PUBLISHED OCTOBER 2026 · FIVE PRICING PAGES, READ A THIRD TIME

We read the same five pricing pages a third time. Where our count is not the same as last time, it is on a page that does not give the same count on two reads of itself — so this page says what it read, and not whose doing it was.

Three readings of the same five pages, fetched and counted by code with no model in it — no buyer, no decider, nobody reading the page. Our readings of Retool’s pricing page and of Notion AI’s have moved since the first of those readings, and nothing else on these pages has. Since the second reading the same two have moved and nothing else has changed either. Neither movement is one this page puts down to a company. Notion AI’s count went 74 controls in the first reading, 74 again in the second, 73 in this one. Retool’s went 231 controls, then 232, then 233, and what this third reading bought is the reason to doubt both of them: that page offers a “Request trial” control on some fetches of it and not on others, so two reads of it can come back one control apart, and that control is the whole of it. Its count is not the same at each of our readings; nothing here says Retool made it so. Notion’s page has since been read the same way, and the note just below says what that found. Every paid price reading on these five pages has reproduced to the character since the second reading: 1 character on Retool, 16 on Linear, 19 on Airtable, 36 on monday.com, 315 on Notion AI, each of them the distance from that page’s first paid price to the nearest words saying what the price is per. The four of them that were read in the first reading agree with that earlier reading to the character as well.

WHAT THIS PAGE NO LONGER SAYS

When this page went up, its title and its first paragraph put the movement in Notion AI’s control count down to Notion, and said that Retool’s was not theirs. The page named the one fact that would overturn that: at the time, no read of Notion’s pricing page we had taken had given any number but the one we published. We have since read that page many more times in single sittings, and it does give another. So each of the pages behind those counts hands out more than one count of itself, and a gap between our own readings of either one cannot be told from which version we were served. Every number on this page is the number it carried when it went up, and every record behind them is published where it was. What has gone is the sentence that said whose doing it was.

HOW THIS WAS READ, AND BY WHAT

Nobody was in the loop. No ghost user walked these pages, no model decided anything, and not one sentence or number here was written by one — which is why this is a reading and not a run, and why there is nothing on it about where a person got stuck.

The instrument: Each page was fetched once and watched until its own list of controls stopped changing, then counted — the same way, by the same code, at all three readings. Nothing was signed up for and nothing on this page is a sentence anybody or anything wrote about these products: the numbers are counts of controls and character offsets, taken off each page as it arrived. Each reading is one fetch of each page, and that is the limit this reading ran into: a page that does not hand the same controls to every fetch of it will look, to a single fetch, like a page somebody edited.
When: The sitting is the unit rather than the date: a pricing page is edited without notice, so a reading is a fact about the moment it was taken and nothing else. These three readings all fall within three days of each other, and this one was taken twenty hours after the second. What is on this page is therefore a statement about that window and nothing wider: it is not an edit frequency, and nothing here says how often these companies touch these pages. The controls and the prices in this reading were taken within the same hour.
Pages read: monday.com/pricingairtable.com/pricingretool.com/pricinglinear.app/pricingwww.notion.com/pricing

WHERE EACH ONE KEEPS WHAT IT COSTS

Counted on the pricing pages themselves.

We opened each pricing page, waited until its own list of controls stopped changing, and counted the controls that send a reader somewhere else in a new tab — recording, for each one, whether it goes to material of that company's own on a separate site or to somebody else's. These are a fact about these pages on 8 October 2026 and nothing else, and anyone can open the same pages and count them again.

Pricing page	Controls on the page	Open a new tab	…to its own material on a separate site	…to somebody else's site
monday.commonday.com/pricing	167	32	8community.monday.com, ir.monday.com, support.monday.com	13
Airtableairtable.com/pricing	103	18	5community.airtable.com, support.airtable.com	8
Retoolretool.com/pricing	233	10	4docs.retool.com	2
Linearlinear.app/pricing	64	6	0	4
Notion AIwww.notion.com/pricing	73	0	0	0

AND HOW FAR THE PRICE IS FROM WHAT IT IS PER

The same pricing pages, read a second way.

