Index updates: dated reviews of the AI detector data

An update is a dated page that says what changed in the index over a stated period, in the same fields the cards and records use: cards added, study records added, figures that moved on a card, rankings whose order changed, and corrections filed. The first update is the 2026 review of every published false positive figure, written on 9 November 2026 at index version 2026.11.3, followed the same day by the Q4 2026 index update and studies digest at version 2026.11.13, with 72 cards and 35 study records in the index. Quarterly updates continue from the first quarter of 2027, one on the index data and one on the studies added, and an annual review closes each year. Every update carries a register ID in the AIDI-UP series, a year in its slug because it is a dated item, and a revision log of its own.

The pages here are not news. A vendor's launch post, a price change or a renamed brand is recorded on the card the day it is read, with the source and the date, and the corrections log carries the dated line; an update gathers those lines over a period and sets them against the study records, so a reader can see in one place whether the evidence moved with the claims. The 2026 false positive review, for example, lists the 28 rows of false positive figures the 35 records print for twelve of the 72 detectors, from 0 of 40 human papers for four tools in Van Vlasselaer 2026 (DOI 10.1007/s40979-026-00226-w) to a 0.9503 rate for the RoBERTa-base baseline in Jabarian and Imas 2025 (NBER w34223), both read 9 November 2026, and then records that Pangram moved from the vendor table into the false positive rate ranking on the strength of the second paper.

An update is built from the index data file and the study records, and from nothing else. Where a figure on an update differs from the figure on a card, the card is wrong or the update is, and the correction is made on both the day it is confirmed. Counts on an update are the counts at the version named in its first paragraph, so a 2026 review that says twelve detectors have study-sourced false positive data is a statement about version 2026.11.13, and the next review will carry its own version and its own count. The methodology page holds the rules an update follows; the studies index holds every record an update cites; and the home board shows the AI detector index the updates describe.

Readers use the updates in two ways. A procurement team re-reading a shortlist a year after writing it can take the annual review and see which figures on its cards came from a 2023 paper on 18 documents and which from a 2025 paper on 1,992 passages, and whether a 2026 study has since tested the tool on long papers. An editor or a journalist quoting a false positive rate can take the table row, which names the study, the base, the threshold and the date in one line, and quote it whole. Each update ends with three questions answered in their first clause and a Sources list with the read date of every URL.

The table below lists every update with its register ID, the year it reviews, and how many cards, records and questions it covers. The list grows by two each quarter and one each year; nothing is removed, and a superseded update keeps its page with a line pointing to the one that followed.

The twelfth wave on 10 November 2026 closed the year with three pieces: the 2027 census, the annual census that counts the cards, the gone, the tested, the open and the held, with the study index by year and kind; the Q3 2027 index update, covering 21 cards, two fields and two held rows across the tenth, eleventh and twelfth waves; and the Q3 2027 studies digest, the five records added with their figures and bases. The census is the annual piece for 2027 beside the 2026 false positive evidence review.