How PickleCrew's guides are made and checked
These guides are written by an AI model under one person's direction and checked by other AI models. This page says what each check proves, and what it does not.
Who writes these guides, and how are they checked?
An AI model, Claude, writes these guides under Jeremy Ferguson's direction, and Jeremy decides what is published. Each page has a written contract, and an AI reviewer that did not write it grades the page against that contract. A second AI grades the design. No human expert has reviewed any page.
| Check | Who or what does it | What it proves | What it does not prove |
|---|---|---|---|
| Written contract | The builder, before building, agreed with the reviewer | The page was built to a stated list of things it must do | That the list is complete |
| Reviewer | An AI model that did not write the page, working from a written role description | A second pass against the contract, on a live preview, that can reject the page | That a person with outside expertise agrees |
| Design review | A second AI model, working from a design brief | Layout, contrast and print were looked at against a stated bar | That real readers find the page clear |
| Automated tests | Code that runs the app's own functions | Calculators, tables and counts match the app's code | That the app's code is right |
| Source log | Kept by the AI that wrote the page | Each product claim is traced to a line of the app's code, and each outside claim to a page read on a date | That the outside pages are right |
Written contract
- Who or what does it
- The builder, before building, agreed with the reviewer
- What it proves
- The page was built to a stated list of things it must do
- What it does not prove
- That the list is complete
Reviewer
- Who or what does it
- An AI model that did not write the page, working from a written role description
- What it proves
- A second pass against the contract, on a live preview, that can reject the page
- What it does not prove
- That a person with outside expertise agrees
Design review
- Who or what does it
- A second AI model, working from a design brief
- What it proves
- Layout, contrast and print were looked at against a stated bar
- What it does not prove
- That real readers find the page clear
Automated tests
- Who or what does it
- Code that runs the app's own functions
- What it proves
- Calculators, tables and counts match the app's code
- What it does not prove
- That the app's code is right
Source log
- Who or what does it
- Kept by the AI that wrote the page
- What it proves
- Each product claim is traced to a line of the app's code, and each outside claim to a page read on a date
- What it does not prove
- That the outside pages are right
What these checks do not prove
The limits come first, and they are not softened.
- The reviewers are AI models. They work from written role descriptions. They are not human experts and not outside auditors.
- Some reviews used a stand-in. During this build, a general-purpose model stood in for the named reviewer, because the configured reviewer could not reach the app. The reviews so far, including the agreement on what these four guides must do, were run that way.
- The printables have not been tested on paper. The sign and the checklist were designed, measured and viewed on screen. Nobody has printed them and read them from across a court yet.
- Rules and sources have dates. Rules are as of the 2026 USA Pickleball Official Rulebook, read on November 2, 2026. Outside pages are as of the day we read them.
- PickleCrew's rating and sit-out behaviour are its own. No rating owner and not USA Pickleball endorses them.
- No referee or coach has reviewed any page.
- Search data does not judge accuracy. Rankings, clicks and search-console numbers are not used to decide whether a page is right.
What happens to a page before it goes live
- A contract. Before building, the builder writes what the page must do and how that will be checked. The reviewer questions it until both agree.
- A build. The AI builds the page from the contract in a private copy of the site.
- A review on a live preview. The reviewer opens the page in a real browser at phone and desktop widths, in light and dark mode, and may reject it. A rejected page goes back to be reworked, and the reviewer has rejected pages on earlier guides.
- A design review. A second AI looks at layout, contrast and print.
- Sources. Each product claim is traced to a line of the app's code. Each outside claim is traced to a page read on a date, or removed. Calculators, charts and tables are generated by code and checked by automated tests against the app's own functions.
- A person publishes. Jeremy decides when a page goes live.
Why a model checking a model is not enough
Google's guidance on generative AI content asks publishers to focus on accuracy, quality and relevance, and to manually factcheck and review AI-generated content before publishing (page last updated October 1, 2026). Most of our checks are made by AI models, so we do not claim that a person checked every claim by hand. The limits above are the honest version of that gap.
What we have not verified yet
As of November 2, 2026 there are 8 open items for the organizer guides, listed below, and 17 older open items in our source log for earlier guides. Before we publish, a script compares this list with our source log.
- The printable sign and the checklist have not been printed on paper and read from across a court.
- The wait table is a model. No real open play session was timed to check it.
- App behaviour was read from the app's code, not run. Running it would write to the shared database, so the waitlist, quick play and invite flow were not exercised.
- The sign's plain-English lines summarise the rulebook's rules. No referee has confirmed the wording.
- Four pages we tried to read were blocked (Salt Lake City's and Knoxville's sheets, Playtomic's and Pickleball.com's guides), so they are not used.
- The published pages' server HTML, rich-result checks and share images have not been checked on the live site.
- The questions in the FAQs come from one day's search suggestions. That shows people type them, not how many.
- The keyboard-only walk, reduced-motion check, measured contrast and a real-phone check are not all done, and the reviewers' final grades are not in yet.
Byline, dates and corrections
The byline "PickleCrew editorial" is a label for the site, not a team of people. Each page shows the date it was last updated.
To report a mistake, email support@picklecrew.net, which the support page lists. We correct the page and change its Updated date.
What changed: November 2, 2026, the first version, published with the first organizer guides: how to start a pickleball group, how to organize pickleball open play and pickleball etiquette.
Frequently asked questions
Has an expert reviewed these guides?
No. No certified referee, coach or other expert has reviewed them. The reviewers are AI models working from written role descriptions, and the limits section above says what that does and does not prove.
Who is the byline PickleCrew editorial?
A label for the site. The pages are written by an AI model under Jeremy Ferguson's direction, and Jeremy decides what is published.
How do I report a mistake?
Email support@picklecrew.net or use the support page. We correct the page and change its Updated date.