Why claims carry confidence labels
Hiring processes vary by country, office, role, and recruiting cycle. An employer may describe the broad stages but omit the assessment vendor; a candidate may name the exact game but only for one application. We therefore label individual claims instead of assigning one confidence rating to an entire page. The employer guides show the label beside the claim and, when available, link the source used to support it.
What each label means
Official means the claim comes directly from an employer or assessment vendor, such as a careers page, candidate FAQ, product page, press release, or branded assessment portal. Official confirms what that source published; it does not automatically prove that every office or role uses the same process.
Multiple sources means at least two independent reports support the same material point. We check whether apparently separate articles trace back to one original report before treating them as independent.
Single source means one identifiable source supports the claim. It may be useful for preparing for a possible format, but it is not presented as a guaranteed or company-wide rule.
Anecdotal means a candidate or forum report provides personal context rather than policy. These reports can reveal timing, questions, or regional variation, but dates and roles matter and individual recollections can be incomplete.
No evidence found means a documented research pass did not find support for a claim, pass mark, game name, or retake rule. It describes the search result, not proof that the thing has never happened. Where we derive a planning window from official dates rather than quote a published deadline, we use a separate estimate or inference label.
How a guide is reviewed
We start with current employer and vendor material, then compare it with dated candidate reports and other sources where the official material is silent. We record the role, region, recruiting cycle, publication or report date, and source URL when available. Employer pages distinguish current statements from historical case studies and preserve conflicting evidence instead of flattening it into one confident answer. Game reconstructions are described as independent practice tools; our scores and percentiles come from our own rules and practice pool, not a vendor’s private model or an employer’s hiring threshold.
How we calculate practice statistics
Public usage totals come from completed production practice sessions through September 24, 2026. “Candidates trained” counts distinct registered users with at least one completed session; “practice sessions” counts completed session records. Identified internal and test activity is removed, and the displayed totals are rounded down rather than projected upward.
The score-lift analysis compares the same person’s first and fifth eligible attempts in the same game and compatible mode. We calculate the percentage change from each first-attempt score to its matched fifth-attempt score, take the median across those pairs, and round down to a whole percentage. In category-wide estimates, a person can contribute a pair for more than one game or mode, so pairs are not a count of individual people.
We include completed sessions with recorded scores in selected performance-scored games and exclude identified internal and test accounts. Comparisons stay within the same scoring era; they do not cross a scoring reset or a change in how a game is scored. Personality and preference exercises are excluded.
Towers: Towers practice scores were a median 11% higher on the fifth attempt than on the first. Based on 286 players, through September 8, 2026 (snapshot read at 22:12:34 UTC).
The Towers analysis ranks completed, scored attempts from July 14, 2026 onward by completion time, with session ID breaking ties. Of 292 players who reached five attempts, six had a zero first score and were excluded because percentage change from zero is undefined. Zero scores on later attempts remain in the analysis.
pymetrics: Pymetrics practice scores improved by a median 14% over five attempts on GAP. Based on 1,153 matched user–game pairs across 10 scored practice games, through September 24, 2026.
HireVue: HireVue practice scores improved by a median 23% over five attempts on GAP. Based on 1,669 matched user–game pairs across 7 scored practice games, through September 24, 2026.
Arctic Shores: Arctic Shores practice scores improved by a median 14% over five attempts on GAP. Based on 834 matched user–game–mode pairs across 13 scored practice games, through September 24, 2026.
The September 24 snapshot, read at 17:30:55 UTC, contains 834 first-to-fifth pairs from 365 players. The unrounded median is 14.29%, and 632 of the 834 pairs scored higher on the fifth attempt. We checked all 15 performance-scored games; 13 had matched fifth attempts. Security uses only its current ten-lock difficulty profile, Direction starts after its September 3 scoring reset, and Team Sales attempts stay within the same opponent mode. Only positive-score completions enter the attempt order, and ties in completion time are broken by session ID.
For the category-wide estimates, only positive-score sessions enter the attempt order. This differs from the Towers analysis above, which retains later zero scores. The category figures are median relative changes, not arithmetic averages or changes in percentile rank.
These results describe people who returned for at least five eligible attempts, not everyone who started practicing. They show an association on Game Assessment Prep, not proof that practice alone caused the change, that employers can detect it, or that hiring outcomes improve. We do not systematically observe candidates’ final application outcomes and do not publish a hiring “success rate.”
Your progress and practice targets
Your progress charts use your own saved attempts. Different practice modes, scoring versions, question blueprints and recorded timing conditions stay separate. Questions can vary or repeat, and adaptive tasks change difficulty with your answers, so a change in raw score is not an equated change in ability.
Practice targets are specific to each assessment, mode and content version. We review published formats, scoring guidance and our own question bank, then set demanding provisional goals with a further stretch goal. Full-test goals retain the standard clock and, where relevant, require success in every section or at harder difficulty tiers. Missing evidence does not count as meeting a target. These goals are not equated to official scores or employer pass marks.
Target details show the source basis, added practice margin and limits. The 50 modeled reference points illustrate the practice scale; they are not results from people or a fitted population. They never count toward candidate, session or peer-comparison totals. Personal consistency on repeated content is rehearsal, not proof of performance on unseen questions. Preference and chance-sensitive tasks have no universal higher-is-better target.
Observed comparisons use all compatible completed practice scores, including replays. Your own attempts and known internal or test accounts are excluded. Each score has equal weight. Global comparisons require scores from at least 25 different learners; school comparisons require 10 within their 42-day window. The displayed sample counts scores, including replays. Different scoring versions and practice setups stay separate. These are practice-score distributions, not the distribution of actual employer candidates or official test norms.
How our practice tests are built
Our CCAT-style practice test uses 50 original questions on a 15-minute clock. Its question count, time limit and calculator rule follow Criteria’s candidate guidance. The mix of verbal, math and logic, and spatial questions is our practice blueprint; the publisher does not disclose the exact distribution of its official forms.
We write our own verbal items and generate quantitative, logic and figure items from original templates. A form records item identifiers and answers so it can be reconstructed for review. Worked explanations show how to solve each item. We do not publish copied, leaked or live assessment questions.
Your practice raw score is the number correct out of 50. Skipped and unanswered questions do not add points, and wrong answers do not subtract them. The results include a skill breakdown and answer review. These items have not been normed or equated to Criteria’s CCAT: a raw score here is not an official CCAT score, cannot be converted using Criteria’s percentile tables, and does not predict an employer’s decision. A practice-score comparison is available only when enough compatible saved scores exist; it is not an official percentile.
Employer-specific instructions still apply. Crossover, for example, re-verifies the CCAT with a proctored retake; its employer guide covers that process.
Your invitation overrides every guide
The invitation in front of you is more current and specific than a general page, including ours. Follow its named platform, deadline, device requirements, accommodations process, and assessment instructions. If it conflicts with this site, follow the invitation and tell us what changed.
Corrections and updates
We update claims when an official page changes, stronger evidence appears, or a source can no longer support the wording. If you spot an error, a stale process, a broken source, or a label that overstates the evidence, use our contact page. Include the page, the claim, the role and region involved, the date, and a source or redacted invitation if you can share one safely. We review the evidence and correct the page rather than treating an individual report as universal policy.