Before Reveal Day, this site sold a forecast of the 2026 round: which strings would be applied for, by whom, and how many applications there would be. On 7 October 2026 ICANN published the real list. This is the forecast scored against it, with nothing rewritten after the fact.
Volume
The forecast published a range, not a number: a named floor of about 1,049 applications and a top-down scenario of about 2,018, presented on its own page as the upper end of the range and not a point prediction. Its front page gave 1,244 (base) to 1,821 (high). 1,614 applications were filed, inside both ranges.
One of the pre-registered hypotheses went further and bet that the total would exceed 2,018. That bet is lost; see H4 below.
Strings
986 distinct strings were filed. 182 of them were in the forecast on one track or the other: 18.5%. The named track (a string attached to a named organisation) had 118 of them; the generic track had 95.
Read the other way, most of what the forecast listed did not arrive: 118 of the 2,365 named-track strings were filed, and 95 of the 502 generic-track strings.
The forecast did worse on declared brands (38 of 332 brand strings seen) than on the rest (144 of 654). A catalogue built from named organisations missed most of the brand strings that were actually filed.
Who files what
The forecast's central promise was pairs: this organisation will apply for this string. Of 2,708 forecast pairs, 18 were filed by the organisation named. That is 0.7%.
Calibration tells the same story. Among the 54 pairs given a joint probability of 60% or more, the string was filed by someone in 14 cases and by the named organisation in 5.
Contention
263 strings have two applicants or more. The forecast's public table marked 102 strings as contested; 12 of those really are. 49 of the 263 contested strings appeared in the table at all.
One thing held up better: of the 23 generic strings flagged as headed for an auction, 14 are contested.
The four pre-registered hypotheses
Four hypotheses were written down and frozen on 15 August 2026, two days after the first public disclosure by an applicant (316 strings) and before ICANN published the list, with the rule that each would be judged as written.
- H1 — registry operators are mis-modelled as brands. Not judged. Judging it requires matching the 11 applicants with 20 filings or more to the forecast's organisations through the alias table. That matching was not done at Reveal.
- H2 — our contention counts are floors. Refuted as written. On the 116 strings of our table that were filed, the real count was above ours 43 times, equal 61 times, below 12 times. The direction held; the claim said "systematically", and it was not.
- H3 — generic recall is bounded by topicality. Not judged. It needed a semantic-zone assignment fixed before Reveal Day, and none was recorded.
- H4 — volume overshoots, driven by portfolio filers. Refuted. It required more than 2,018 applications; 1,614 were filed. Its second leg held: the 4 applicants with 50 filings or more account for 27.6% of the round.
Two refuted, two not judged, none confirmed.
What was wrong
The forecast reasoned from organisations to strings: who has a brand, a dispute history, a reason to apply. The round was filed the other way round. Four applicant names account for more than a quarter of it, portfolio applicants file common words by the dozen, and 443 applications share their mission wording with five or more others. None of that is visible from a list of organisations.
The catalogue also had no way to see an applicant it had never heard of.
What happens to it
The forecast pages stay online at forecast.tld.report, unchanged, with a banner pointing here. Nothing in them has been edited to look better in hindsight, and the frozen hypotheses file is the one committed on 15 August.
This site now does something else: it reads the filings.
Sources
- ICANN Reveal Day export of 7 October 2026 (2,783 rows: 1,614 primary applications, 1,169 replacement strings).
- The forecast as frozen: catalogue v7, generic track, public contention table.
- The pre-registration file
data/prereg_hypotheses.md, committed on 15 August 2026 and unchanged since (sha256cb3cde4e7f36c4e2e033d71b21218ee8298bd32dc3b2bf41008fe968ccd15887), and the scoring scriptscripts/score_reveal_vs_predicted.py. Both are in this site's repository, which is not public. - The forecast as it was sold.