1,674 numbers are visible on our website, and the page with the most of them is a fiction that adds up
A content site is an accumulation of claims, and the claims that can embarrass you are the numeric ones. Nakodo's content rules say it plainly: claim only our own behaviour, which is in the code, or something with a link
A content site is an accumulation of claims, and the claims that can embarrass you are the numeric ones. Nakodo's content rules say it plainly: claim only our own behaviour, which is in the code, or something with a linked official source, and no invented statistics. I had never checked whether the pages obey that, so I counted every number a reader can see.
Method, and three traps
Visible text of all 76 sitemap pages, with header, footer and nav stripped, then:
NUM = re.compile(r"(?<![\w/#-])(?:Β£|\$|β¬)?\d[\d,]*(?:\.\d+)?%?")
The lookbehind matters. Without it the census counts the digits inside slugs, ids and hex colours, and a page about "top 10 tools" scores the same as its URL. The second trap is the furniture: our footer links are on every page, so an uncleaned count makes "2026" the top number on the site 76 times over. The third is the pricing page, which localises its prices, so a curl from a shell measures whatever currency the edge picked for that request, not what the page "says".
The census
numerals in body text 1,674
distinct numerals 282
median per page 14
most: /how-it-works 110
/guides/youtube-engagement-rate 98
/guides/how-much-to-pay-tiktok-and-instagram-creators 78
home 76
least: /guides, /tiktokers, /instagrammers 2 each
By kind:
| Kind | Count |
|---|---|
| plain numbers | 1,166 |
| dates | 163 |
| step numbers (01 to 06) | 145 |
| percentages | 121 |
| money | 79 |
282 distinct numerals is the number I found most useful. A site with thousands of distinct figures is quoting a dataset. A site with 282 is making the same handful of points, with worked examples.
Almost nothing is interpolated, and that is the right answer
The content registry is 77 files and 13,660 lines of prose written as data. It contains 1,574 numeric literals, and exactly five of those files interpolate anything at all:
files with ${...} in them: 5 of 77
interpolations: 87
Nearly all of them are in the one hand-written page that explains how the product works, which pulls its constants straight from the modules that implement them. Every other number in the registry is typed into a string.
That sounds like an accident waiting to happen until you look at what the numbers are. Here is the most product-adjacent sentence in the whole registry, from the agencies page:
a: "There's a free plan with no card required... Plans are measured by new creators
contacted (a month on Free, each weekday on paid plans), and paid plans raise that
limit and the number of campaigns. Current prices and every limit are on the
[pricing page](/pricing).",
It describes the shape of the limits and refuses to name one. Searching the registry for our actual plan numbers finds nothing: no price, no monthly cap, no daily cap. They live in one module, they are rendered on /pricing and the home page from that module, and the prose everywhere else points at the page instead of copying it. 1,574 typed numbers, and not one of them is a product fact that could go stale.
Who vouches for the rest
If the numbers are not ours, they are somebody's. Counting external links on the pages that carry numbers:
help.instagram.com 37 asa.org.uk 20
support.google.com 23 ftc.gov 15
trends.google.com 17 gov.uk 13
tiktok.com 11 ads.tiktok.com 8
And the hedging. Counting numerals with a hedge word ("roughly", "about", "usually", "commonly", "up to", "hypothetical", "example") within 90 characters:
hedged: 295 of 1,674 (18%)
18% sounds low until you see where the rest of the hedging went: into table captions. The size-band table on micro-influencers is captioned "Commonly used size bands. Conventions, not a standard", which hedges 12 numbers at once from outside my 90-character window. The engagement-rate tables do the same with "Hypothetical numbers for one channel". The measurement undercounts the practice, and the practice is the thing I wanted to check.
14 pages carry ten or more numbers and no outbound link at all. Going through them, they split cleanly: the four calculators and how much to pay a YouTuber are arithmetic, where the numbers are an example being worked through in front of you (40,000 Γ· 1,000 Γ $25 = $1,000), and the audience pages describe our own product. Neither needs a citation. That leaves zero pages quoting an unsourced statistic, which is the result I was hoping for and did not expect.
The densest page is invented
The home page has 76 numerals and no outbound links, and almost all of them belong to three creators who do not exist. They are a demo of what a result looks like: No. 0147 Open Rate Lab, 48K subscribers Β· 21K avg views Β· 4 uploads a month, and two more.
So I checked the fiction for internal consistency, which felt absurd until it passed:
"4 uploads a month" vs "12 uploads in the last 90 days" consistent
"21K avg views" vs "median 21K views" same number
"157 of the last 412" = 38.1% vs "Questions 38%" consistent
Questions 38 + Discussion 27 + Praise 29 + Spam 6 = 100
English 74 + German 12 + Other 14 = 100
"Three in four write in English" vs 74% consistent
Somebody made up a channel and then made its numbers add up. Nothing enforces that: the percentages are literal arguments in a component, the stats line is one hand-typed string, and the two sets live in different files.
The one place it has already gone slightly wrong is here:
{
no: 163,
name: "Formula Desk",
score: 30,
stats: "9K subscribers Β· 4K avg views Β· 2 uploads a month",
note: "Email not found... It scores 30 of 100: a related topic, but a small audience.",
}
30 appears twice in one object: once as data and once inside an English sentence. Change the first and the page contradicts itself in the same card, and no test can see it, because a test that reads the note is a test that parses prose.
What I am changing
- The sample note gets the score interpolated from the field above it. One template literal, and the duplicate is gone.
- The invented demo stays invented, and I will write down the arithmetic it satisfies as a comment, because the next person to tweak "4 uploads a month" will not know that another file says 12 in 90 days.
- A script, not a test: list every numeral on every page that has no hedge, no nearby link and no arithmetic around it, for a human to read before a content batch ships. Today's answer is a short list. The value is in seeing it get longer.
- The registry keeps typing its numbers and refusing to type ours. "Current prices and every limit are on the pricing page" is the cheapest correctness guarantee in the whole codebase.
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.