An in-page link is the least glamorous navigation on the web and one of the easiest to get subtly wrong. Click a table-of-contents entry and the heading you asked for ends up tucked under a sticky header, or three pixels off, or the page scrolls somewhere plausible and you cannot tell whether it worked. None of that shows up in a code review, because the bug is a number and the number is in a stylesheet.
So I measured every one of them on Nakodo, which has 76 public pages.
The census
const links = [...document.querySelectorAll('a[href^="#"]')].map(a => a.getAttribute("href").slice(1)); 515 same-page anchors across 68 pages. The eight pages with none are the hub pages, where the sections are cards rather than prose: pricing, guides, blog, about and the four index pages for platforms and competitors.
There are another 64 links that carry a hash to a different page, almost all of them into the glossary: a guide that uses the term CPM links the word to its definition. Every one of the 515 and every one of the 64 points at an element that exists, which I half expected. The same-page ones are safe by construction, because the contents list and the sections are both generated from the same array. The cross-page ones are only half checked:
// A glossary anchor must name a real term; other anchors only need their page. const [path] = href.split("#"); assert.ok(path === "/glossary" ? resolveLink(href) || href === "/glossary" : resolveLink(path), ...); So the 59 links into the glossary are validated down to the fragment, and the five links into a section of another page (four into the privacy notice, one into a guide) are validated only as far as the page. Those five happen to be correct today. Nothing would tell me if they stopped being.
The sweep
For each page, scroll to the top, set the hash, read where the target ended up:
for (const id of links) { scrollTo(0, 0); await sleep(30); const el = document.getElementById(id); location.hash = id; await sleep(80); rows.push({ id, top: Math.round(el.getBoundingClientRect().top), smt: parseInt(getComputedStyle(el).scrollMarginTop), atBottom: Math.round(scrollY) >= Math.round(maxScroll) - 1, active: document.activeElement.tagName, }); } scroll-margin-top is the whole trick of a good hash landing: it tells the browser to stop short of the element by that much, so the thing you clicked is not flush against the top edge. Three values are in use across the site:
| scroll-margin-top | Targets | Where |
|---|---|---|
| 80px | 483 | content sections and glossary entries |
| 24px | 31 | privacy and terms |
| 64px | 1 | one section on the home page |
And the landing itself:
| Anchors | |
|---|---|
| Target top equals its scroll-margin exactly | 513 |
| Target top is short of it | 2 |
The two that fall short
Both are the last section of a legal page: #changes on the privacy notice stops at 56px instead of 24, #contact on the terms at 44px. Neither is a bug in the link. The browser scrolled as far as the document goes and ran out of page, so the final section cannot be lifted to its mark.
The fix is to give the document somewhere to go, either scroll-padding-bottom on the scroller or enough trailing space after the last section that the bottom of the page is not the bottom of the content. I have not applied either, because 32 and 20 pixels of difference on the final section of a legal page is the most harmless defect this sweep could have found, and saying that out loud is more useful than quietly fixing it and claiming 515 of 515.
Where 80px comes from
This is the part I did not expect. Here is every sticky or fixed element on all 76 public pages, by computed style:
(none) The marketing header is 65px tall and it scrolls away with the page. So the 80 pixels that 483 targets reserve are not clearing anything. The number has a source, though:
// src/app/(app)/layout.tsx, the layout behind the login <header className="sticky top-0 z-40 border-b bg-background/95 backdrop-blur ..."> <div className="mx-auto flex h-14 w-full max-w-7xl items-center gap-8 px-4 sm:px-6"> Behind the login there is a sticky header, it is h-14 (56px), and the app's sticky side panels sit at lg:top-20, which is 80px. That is where the convention comes from: top-20 is the number this codebase uses for "below the sticky header with some air", and when the marketing pages needed a hash landing they took the same number.
It happens to be a good choice for a page with no sticky anything. 80px of space above a heading reads as deliberate rather than as a jump that overshot, which is the actual goal. But the honest description is that 483 targets on 68 pages are spaced by a measurement borrowed from a header they will never meet, and if the marketing header ever becomes sticky, the number that keeps working will be right by luck rather than by derivation.
Three things that do not happen
Nothing scrolls smoothly. The two stylesheets this site serves are 154,382 bytes together and contain zero scroll-behavior declarations, so every hash landing is an instant jump. That is also why none of this needs a prefers-reduced-motion gate: there is no motion to reduce.
Nothing highlights what you landed on. Zero :target rules in the same 154,382 bytes. You click "Red flags and what they might mean", the page jumps, and the only evidence that the right thing happened is that the heading is near the top. On a long guide that is fine, because the heading is large and the landing is exact. In the glossary, where you have arrived at one definition in a list of 27 near-identical ones, a brief tint on the target would do real work.
Focus never moves. This is the one that matters:
active element after all 515 landings: BODY (515 of 515) The hash target is a <section> with no tabindex, so the browser scrolls it into view and leaves focus where it was. A sighted mouse user does not notice. Somebody using a keyboard clicks a contents entry, the view moves, and their next Tab continues from the navigation list they just left rather than from the section they asked for. The usual remedy is tabindex="-1" on the target plus an explicit focus(), and it is now the top item on my accessibility list for these pages.
Two small structural notes
The thing a hash points at is the section, not the heading, and the two are tied together rather than merged:
<section id={s.id} aria-labelledby={`${s.id}-h`} className="scroll-mt-20 space-y-5"> <h2 id={`${s.id}-h`} className="text-[26px] leading-8">{s.heading}</h2> So each section spends two ids, x for the link and x-h for the accessible name. Across all 76 pages that is 1,400 ids, and the sweep found zero duplicates, which for a site assembled from a registry of content blocks is the kind of thing worth checking rather than assuming.
And the contents list is a plain anchor:
<a href={`#${s.id}`} className="text-link underline-offset-4 hover:underline">{s.heading}</a> Every other internal link on this site is a next/link, which prefetches and navigates client side. A same-page hash has nothing to prefetch, so it is the one category of internal link where the raw element is the right element. The cross-page ones do go through the router, and they land identically: clicking the CPM link inside how much to pay YouTubers ends with the definition at exactly 80px, as does a cold load of the same URL, as does pressing Back afterwards, which returns to the top of the page and drops the hash.
What I am changing
Focus management on the 515 targets, and a :target tint for the glossary where one entry looks like every other entry. The two legal-page landings and the borrowed 80 can stay as they are, now that they are written down rather than assumed.