A price on its own is not what something costs: the same number means one thing charged by the month and another charged for a year up front. So we read the rendered text of the same pages again on 8 October 2026 and measured one distance — from the first paid price the page shows to the nearest place that page names a billing period.

Pricing page	First paid price → nearest billing period	Closest any paid price gets	Paid prices within two lines of one	Periods the page names
Retoolretool.com/pricing	1 char	1 char	2 of 13	annual
Linearlinear.app/pricing	16 chars	16 chars	2 of 2	annual
Airtableairtable.com/pricing	19 chars	1 char	4 of 11	annual + monthly
monday.commonday.com/pricing	36 chars	9 chars	12 of 27	annual + monthly
Notion AIwww.notion.com/pricing	315 chars	315 chars	0 of 5	annual + monthly

WHY WE TRUST THIS ONE

There are three ways of asking the question in that table — the first paid price, the closest paid price, and how many of a page's prices sit next to a phrase naming a period — and the tool that assembles this page refuses the whole group unless all of them single out the same page. The reading was chosen after these numbers were on screen, and a finding that depends on which reading we picked is a finding about us rather than about anybody's product. It did not have to agree, and it did:

The odd one out on every reading — Notion AI

What this is not: a claim that any of these pages fails to say what its prices are per. Each one says so, and the column on the right shows which periods each names. What is measured here is how far a reader travels between the two, on one rendering of one page — and anybody can open the same page and measure it again.

WHAT THIS PAGE LEAVES OUT

What we read and are not publishing.

How often these companies edit these pages: A third reading inside three days cannot answer that. What is above is a comparison of these particular readings, and a rate would need readings spread over weeks rather than days. The records behind all three are published, so a later reading extends this series rather than restating it.
A column of the earlier readings beside this one: Every number this page states about an earlier reading is in the sentences above, and both of those readings are published in full at their own addresses, untouched by this page. What is not here is a table restating them: the fields that held still would be the same number printed three times, and the ones that moved since the first reading are already above with both of their values.
How often Retool’s page hands out the control that comes and goes: That it does is what this reading found: reads of Retool’s page taken in one sitting do not all give the same count, and the control they disagree about is the same one each time. How often each version is served is a rate, and a rate needs those repeat reads published as records in their own right, which they are not — what is published here is one fetch of each page, and that is what every figure above is. So this page says that count is not evidence about Retool and prints no rate for it.
What the edits actually said: We count the controls on a page and locate its prices; we do not read the difference between two versions of its words. And a control count that is not the one we read last time does not establish that anybody touched the page at all: a page can hand out more than one count of itself without anybody having touched anything, which is what going back to look found on both of the pages whose counts are not the same at every reading here. Reading the edit itself is a different measurement and this page is not it.
What a page does when a billing-period control is pressed: Our own instrument recognises a billing-period label on some of these pages and not on others, which is a limit of ours rather than anything about them, and a reading across the group is worth nothing while part of the group is beyond what we can read. The pages where it does read are already published, with their readings, on the three-product comparison.
Anything a person did: Nobody walked these pages. There is no buyer here, no model decided anything, and not one sentence on this page was written by one — so there is nothing here about where somebody got stuck, and these numbers are not evidence about what a reader would do.

THE FINDING

What this page can say that neither of the earlier readings could is which of its own numbers are facts about a company and which are facts about us — and the first answer it got was the second kind. A reading of a page is evidence about that page only if a further read of it would come back the same, and on Retool’s pricing page it would not: that page hands out a “Request trial” control on some fetches of it and withholds it on others, and that control is the whole of the gap between the three readings above. A count that comes and goes with the fetch is not an edit, so this page attributes that count to nobody. Notion AI’s we found no such control on at the time: its count held still across the earlier readings, came back lower here, and no read of it we took in that window gave a number other than that one. That was weaker than it sounds — not finding a second version of a page is not proving there is none — and the note above says what came of going back to look. So this page treats them the same: it prints both counts at every reading it has, and leaves the subtraction, and the reason to doubt it, with the reader. The page that was already the furthest of the five from saying what its price is per is no longer one of the ones holding still. There is no verdict in that and no advice attached to it: a page being worked on is not a page getting better, and a page holding still is not a page being neglected.

← Everything we have published, including the runs with a buyer in them