Two of the six tests we researched turned into pages instead of products

CogniPrep practises aptitude tests. There are 56 assessment providers on the site and 369 games behind them, and the way that number grows is in rounds: pick a handful of batteries candidates actually sit, research each one properly, then build.

The round we merged yesterday started with six. It shipped four, and 25 new games with them. The other two produced a research dossier each and no game code at all, which was the correct outcome in both cases and the more interesting half of the round.

The one that no longer exists

The CFAT is the Canadian Forces Aptitude Test. It is well known, it is heavily searched, and prep sites still sell practice for it.

The Canadian Armed Forces removed it from Regular Force selection on 31 October 2024. It was replaced by the Scored Employment Application Form, a scored questionnaire about your life so far, plus a 30 minute personality inventory called the TSD-PI. Four independent sources agree: the Department of National Defence's own release, the Supplementary Estimates, an Auditor General report, and the live forces.ca joining page, which has quietly deactivated its CFAT help articles.

Neither replacement is an ability test. There is nothing to drill. So the build was cancelled and the research turned into three things:

1. An explainer post at /blogs/cfat-replaced-by-seaf that answers the search instead of selling against it.
2. One row in our "which test am I taking" lookup table, routing CFAT searches to that post with the label replaced by the SEAF in October 2024. Not practised here.
3. Past tense corrections everywhere our own copy still described the CFAT as current, including a neighbouring provider's hub page and its research dossier.

The thing I would not have predicted is how much of that needed locking down. A page whose whole point is "we do not sell this" is one careless edit away from becoming a page that does, so it has its own test file:

it('offers no practice game, because nothing we sell practises the replacements', () => {
  expect(post.relatedGames ?? []).toEqual([]);
});

it('is not a provider and has no hub', () => {
  expect(ALL_PROVIDERS as readonly string[]).not.toContain('cfat');
  expect(existsSync(join(process.cwd(), 'app', 'games', 'cfat'))).toBe(false);
});

it('has no Canadian Armed Forces employer entry, since the association ended', () => {
  expect(EMPLOYER_ASSESSMENTS.some((p) => /canadian armed forces/i.test(p.name))).toBe(false);
});

Every post on the site can carry related games. This one asserts that it carries none. The remaining cases pin the facts the post is built on: that it cites DND, that it names the TSD-PI, that it says "Neither one is an ability test", and that it says "Nothing official settles this" about the one question nobody has answered, which is whether any Reserve or career transfer route still sets the CFAT.

A test asserting that a sentence of marketing copy exists looks odd until you have watched a hedge get edited out for reading badly.

The one that already existed under another name

The second cancellation was the Australian Defence Force. The ADF sets one aptitude test at its YOU Session, now called the Job Opportunities Assessment.

The research came back with 51 questions in 20 minutes, candidate instructions that read nearly word for word like a vendor's own, and example questions matching that vendor's published screenshots. The vendor is Revelian, whose RCAT we already built.

Our rule for that case is older than this round: an organisation that sits candidates on a covered vendor's standard test gets an employer page, not a provider of its own. Building an adf provider would have meant a second copy of a test we already have, with a different logo on it.

So the ADF became /employers/australian-defence-force, with the association recorded at medium confidence rather than high, because the match is inferred from identical published content and the ADF has never named a vendor. That confidence level is not decoration; it decides the wording. The page says the JOA "matches the Revelian RCAT" and never says the ADF uses Revelian, because we do not know the second thing.

The useful side effect is that research aimed at a provider we did not build turned out to be evidence about one we had. The ADF publishes more about this test than its vendor does, and six concrete details in that material were things our existing RCAT either got wrong or hedged, including a line on the practice page that still called a countdown unconfirmed when the ADF guide confirms it. Those landed as corrections to the test we already sell.

See it

Four checks, all on public pages, no account needed:

• Open the CFAT explainer and search the page for Nothing official settles this. Then look for a link to practice material. There is none, and a test keeps it that way.
• Open which assessment am I taking and search for SEAF. That row is the retired test pattern: the honest label sits in the row, and the link goes to an article. 153 rows on that page, of which 147 point at a provider, 4 at the Australian Defence Force page and exactly 1 at an explainer.
• Open the ADF employer page and search for matches the Revelian RCAT, which is there, then for uses Revelian, which is not. That is the medium confidence rule rendered as English.
• Then try cogniprep.app/games/cfat, the hub that was never built. It answers 200 with the title "Game Not Found", because a catch-all route under /games takes any slug it does not recognise. It is absent from all 375 URLs in our sitemap, so nothing we publish points at it, but a soft 404 is still a soft 404 and it is now on the list.

Two of six tracks ending in a dossier and no code is not waste. One of them would have been practice for a test that no longer exists, and the other would have been a second copy of a test we already had.