Methodology
How we score prep platforms
10 criteria, 7 weightings, and a rule that every published figure must carry the words we read and a link to where we read them. The weights were fixed on 3 September 2026 — before any platform was scored.
Cred Farmer publishes this comparison and appears in it. That is precisely why this page exists: a ranking produced by one of the ranked parties is worth only as much as your ability to check it.
The order things were done in
This matters more than any individual weight. The 10 criteria and all7 weightings were written and committed on 3 September 2026. Only afterwards was any platform scored, and the dataset was assembled on 3 September 2026. A weighting chosen after you can already see the table is not a rubric, it is a rationalisation — and the natural temptation for a publisher who is also a competitor is to keep nudging the weights until the answer comes out right.
The rubric lives in its own file with its own history, so that claim is checkable rather than merely asserted. Any change to a weight has to appear in the changelog at the bottom of this page with the date and the reason.
The 10 criteria
Each platform is rated 0–5 on every criterion. A weighted total out of 100 is then computed at the moment you look at it, from whichever weighting you have selected — the scores are stored, the ranking is not.
- 0 — No Claude certification content
- 1 — Partial coverage of a single exam
- 2 — One exam, covered properly
- 3 — Two exams
- 4 — Three exams, or all four unevenly
- 5 — All four, in one product, at comparable depth
Scored on what one account or purchase gets you. A network of separate sister sites is not one product, and is scored as the coverage of the site being reviewed.
- 0 — None
- 1 — Scattered articles
- 2 — A guide or article set with no objective mapping
- 3 — Organised lessons, a manual or an eBook covering the blueprint
- 4 — Full curriculum mapped to the published exam objectives
- 5 — Full curriculum mapped to objectives, plus practical exercises or labs and a reference layer (glossary, cheat sheets)
Written and self-study material only. Recorded video and live instruction are scored separately under Teaching, because a 0-5 scale that had to hold both could not express the difference between 'no video at all' and 'a full video course' — and did not: it let a platform with no video top the video-led profile.
- 0 — Nobody teaches you. Written material only
- 1 — Incidental video — a walkthrough or promo rather than taught content
- 2 — Video covering part of the syllabus, or only one of the exams covered
- 3 — A full recorded video course covering the syllabus
- 4 — Full recorded video across the exams it covers, with exercises or practice built in
- 5 — Live instruction — scheduled or private classes with an instructor
Cred Farmer scores 0 here and cannot score higher while it ships no video and no instructor. This is the criterion the video-led profile is built on.
- 0 — None
- 1 — A handful of sample questions
- 2 — A bank under ~250, or a bank with no explanations
- 3 — A few hundred questions with explanations
- 4 — Several hundred, domain-mapped, with per-option explanations
- 5 — 1,000+ domain-mapped with per-option explanations, or equivalent depth per exam covered
Volume alone does not reach 4. Domain mapping and explanations are what make a bank usable. We have run no blind quality review of anyone's questions, including our own, so nothing here scores question QUALITY — only published, checkable structure.
- 0 — No mock
- 1 — Short quizzes only
- 2 — A partial-length timed paper
- 3 — One full-length timed paper at the real item count
- 4 — Multiple distinct full-length timed papers, blueprint-weighted
- 5 — Multiple distinct full-length papers per exam covered, with review, flagging and scored reporting
- 0 — No feedback beyond a score
- 1 — Overall score history
- 2 — Per-domain score breakdown
- 3 — Per-domain breakdown that routes you to targeted practice
- 4 — Objective-level diagnosis with a generated study plan or adaptive difficulty
- 5 — The above plus a readiness signal that is gated on measured performance
Level 5 requires the readiness figure to be resistant to activity — a score that rises because you read more, rather than because you answered more correctly, is scored at 4.
- 0 — None
- 1 — Bookmarks or a saved list
- 2 — A static flashcard set
- 3 — A mistakes deck that re-serves what you got wrong
- 4 — Scheduled review across the syllabus
- 5 — Spaced repetition with a published algorithm, scheduling each item individually
- 0 — Anonymous, undated, or markets leaked or recalled exam content
- 1 — Anonymous but makes no false claims
- 2 — States independence from Anthropic
- 3 — States independence and dates its content
- 4 — Cites the published exam guides or documentation it is built from
- 5 — Cites sources per claim, dates them, and publishes a correction route
Marketing questions as taken from real or previous exams scores 0 regardless of everything else. It is the one thing on this page that can cost a reader their certification.
- 0 — Paid with no trial and no published price
- 1 — Paid, no free tier
- 2 — Paid with a trial or a small free sample
- 3 — Substantial free tier behind a signup
- 4 — Free and complete behind a signup, or inexpensive one-off lifetime pricing
- 5 — Free and complete with no account required
A login wall is real friction and is scored as such, including ours. Cred Farmer is free but requires an account, so it scores 4 here and cannot score 5 while that is true.
