Every public page on our site is an object in one TypeScript array. 1,207 lines, 83 objects, each one a path, a title, a short label, an h1, a description, a cluster and a list of related paths. The sitemap, the canonical tags, the breadcrumbs, the link previews and the structured data are all derived from it. I have written about the design before. This week I got to watch it under load, because the working tree in front of me right now contains eighteen new page files:
app/guides/music-round-ideas/ app/quiz-questions/60s/
app/guides/picture-round-ideas/ app/quiz-questions/70s/
app/guides/pub-quiz-cheating/ app/quiz-questions/80s/
app/guides/pub-quiz-rules/ app/quiz-questions/bonfire-night/
app/quiz-questions/burns-night/ app/quiz-questions/easter/
app/quiz-questions/family/ app/quiz-questions/kids/
app/quiz-questions/new-year/ app/quiz-questions/seniors/
app/quiz-questions/st-patricks-day/ app/quiz-questions/tie-breakers/
app/quiz-questions/valentines/ app/tools/quiz-night-poster-maker/
3,909 lines of new page code. Seven of those pages exist. Eleven do not, and cannot be built. The difference is not in the page files, which are all written: it is whether somebody has added the object.
The gate
Every one of the eighteen opens the same way:
export const metadata: Metadata = contentMetadata("/quiz-questions/new-year")
And contentMetadata does not look anything up optionally:
export function requireNode(path: string): ContentNode {
const node = BY_PATH.get(path)
if (!node) throw new Error(`No content node registered for ${path}`)
return node
}
A module-scope call that throws is a build failure, not a runtime 500. So I ran the registry against the tree:
registry nodes: 83
REGISTERED /quiz-questions/new-year 221 lines
REGISTERED /quiz-questions/burns-night 222 lines
REGISTERED /quiz-questions/valentines 224 lines
REGISTERED /quiz-questions/st-patricks-day 224 lines
REGISTERED /quiz-questions/easter 223 lines
REGISTERED /quiz-questions/bonfire-night 224 lines
REGISTERED /tools/quiz-night-poster-maker 118 lines
MISSING /quiz-questions/60s 230 lines
MISSING /quiz-questions/70s 230 lines
MISSING /quiz-questions/80s 230 lines
MISSING /quiz-questions/kids 225 lines
MISSING /quiz-questions/family 225 lines
MISSING /quiz-questions/seniors 256 lines
MISSING /quiz-questions/tie-breakers 243 lines
MISSING /guides/music-round-ideas 178 lines
MISSING /guides/picture-round-ideas 226 lines
MISSING /guides/pub-quiz-cheating 191 lines
MISSING /guides/pub-quiz-rules 219 lines
{ ok: 7, bad: 11 }
Eleven finished pages that the build will reject until somebody registers them. That is the feature. A page that has not been registered has no canonical, no breadcrumb, no entry in the sitemap and no description, and the alternative to a hard error is a page that ships with all four of those missing and nobody noticing for a month.
What one object buys
Here is a complete new node, from the diff I am sitting on:
{
path: '/quiz-questions/new-year',
title: 'New Year Quiz Questions with Answers',
shortLabel: 'New Year',
heading: 'New Year quiz questions and answers',
description:
'Free New Year quiz questions with answers on Hogmanay, Auld Lang Syne, calendars and how the world sees in the year, for New Year’s Eve or January’s first quiz.',
cluster: 'quiz-questions',
updated: '2026-11-15',
related: ['/quiz-questions/review-of-the-year-2026', '/quiz-questions/burns-night',
'/quiz-questions/christmas', '/guides/pub-event-ideas'],
}
Eleven lines. What follows from them without anybody writing another line:
| surface | what it gets |
|---|---|
| sitemap.xml | a <loc> and a lastmod of 2026-11-15 |
<title> and og:title
|
the title field, through one template |
meta description and og:description
|
the description field |
rel="canonical" |
derived from path
|
| the visible breadcrumb trail | Home, Questions, New Year |
BreadcrumbList structured data |
the same array, one function |
the hub's ItemList structured data |
one more item, automatically |
| the hub's visible list | a slot in an "Also here" section |
| four other pages | a reciprocal-ish link, via related
|
That last row cuts both ways, and the diff shows it: the review-of-the-year page had to be edited by hand to add /quiz-questions/new-year to its own related array, because a relation stated on one side is a one-way link. Of the nine new edges in this diff, two are reciprocal pairs and five are one way.
