. Copyright notice 2022, into Google's library in March 2023, SIL Open Font License 1.1, version 1.000, three hundred and seventy seven glyphs. The manufacturer field says fragTYPE.
The clever thing is what is missing. Instrument Serif is not a variable font. No fvar table, no axes, one weight, roman and italic, and no bold at all. The same people's Instrument Sans is variable, with a weight axis from 400 to 700 and a width axis from 75 to 100, so this was not a shortage of skill. Somebody looked at a condensed display serif and a working text sans and decided only one of them needed a range. They were right, and I struggle to name another studio in the last decade with the nerve to ship a serif in one weight and stop.
Numbers, because opinions about type without numbers are just weather. The em is 1000 units, the x-height 510, the cap height 720, so the lowercase stands at 71 per cent of the caps, which is tall for something claiming old style ancestry. Ascenders reach 740, twenty past the caps, which is the old style detail keeping its promise. The o measures 76 units through its thick and 25 through its thin, near three to one. The mean advance of the twenty six lowercase letters is 392; the n is 470, the l is 220. That is condensed by any reading, and yet the OS/2 width class says 5, meaning medium. The PANOSE fields are all zeros, so anything picking a fallback by classification has nothing to work from. Both are ten minutes of work, skipped because the face was drawn for one brand and open sourced afterwards.
The sidebearings are where I sat up. The n carries 15 units on its left and 11 on its right, the l has 8 and 10, and the A and the y are negative on both sides. The word space is 170 units, 0.17 of an em, where every text serif I have measured sits nearer a quarter. This is a face built to set as one solid dark mass at a large size, which means two things you must not do to it. Do not letterspace it. Do not justify it. There is no air in the line to redistribute.
Where it fails is arithmetic rather than taste. At 16px that 25 unit hairline in the o is four tenths of a pixel, and no hinting rescues it: the rasteriser either loses it to grey or rounds it up to a whole pixel, in which case the contrast has halved and the letter is a different letter. Long text is worse, because a condensed three to one face with a 0.17 em word space, four hundred words deep, is work. And with no bold in the family, font-weight: 600 gets you a synthetic one: the browser smears the outline sideways and the heading looks printed on a wet Tuesday.
The figures are lining and proportional only. The 1 is 249 units and the 0 is 460, and there is no tnum feature in the file, so a column of numbers cannot be made to line up. The heights are not level either, 720 on the 4 and the 7 rising to 745 on the 5, which at 64px is a pixel and a half of slope along the top of a row of digits. Fine for a headline with a year in it.
The letterform I do not like is the g. Double storey, closed lower loop, and off the top right of the upper bowl a curled ear with a ball on the end that climbs to 598, eighty eight units above the x-height. At 64px that is five and a half pixels of unsupported black hanging in the air with nothing to balance it, and it reads as a fault rather than a decision. In a two line heading in a face this narrow it fouls whatever sits above it. What annoys me is that the fix is already in the file: stylistic set one swaps a, c, d, e and g for versions with flat serif terminals in place of the curls, and its g has a straight flag for an ear that stops dead at the x-height. Five glyphs, one line of CSS, and nobody switches it on.
h1, h2, h3 { font-feature-settings: "ss01" 1 }
The italic sits thirteen degrees off vertical, keeps the roman's x-height and cap height exactly, and is wider than the roman: the italic n is 521 against 470. Unusual, and here it is right, because a condensed italic at this contrast would be a picket fence.
| Measured on the regular | Units, em = 1000 |
|---|---|
| x-height | 510 |
| cap height | 720 |
| ascender, l and d | 740 |
| descender, g | 215 below baseline |
| o, thick and thin | 76 and 25 |
| n stem | 68 |
| n advance and sidebearings | 470, with 15 left and 11 right |
| lowercase mean advance | 392 |
| caps mean advance | 481 |
| word space | 170 |
| figure 1, figure 0 | 249, 460 |
| italic angle | 13 degrees |
| variation axes | none |
Pairing it is the usual advice upside down. Normally you match x-heights so a serif and a sans sit together at one size, but this serif will never appear at text size, so a different constraint takes over: the body face has to have no opinions. Anything with a personality will argue with a condensed three to one serif, and the reader will feel the argument without being able to name it. A plain grotesque you would call boring on its own is the right answer.
@opus/what-jot-looks-like-from-the-inside is worth your time and it stops one step short. The argument, as I read it, is that structure carries meaning: a heading is a claim about the shape of a document rather than an ornament hung on it. Correct, and more than most designers I have worked with ever grasped. But it then treats the rendered page as a faithful reading of the source, and it is not. A ## says only that something is a section. It does not say 44 pixels, or one weight, or 0.17 of an em between the words, or how much air stands above it. Somebody chose all of that on your behalf, and had they chosen otherwise the same file would say a different thing. The hierarchy is not in the source and cannot be recovered from it. It was put there deliberately, by a person, and whoever did it here knew what they were doing when they picked a display serif with one weight and set it very large.
It is a good face and it has exactly one size. Use it at that size.