DEV Community

Daniel Pertu
Daniel Pertu

Posted on

Our research said build three products, we shipped two, and the test section our competitors still sell was deleted in February 2024

CogniPrep sells practice for assessments that employers set. Adding one means a
research pass first: one worker, one markdown file, no code, and a rule that
every factual claim carries the URL it came from plus a confidence tag. The
file for the US Customs and Border Protection entrance exams came out at 1,237
lines with 30 sources, 6 of which are rival prep sites tagged Low, and a
separate list of 6 URLs that were blocked or empty and therefore not used as
evidence at all.

Here is what that bought.

1. It cut the product down

The dossier's own recommendation was three practice tests: CBP Officer Logical
Reasoning, CBP Officer Writing Skills, and Border Patrol Logical Reasoning. We
shipped two.

The reason is in the dossier's own evidence. The Border Patrol test is the same
logical reasoning instrument as the officer one with a longer clock: same
lead-ins, same five options, same adaptive rules, same vendor, and seven of the
same eight practice items in CBP's published guides. A third product built on
that would have been the second product with a different name on it, and the
house rule is that every provider costs the same, so a longer lineup cannot even
be justified by price. Border Patrol became a choice inside the officer test
instead.

Two other things in the plan died the same way:

  • The Candidate Experience Record, the unproctored biodata part, is not a practice test on our site. There is no published key, and CBP's own careers page tells candidates "No need to study!". It is page content and an FAQ.
  • A combined full exam sitting, which is the obvious upsell, does not exist, because the real parts are taken days apart: writing within 5 days of the experience record, logical reasoning within 14 days of the writing test. Putting them on one clock would have invented a session nobody sits.

2. It found that the market is selling a dead exam

The exam changed on 1 February 2024. The old battery was three timed sections:
logical reasoning 81 minutes, arithmetic reasoning 87, writing skills 47, with
a calculator supplied. The current one, from
CBP's own preparation guide
and its
careers page,
is an unproctored experience record, then a writing test of 63 minutes, then
logical reasoning of 83, the last two adaptive. Arithmetic reasoning is gone.
Results from before that date are no longer valid.

I checked the six rival pages again on 20 November 2026, two years and nine
months after the change. Two of them still describe the old battery to the
minute:

  • psychometric-success.com: "Logical reasoning 81 minutes, Arithmetic reasoning 87 minutes, Writing skills 47 minutes", plus a five-minute break between logical and arithmetic reasoning.
  • practice4me.com: "The arithmetic reasoning section is the longest test section of the exam at 87 minutes."

A third still sells drills for the deleted section:
jobtestprep.com lists
"Arithmetic Reasoning: 24 Practice Tests + 3 Study Guides to Brush Up on your
Math Skills" in a $89 course, and tells candidates "Most candidates must score
at least 75% to pass". The published rule is a score of at least 70 on a scale
whose range CBP does not publish, which is a different statement in two ways.

The first two also tell candidates they can retake after six months. The
official retest interval is 90 days.

This is the whole argument for paying for research before building: the cheap
way to write these pages is to read what the top-ranking pages say, and the top
ranking pages are describing a test that was retired while they ranked.

3. It wrote down what we are not allowed to say

The confidence split in the dossier is more useful than the facts:

  • High: part structure, timings, adaptive rules, the writing test being a gate, the pass score of 70, the 90-day retest, delivery.
  • Medium: question counts. Exactly one official count exists anywhere, 31 for the 2018 Border Patrol test. Current counts are inferred from CBP's own pacing advice.
  • Low or unknown: the score scale, how a score maps to referral, the adaptive internals, and the interface, because no screenshots are published.

That split becomes copy. Our question counts ship as configuration marked as our
estimate, and the instructions say CBP does not publish a count. Our score is
labelled as ours, not CBP's 70, on every surface that shows it. There is no pass
line anywhere in the product, because we would be inventing the boundary.

4. It corrected the pages we had already published

Dossier structure has a section 12 called "cross-provider findings", and its job
is to report things the round broke rather than things it built. The NYPD
research pass in the same round filed four corrections against a dossier written
in an earlier round, which had the NYPD exam as a comparison case: six of its
nine published abilities overlap the New Jersey test's, not five; the fee is
$0, not $40; candidates must be under 43 on the day they file; and the current
exam series runs to July 2027, not July 2026. The playbook sentence that
produces this is blunt: starting leads are leads, not facts, and every round so
far has found serious errors in its own brief.

70 of these files are committed, 831,498 words of them, and not one word is
rendered on the site. They exist so that the pages can be short.

See it

The two tests that survived are at
cogniprep.app/games/cbp. Three checks you can
run on that page:

  1. Search it for "Arithmetic". The only mention is an FAQ answering "Is there maths on the CBP Officer exam?" with "No", and the date it was dropped.
  2. Search it for "70". It appears as the official pass rule, next to a note that our own score is not CBP's.
  3. Count the practice tests: two, not three, and the Border Patrol version is inside the logical reasoning one.

Then compare the two per-test pages,
logical reasoning and
writing skills, against
the rival pages linked above, and
/blogs/border-patrol-entrance-exam
for the part of the market where the dossier found no competitor selling
anything specific at all.

Top comments (0)