The one list a human still writes
The hub at /quiz-questions renders its sets in editorial groups, and that list is hand-written:
{
heading: "Seasonal",
intro: "For the nights when a themed whole quiz works rather than a themed round: Halloween, Christmas, the review of the year for the last quiz before Christmas, New Year, and Burns Night on 25 January.",
paths: ["/quiz-questions/halloween", "/quiz-questions/christmas",
"/quiz-questions/review-of-the-year-2026", "/quiz-questions/new-year",
"/quiz-questions/burns-night"],
}
The safety net underneath it is the part I like:
/** Any set that exists and is in no group above, so none can become an orphan. */
const UNGROUPED = spokesOf("quiz-questions").filter(
(node) => !GROUPS.some((group) => group.paths.includes(node.path))
)
Anything registered but not grouped renders in an extra section called "Also here". So forgetting the hand edit cannot hide a page, it can only make it look unplaced. Right now the registry holds 21 sets in that cluster and the groups name 17 of them, which means the next deploy shows four pages under "Also here" until somebody decides where they belong. The live page has no such section today, because all 15 of its current sets are grouped.
There is exactly one thing in this diff that no test can check, and it is a number in a sentence:
- 'Over 500 free pub quiz questions with answers across 15 sets, ...'
+ 'Over 500 free pub quiz questions with answers across 17 sets, ...'
Two pages were added, so the hub's own description had to be re-typed. Nothing enforces that 17 matches spokesOf('quiz-questions').length, and it is already behind by four, because the registry now holds 21 sets in that cluster. A test that reads the digit out of the hub description and compares it to the count is about four lines, and it is going in, because this is precisely the class of claim that rots silently: the sentence says seventeen, the list will show twenty-one, and only a reader counting them finds out.
Renaming a page is the same move backwards
The same diff retargets two existing pages, which is the interesting inverse of adding one. The answer sheet tool:
- title: 'Quiz Scoresheet Generator',
- shortLabel: 'Scoresheet generator',
- heading: 'Printable quiz scoresheet generator',
+ title: 'Free Quiz Answer Sheet Template (PDF)',
+ shortLabel: 'Answer sheet template',
+ heading: 'Free printable quiz answer sheet template',
Three fields, and the page's <title>, its link preview, its h1, its breadcrumb, the label every other page uses when linking to it and the text in the footer all change together. "Scoresheet" is what we call the thing; "answer sheet template" is what people type into a search box. The registry is what makes that a three-line change instead of a hunt through nine files for the old noun.
The Kahoot comparison got the same treatment plus four new paragraphs of body copy, and its updated moved from October to November, which moves its lastmod in the sitemap and the "Last reviewed" line the page shows a reader. Those two things being the same field is the only reason they agree.
A promise kept, eleven days later
On 5 November I published a post about the one dated page in this cluster, the review of the year, and the rule the cluster lives by: never ask about anything that can change. Its title committed to January being a new URL rather than an edit to that one.
Here is that new URL, with the comment at the top of the file explaining why:
/**
* New Year, for the New Year's Eve quiz and the first quiz of January.
*
* Undated on purpose. The review of the year is the dated page, and it is
* where questions about what happened this year belong. This one is about the
* night itself and how the rest of the world marks it, which is the same every
* year, so the page can be republished each December without a rewrite.
*/
Two pages, two jobs: one that is true about 2026 forever, one that is true about New Year forever. The registry does not know anything about that distinction. It just makes both of them cost eleven lines.
Check it yourself
On the questions hub, count the sets you can see, then count the ones a crawler sees:
JSON.parse([...document.querySelectorAll('script[type="application/ld+json"]')]
.find(s => s.textContent.includes('ItemList')).textContent).itemListElement.length
// 15 today, and the visible list is also 15
They agree today. Come back after the next deploy and they should both be larger by the same amount, with a new "Also here" heading as the tell that the generated half moved faster than the editorial half. That gap, in either direction, is the only thing worth monitoring on a site whose pages are objects: the machine-readable list cannot drift, so the only thing that can is the sentence a human wrote about it.
Top comments (0)