Why Claude practice scores swing between tests, and what one bad paper means

Checked against the official sources on 6 November 2026: Anthropic’s certification FAQ and the four v1.0 exam guides. Points per question are arithmetic from the guides’ item counts.

Why do my practice scores swing so much from one test to the next?

Scores on practice papers for a Claude exam swing for four reasons you can check. A paper is a small sample, forms may differ in difficulty, a repeat sitting carries memory of the items, and an unfinished paper counts its blanks as wrong. None needs your knowledge to have changed, so the size of a swing alone proves little.

• Small samples. A paper holds a few dozen items from everything the blueprint lists. A different draw holds different items you happen to know. Cred Farmer’s 30-item quick sitting is a different draw each time.
• Different forms. Anthropic’s FAQ says scaled scoring “equates scores across exam forms that may have slightly different difficulty”. A raw practice count has no such adjustment, though Cred Farmer’s signed-in forms can show that form’s learner average.
• Repeat sittings. A re-sit has already shown you the items, so it cannot be read beside a first sitting.
• Unfinished papers. On Cred Farmer’s papers an unanswered question is marked wrong, so count what you answered first.

Inconsistent mock exam scores are weak evidence until you know which of these you are looking at.

Claude practice scores: what one question is worth on each paper

One question on a practice paper for a Claude exam is worth 100 divided by the item count, in percentage points: 1.67 on 60 items, 1.89 on 53 and 1.59 on 63. Three questions either way is about five points, and five questions either way is about eight or nine.

The last two rows are Cred Farmer’s shorter free papers, so never set one beside a full-length paper. Three questions is a yardstick, not a measured margin: no official source says how far scores move by chance.

Cred Farmer’s signed-in report adds a second number: a simulated score on a straight line from 100 to 1,000 and a Pass or Not Pass grade against 720, neither calibrated against the real exam. One question is 15 simulated points on 60 items, so three either way is 45, which can flip that grade.

A domain line moves further: see what a Claude exam domain score means.

How do I keep one bad practice test from wrecking my confidence?

To keep one bad practice test from wrecking your confidence in a Claude exam plan, treat it as one reading. Convert the drop into questions, check it was a finished first sitting, read the raw count by domain weight, then plan seven days from that. The order is Cred Farmer’s advice, not an official rule.

The bands and their seven-day plans are in how many mock tests to sit, and the weights are in Claude certification domain weights. A low first sitting still shows what to study next.

This is exam technique and nothing more: anything beyond exam nerves is a matter for a professional. If a condition affects how you sit an exam, Pearson’s accommodations process is the official route: see ID and accommodations.

Where did my confidence go between the practice tests and the real exam?

Confidence built on practice papers can fade at a Claude exam for reasons you can check, though nothing published measures that fade. Repeat sittings flatter a practice score, the real form is unseen, and 720 is a scaled score rather than a percentage, so the two numbers never shared a scale.

Repeat sittings. Cred Farmer’s stats page counts only a learner’s first answer to each question and says “Repeat sittings score several points above first sittings, which is memory rather than learning”. Read on 6 November 2026, snapshot of 6 November 2026.

That stats line describes Cred Farmer’s practice bank and a self-selected group: not the real exam, not a pass rate, with no outcome or advice attached. Cred Farmer’s advice, which that line does not prove, is to keep first sittings and repeats in separate columns.

The unseen form. Anthropic’s FAQ says its earlier practice exam was retired in the move to Pearson. Cred Farmer writes its own items from the exam guides, and their difficulty is not matched to the exam. No practice paper previews the real form.

The scale. The real exam reports 100 to 1,000 with 720 to pass. Anthropic publishes no conversion from raw answers, so a practice percentage cannot be set against it: the Claude certification passing score explained says why. If nerves on the day were the problem, the plan for Claude exam nerves has a routine.

Which swings mean something: patterns that survive a change of paper

A swing on a Claude exam practice paper means something when the same pattern survives a change of paper. That can be the same heavy domain missed on two different forms, repeated misses on multiple-response items, or blanks on two papers. A total alone says less than any of these. They are Cred Farmer’s advice, and none predicts a result.

• The same heavy domain on two different forms. Different items and the same gap are harder to put down to the draw. Heavy means a large weight in your exam guide: a small domain’s line moves too far on one question.
• Repeated select-count misses. All four v1.0 guides say each item states how many responses to select, so a wrong count is avoidable. Cred Farmer’s papers mark such an item right only when the whole set matches; no guide describes partial credit on the real exam. If this keeps happening, drill the count check.
• Blanks on two papers. Blanks on two first sittings may point at the clock: see pacing on the Claude exams.

How many papers settle it is not published. Cred Farmer’s mock plan uses three first-attempt sittings.

What to write down after every paper, and why the reason comes first

After every practice paper for a Claude exam, write down the date, the paper, whether it was a first sitting and how many items you answered. Add the raw count by domain, and the reason for each miss before you read the rationale. Cred Farmer’s free runner saves no result, so your copy is the record.

The reason comes first because hindsight rewrites it. Common Claude exam mistakes has the log and its labels.

Then read the log, not the totals. Does the same domain keep appearing? Does the same reason? The Claude certification study guide shows where a timed sitting falls in a plan, the practice exams guide lists the papers, and when to book the Claude exam shows how a sitting feeds a date.

Key takeaways

• One question is 1.67 percentage points on a 60-item practice paper. It is 1.89 on 53 items and 1.59 on 63, so three questions either way is about five points.
• One question on 20 items is worth three on 60 items. One question is 5 points on Cred Farmer’s free CCAR-F mock, and three questions are 15.
• A repeat sitting does not belong beside a first sitting. Cred Farmer’s advice is to treat a re-sit as revision.
• No practice percentage converts to the 720 passing score. Anthropic publishes no conversion from raw answers to the 100 to 1,000 scale.
• A pattern that survives a change of paper means more than a total. Cred Farmer’s signals are the same heavy domain on two forms and repeated select-count misses.