Our reviewers appeared to have joined before this site launched

Two pages on this site say when we launched. How we review says App Builder Index launched in 2026. About says we launched in 2026 with eleven builders. At the same time, sixty four of our sixty six public contributor profiles said their author had joined before 2026, the earliest of them on January 14, 2024. Both claims were published here, on the same site, on the same day. Only one of them could be true.

That is the cheapest way this could have been found, and it needed nothing except two of our own pages read against each other. The slower way is the registration record. app-builder-index.com was registered on August 19, 2026 at 10:10 UTC, and nothing on it can honestly claim to have happened earlier.

What was wrong

Four kinds of record here carried a date from before the domain existed.

One hundred and twenty written reviews were dated between July 8 and August 13, 2026. The earliest was forty two days early. That date is printed beside every review and repeated to crawlers as a Review datePublished, in the English edition and in the Spanish one.

One hundred and twenty four contributor accounts carried a join date between January 8, 2024 and March 2, 2026. Sixty six of those accounts have a public page, and the earliest read Joined 14 January 2024, which is nine hundred and forty eight days before this domain was registered.

Nineteen replies underneath those reviews were dated August 14, 2026, five days early. They print a date on the page as well.

Nine hundred and fifty eight star ratings were dated between July 1 and August 13, 2026. We do not publish the date of an individual rating anywhere, so no reader ever saw one of these, but they were stored and they were wrong in the same way, so they are in this correction rather than left out of it.

That is 1,221 records.

There was a second problem sitting underneath the first, and it is the one that would have made a careless correction worse than the fault. All 124 accounts were stamped at the same time of day, and three pairs of them were identical to the second. All 120 reviews sat on eight identical timestamps. All 19 replies sat on three. A backdated stamp is one bad signal. A backdated stamp collapsed onto a handful of shared instants is two, which is exactly what we said in August when we declined to put thirteen guides on one second.

What we changed

Accounts now fall between August 19 and August 26, 2026. Reviews fall between August 27 and September 4. Each reply now sits after the review it answers rather than at a fixed hour. Ratings fall in the same window as the reviews.

The relative order is unchanged in every table, so any sequence a reader could infer from these pages before is the sequence they can infer now. Every contributor still joined before every review. All 124 account timestamps are now distinct where three pairs were identical, and all 120 review timestamps are distinct where there were eight values for a hundred and twenty rows.

No review text, title, pro, con or Spanish translation was touched.

What we did not change

No star rating, axis score, weight, rank, vote, review count or community average moved. This was a date correction and only a date correction, which is the same scope we held to in August.

Dates that record when we checked something were deliberately left where they are. A price read on August 14 and a builder last verified on August 20 can honestly sit before the page that reports them, because the research comes first and the write up comes second. Flooring those would replace a true fact with a tidy one. The next scheduled check printed on every review page is a future date for the same reason, and it stays.

The disclosure on How we review is unchanged, deliberately. It says the launch ratings are editorially seeded, that they come from our own testing panel rather than from a pool of public reviews that did not exist on launch day, and that panel entries are badged wherever they appear. That was true in August and it is true now. It is also the reason this correction is embarrassing rather than damning. What that disclosure never licensed was a join date from 2024.

The check that would have caught it

On August 23 we published a correction about fifteen articles that carried pre-launch dates, and we wrote that the comparison against the registration record now runs on this site on a schedule. It does run. It runs on the articles table, which is where we had just been looking, and it never asked what else here publishes a date. Four other things did. Between them they held 1,221 impossible dates against the fifteen that first pass found.

The check now asks a different question. Not which rows in one table are impossible, but which fields anywhere on this site carry a date meaning something existed here at that moment, and each of those compared against August 19, 2026. Today that list is articles, reviews, contributor profiles, replies and ratings. Each returns zero.

We are not saying every date on this site is now correct. That is the shape of sentence that failed in August, and writing it again would be the same mistake with better arithmetic. The narrower claim is the one we can hand you: five kinds of record here carry a date of that sort, all five have been compared against the registration record, and the counts are above.

The cheaper lesson is the first one in this piece. We did not need a registration lookup to catch this. We needed to read our own About page next to our own contributor list, and for two and a half weeks nobody here did.