The record
What changed, and when
The instrument is versioned, and its descriptors can change under somebody who has already answered. So every change to it is dated here, along with the articles it was built out of. The articles came first, over five months. The instrument came after, in three days.
2 September 2026
The instrument
The instrument gets a public history, and asks how to be better
This site argues that a claim nobody can check is not a claim, and the instrument was the one thing on it with no public record. This page is that record: every deploy since the first, and the publications it was built out of, on one timeline because the sequence is the point. The deploy counts are the project's own pipelines, and the sentences are written, because neither a commit nor a push records what changed for the person reading the page. The report now also carries the only thing this site has ever asked a visitor for: whatever would make the instrument better, from somebody who has just finished using it. That ask is why the contact address changed too. It is no longer assembled when the page loads, so it is not sitting in the document waiting to be harvested, and it appears only once a person clicks, types or touches something.
1 deploy
2 September 2026
The instrument
The first reader from outside, and the two questions he could not answer
Somebody outside the project ran all twenty questions and exported a PDF. Two of them stopped him, and both times he was standing on the answer: not knowing is a 1, and the rule that says so was stated once, on the scope screen, where it had long scrolled off. So the rule is now on every question screen, each dimension asks its question once where it can actually be seen, and the report draws your own twenty scores instead of an example nobody scored.
See the profile on your report1 deploy · 11:00
1 September 2026
The instrument
The landing page leads on the codebase, not on delivery
It opened on what AI does to delivery speed, which is the argument every other site is already making. It now opens on what AI puts into the codebase, and it rests on six pieces of published research instead of one. A page whose whole claim is that other people's rigour is checkable has to survive being checked itself.
Read the landing page2 deploys · 08:26 to 09:55
31 August 2026
The instrument
One anonymous row per completed assessment
A finished assessment sends twenty scores, the scope, and the industry and size the visitor chose to give. No email, no identifier, and no field that could join two rows to one person. Nothing leaves the browser until an assessment is complete, the scope name you type never leaves it at all, and what is sent is the report you are looking at, checked against storage as it was read.
Read what is stored and sent2 deploys · 22:43 to 22:57
31 August 2026
The instrument
The report leaves the browser as a PDF
The same nodes that are on screen are laid out into A4 pages, so no part of this is designed twice, and the export is never behind an email address. The pages are shown before the print dialogue is offered: a modal in front of a document nobody has read yet makes dismissing it feel like cancelling the export.
Export one3 deploys · 22:01 to 22:11
31 August 2026
The instrument
All twenty dimensions, and a page for each
The remaining fifteen were authored, which is the majority of what this project costs, and every dimension got a permanent page carrying its argument rather than only its ladder. The flow started asking industry, size and country before question one, and the site started answering crawlers and answer engines in plain text rather than leaving them to guess.
Start the assessment11 deploys · 20:20 to 20:55
31 August 2026
The instrument
Five dimensions, scored end to end
Scope first, then one question at a time, then a report that refuses to name a floor from five of twenty. Two reading formats were built and put side by side: the whole situation in prose on every rung, against the dimension explained once and five rungs of a line and an observable. The second shipped and the first was deleted rather than kept as an option.
How it is scored1 deploy · 19:22
31 August 2026
The instrument
The site exists
A landing page arguing that your readiness is your floor and not your average, the ladder it is scored on, the method published before the instrument, and the legal and AI content pages. The build checks shipped with it: a price, a leaked contact address or a missing doctrine line fails the build rather than reaching the site.
Read the method2 deploys · 17:34 to 17:38
25 August 2026
The series
Twenty Questions, One Honest Number
The piece that proposed this instrument. One ladder, defined once, for all twenty dimensions, and the rule the site is built on: your floor is the verdict, because a total lets a 5 somewhere else disguise a 1.
18 August 2026
The series
Converting Cultural Agreements into Hard Guardrails
A cultural agreement and a hard guardrail are the same rule at two different altitudes. The piece asks what it would take to move each of yours down to the altitude where obeying it is not a choice.
11 August 2026
The series
Architecture enforcement in CI
Does the build fail on an architectural violation anywhere in your scope, or do you rely on a reviewer noticing the one line that inverted a dependency?
4 August 2026
The series
Documentation integrity
If a new developer or an agent followed the README of any repository in your scope from a clean machine, would they get a running system, and would you bet on it?
28 July 2026
The series
Dependency hygiene
Of the dependencies running in production anywhere in your scope, how many did a human actually choose?
14 July 2026
The series
Release discipline
When something you released last month changed behaviour, how did the people depending on it find out? And would that still work at the rate your scope now ships?
7 July 2026
The series
Change absorption
Of everything merged anywhere in your scope last month, how much was reverted or rewritten within two weeks, and is that share rising?
30 June 2026
The series
Semantic drift
Does the same concept have one name across your scope or several, and do you know whether it is worse than a year ago?
23 June 2026
The series
Organizational design
Pick a recent feature that crossed team boundaries anywhere in your scope. If AI doubled what every one of those teams produced tomorrow, would it arrive sooner, or would the queues between them simply grow?
19 June 2026
The series
The Hard Part of Enterprise AI Was Never the Models
A field report rather than a dimension: three days inside one enterprise software house, and what it took to move AI out of individual teams and into the organisation around them.
16 June 2026
The series
Cognitive load per module
Does anything in your scope know how tangled its worst module is, or only your people?
9 June 2026
The series
Hard guardrails vs cultural agreements
Take the standards your teams rely on anywhere in this scope and sort them into two columns: the ones a machine refuses to allow through, and the ones people have agreed to. Do you know how long the second column is?
2 June 2026
The series
Test meaningfulness
Take a function anywhere in your scope, move a boundary by one or flip a comparison, and run the suite. Would anything fail, or would it stay green?
26 May 2026
The series
Velocity and the constraint
If merge-request volume doubled tomorrow anywhere in your scope, which stage fills up first? And do you know that, or are you guessing?
19 May 2026
The series
Decision memory (ADRs)
Pick an architectural decision taken anywhere in your scope in the last two years. Can somebody find the reasoning in the repository, or only in the person who made it?
12 May 2026
The series
Security assumptions
Pick a security assumption your teams hold anywhere in this scope. Where in the codebase would an agent learn it?
5 May 2026
The series
Trunk-based development
How old is the oldest branch alive anywhere in your scope right now, and what would it cost to merge it today?
1 May 2026
The series
Requirements quality
Could an agent pick up the top item of any backlog in your scope and build the right thing without guessing?
28 April 2026
The series
Code ownership and review routing
If the busiest team in your scope vanished tomorrow, how many open changes would have nobody left qualified to review them?
21 April 2026
The series
Domain language consistency
If you asked three teams anywhere in your scope to define your most contested domain word, would you get the same answer?
14 April 2026
The series
CI reliability
Somewhere in your scope a pipeline went red this week. Did anybody investigate it, or did they press rerun, and would you defend that choice?
7 April 2026
The series
Technical debt
Somewhere in your scope is a system everybody agrees needs fixing. Does your AI plan carry a line item for fixing it, or an assumption that the acceleration will pay for it later?
31 March 2026
The series
Repository structure
Across everything in your scope, does your repository setup help an agent reason about the system, or force it to guess?
The entries at the top of this page exist because one reader wrote in about what had been hard to use. If you have something that would make the next entry better, send it by email or on LinkedIn.
Nothing on this page is generated. Git records commits and GitLab records pushes, and neither of them records what changed for the person reading the page, so the sentences here are written and the deploy counts come from the project's own pipelines.