Methodology version 2.1.0. Pilot values: weights and thresholds are tuned as outcome data arrives.
Every area score is the probability-weighted estimate of what you know, corrected for guessing, with an 80% range. Six questions cannot pin your level down, so the range is wide; thirty can, so it narrows. Each area carries an evidence label (Low, Developing, Strong), and a score stays “Provisional” until every area has enough evidence.
On a four-option question a pure guess is right a quarter of the time. We subtract that chance, so random guessing scores about zero rather than 25. Numeric and written answers have no chance correction.
The diagnostic and reassessments use fresh questions under timed, no-feedback conditions. Workouts teach and schedule review but do not change your headline score, so it cannot be inflated by practising what you have just been shown.
Each target is assessed separately, and your history is kept per target. Technical readiness blends the areas below. Behavioral readiness is the average rubric score across your latest answer to each prompt, shown once you have answered at least four, including the core motivation and CV prompts. Cases are reported separately again. They are never averaged together.
The top tier needs a technical score of 75 or more, a lower bound of 55 or more in the core areas for your target, developing evidence in every area, a behavioral score of 70 or more (not required for quantitative research), and recent worked demonstrations of the key calculations for your target. The dashboard shows which of these you have met. No overall label is shown while a mandatory area has no evidence. The benchmark is our own pilot standard: it describes your preparation and does not predict an interview or hiring outcome.
A change is called real only when the old and new 80% ranges do not overlap. Otherwise we say it is within noise, so you do not over-read a bounce of a few points.
Numerical answers are checked by exact arithmetic against a stated tolerance, never by an AI model. Percent answers are in percentage points (enter 16.7 for 16.7%). If a later step is right from your own earlier answer, you get half the marks for it, so one slip is penalised once, not at every step. A blank answer is not the same as zero, and we never guess what an unreadable entry meant. Written answers get feedback from an AI model against a fixed rubric, scored 0 to 2 per criterion, and each credited point must be backed by a line from your own answer. If the grader is unavailable your answers stay saved with the numerical result and feedback follows; a technical failure is never scored as zero.
You see numerical accuracy and written feedback side by side. A simple case practice score mixes them 50/50 for that one exercise. It is not a readiness score and is not added to your technical score. Only your first attempt on a case, on reviewed content, counts as independent evidence; repeating a case or its variants is practice.
Each answer is scored 0 to 4 on five criteria: structure 15%, specificity 25%, ownership and judgment 25%, result 20% and reflection 15%. Motivation and situational questions use their own anchors for the same five criteria.
New questions and cases start as drafts and are published only after a practitioner has reviewed them. Where reviewed content for your target is limited, we show a shorter practice list and say why, and we do not offer a diagnostic until it can be a fair one. Questions written for another target are never used to fill the gap. Draft content is only ever shown to accounts that have been given preview access, it is labelled, and it never counts towards your results.
These scores are a model of your preparation, not a validated prediction of hiring outcomes. The behavioral grader is an AI model working from a published rubric; it can be wrong, and its feedback is guidance, not a verdict.