- 0 — Individual only
- 1 — Multiple individual purchases
- 2 — Published multi-seat pricing
- 3 — Multi-seat with a cohort view
- 4 — Team management, reporting, and enterprise purchasing
- 5 — The above plus contracted delivery — private cohorts or scheduled instruction
The 7 weightings
There is no single correct weighting, which is the whole reason there are seven. A candidate sitting one exam on a zero budget and a team lead buying for fifteen people are not looking for the same product, and a page that gives them one ranking is wrong for at least one of them. Every column sums to 100; the build fails if it does not.
| Criterion | Balanced overall | I want one system for all four exams | I only need one exam | I want the best free option | I learn from video or an instructor | My exam is in two weeks | I am preparing a team |
|---|---|---|---|---|---|---|---|
| Certification coverage | 15% | 22% | 0% | 0% | 5% | 0% | 18% |
| Structured learning | 12% | 12% | 18% | 12% | 12% | 6% | 5% |
| Taught delivery | 5% | 4% | 6% | 3% | 45% | 2% | 6% |
| Practice depth | 15% | 13% | 26% | 20% | 10% | 20% | 8% |
| Mock-exam realism | 15% | 13% | 20% | 15% | 10% | 30% | 12% |
| Diagnostics and personalisation | 12% | 11% | 12% | 6% | 3% | 24% | 17% |
| Retention system | 7% | 7% | 6% | 4% | 2% | 10% | 0% |
| Evidence and transparency | 7% | 6% | 6% | 10% | 5% | 5% | 5% |
| Value and access | 7% | 8% | 6% | 30% | 8% | 3% | 3% |
| Team and operational readiness | 5% | 4% | 0% | 0% | 0% | 0% | 26% |
| Total | 100% | 100% | 100% | 100% | 100% | 100% | 100% |
Note the zeroes. Certification breadth is worth nothing to someone sitting a single exam, so it is weighted at zero in three of the seven profiles — which is the main reason Cred Farmer does not top most of them. A rubric that never zeroes anything is usually one built to protect a particular result.
The evidence rule
Every figure published about another platform carries three things: the words we actually read, a link to the page on that platform's own domain where we read them, and the date we read them. A build that contains a figure missing any of the three fails before it can ship. So does one containing a figure checked more than 100 days ago — a stale number with our name on it is indistinguishable, to a reader, from a current one.
Where a platform does not publish something, the table says Not published rather than guessing or inferring it. That is a real and common answer, and it is more useful than a confident number that turns out to be wrong.
Third-party sources are not accepted for facts about a business. A blog post saying a platform has a thousand questions is hearsay; that platform's own page saying so is evidence.
What we will not claim
Written down so that relaxing one later is a visible change rather than a quiet one.
Any pass rate, ours or anyone else's
It would need verified outcome data from a meaningful number of candidates, tied to which platform they actually used. Nobody in this market has that, and a pass rate without it is a marketing number.
That one platform's questions are more accurate than another's
That needs a blind expert review against the published objectives, run identically across every bank. We have not done one. Every criterion here therefore scores observable structure — domain mapping, explanations, item counts — never quality.
That any platform here is official, endorsed, or affiliated with Anthropic
None are, including this one. Anthropic publishes the exam guides; Pearson VUE delivers the exams.
That any platform guarantees a pass
No preparation product can, and a product that says so is telling you something about itself.
That AI assistants rank us first
It would only be meaningful with the prompt set, the models, the dates and the raw results published alongside it. Without those it is unfalsifiable.
Where this comparison is weak
- It scores what platforms publish, not what they deliver. A site that documents its features well scores better than an equally good one that documents them badly. We have no way to correct for that from the outside.
- Nothing here measures question quality. Running a blind review across every bank would be the single most useful improvement to this page, and it has not been done. Until it is, a large well-documented bank of mediocre questions scores the same as a large well-documented bank of excellent ones.
- The 0–5 ratings are judgements. The scales above are written to make them repeatable, and each rating records which facts it rests on, but two careful people could differ by a point. Treat small gaps as noise.
- Paywalled products are assessed from outside the paywall. Where a platform's best material is behind a purchase, we score its published description of that material, which may understate it.
- We are not a neutral party. Everything above is designed to limit how much that can distort the result. None of it removes the conflict.
Corrections
If a figure about your platform is wrong, out of date, or missing context, email hello@credfarmer.com with a link to the page that shows the correct position. Corrections are made from the source dataset, so a fix lands on the master table and every head-to-head page at once, and the "checked" date moves with it.
This applies to our own row too. If you think a rating for Cred Farmer is generous, that is worth an email as well.
Changelog
- — Initial rubric. Nine criteria, seven reader profiles. Weights fixed before any platform was scored.
- — Structured-learning scale, level 5: replaced 'video, labs or instructor-led' with 'recorded video or a live instructor', so that written labs and printed manuals score 4. No weight changed. Recorded because the edit was made after the rubric was frozen but before any platform — including ours — was scored, and it moves Cred Farmer DOWN from 5 to 4 on this criterion.
- — Split taught delivery out of Structured learning into its own criterion, Taught delivery, and re-weighted all seven profiles to include it. THIS WAS A DEFECT, NOT A PREFERENCE: with nine criteria, the only difference between 'no video at all' and 'a full video course' was one point on a 0-5 scale, and the arithmetic duly ranked Cred Farmer — which ships no video and no instructor — first for the reader who says 'I learn from video or an instructor'. A rubric that produces that answer is broken. Cred Farmer scores 0 on the new criterion and now loses that profile outright. Scores for Tutorials Dojo, Learning Tree and Udemy were re-read at the same time, because their Structured learning 5s had been awarded FOR the taught delivery that has now moved out.
Not affiliated with, or endorsed by, Anthropic or Pearson VUE.