CogniPrep covers 73 assessment providers with 465 practice tests. The public
site has 1,119 URLs in its sitemap, of which 423 are per-test pages at
/games/<provider>/<test>.

465 and 423 are different numbers on purpose, because a page exists for a phrase
people search for, not for a thing we happened to build. The practice tests
themselves sit behind an account and are not indexed, so the page is the only
public artefact either way, and the question "does this get a page?" is purely
editorial.

Measured against the catalogue today:

  • 445 of the 465 tests are covered by a page.
  • 20 are not.
  • 16 pages cover more than one test, and the largest covers 6.

The 20 with no page share one property

Here is the list, by name:

TSA CBT Full Sitting CAT-ASVAB Full Battery GIA Full Battery PSET Full Mock Paper Army Cognitive Test (Full) NFA Full Written Tests CAT-Test (Full Sitting) PELLETB Full Test (online format) Scenario-Based Test (full sitting) Full POST (Sections I to III) Full EIAT (100 Questions) Full LEAB Ability Test DAA Full Assessment Full SAEE Practice Exam Future Potential Blended Assessment Full NYPD Police Officer Practice Exam Connector Ability (Full Test) Full Aptitude Test (Both Tests) Cognitietest (Full CCT) GCAT General Reasoning (NZ Police) 

Every one is a full sitting: the whole battery, in exam order, on the real
clock. These are usually the most valuable thing on the provider's hub, and they
get no page of their own, because the phrase they would be built for is one we
judged nobody types. The searches are for a section ("SAEE logic based
reasoning") or for the exam itself ("Secret Service exam"), and the second of
those belongs to the hub. A page built for a phrase nobody types has no inbound
demand and competes with the hub for the one phrase it could win.

So full sittings live on the hub, next to the sections they are made of. The
rule reads as: a full sitting gets a page only if candidates search for it as a
thing in itself, which happens, but not often.

The 16 going the other way

The mirror of the rule is that several tests share one page when they are forms
or levels of one named test that people search for as one thing. The extreme
case is the Revelian Cognify battery: six named mini-games, one page, because
the invitation says "Cognify" and the candidate searching has no idea which of
the six they will be given.

page tests covered
/games/revelian/cognify 6
/games/mckinsey-solve/redrock 4
/games/revelian/cognitive-ability-test 4
/games/clevry/verbal-reasoning 3
/games/criteria/ccat 3
/games/predictive-index/cognitive-assessment 3
/games/test-partnership/mindmetriq 3

The tell is always the invitation email. If the employer's email names a battery,
the battery gets the page. If it names a section, the section does.

Then there is what never becomes a product at all

The same policy decides what we refuse to build, and that list is explicit:
video interviews, human-marked exercises such as report writing and essays,
physical and medical stages, apparatus and psychomotor tests with joysticks or
pedals, language-proficiency tests, and questionnaires with a key the vendor has
never published. Anything we cannot score faithfully.

None of those disappear from the site, because candidates still have to sit them.
They become page content. The clearest example from the most recent round is the
unproctored biodata part of the US Customs and Border Protection exam: it is the
first of three parts, it has no published key, and CBP's own careers page tells
candidates there is nothing to study. So it is a section of the hub and an FAQ
entry, and it is not a practice test. A product there would be a guess wearing a
score.

What one round adds

The four providers in the November round shipped 16 practice tests and 14 pages.
The two without pages are the full NYPD and full SAEE mocks. Per provider, the
public surface is uniform: one hub, one page per named test, one "can you cheat
this?" page, one guide aimed at a search intent the hub cannot serve. That came
to exactly 26 new sitemap entries.

A page in this family is allowed to contain search intent and advice, and nothing
else: what the test is and how it works, a method as steps, a worked example in
the real format, tips, one short section stating our own question count and clock,
then three to five questions phrased the way people ask them. No source
narration, no visible source list, no hedging paragraphs. Where the vendor
publishes nothing, the page says "not published" rather than guessing, and where
a figure comes from somewhere other than the vendor it says "reported". A page
that could only say "not published", or that would repeat another page, is not
built. Little published information means a shorter page, not a padded one.

See it

The two halves of the rule are a browser tab apart.

  1. A full sitting with no page: open /games/saee and find "Full SAEE Practice Exam" sitting beside the three sections that do have pages (logic, language, observation). There is no /games/saee/full and there should not be.
  2. Six tests on one page: /games/revelian/cognify names all six and says so in one line: "We have all six games, played separately".
  3. The thing that is page content rather than a product: search /games/cbp for "Candidate Experience Record". It is described in full and offered nowhere.
  4. The arithmetic, if you want to check the shape claim yourself:
curl -s https://cogniprep.app/sitemap.xml \ | grep -oE '<loc>[^<]+</loc>' | wc -l # 1,119 curl -s https://cogniprep.app/sitemap.xml \ | grep -oE '/games/[^/<]+/[^/<]+' | wc -l # 426: the 423 test pages plus 3 others 

The three extra deep URLs are score benchmark pages, which are not test pages and
are a nice illustration of why counting a page family with a URL pattern instead
of its own registry gives you the wrong number.