How these assessments actually work

MindScore builds performance tasks and structured self-report assessments, scores them deterministically and writes back a plain-language interpretation. It is a consumer education product.

MindScore is not a clinical or medical instrument. It does not diagnose any condition, it does not screen for one, and it is not a substitute for assessment by a qualified professional. It is also not designed or validated for employment selection, academic placement or any other consequential decision about a person.

The challenge is a simple visual-search task. After an unpredictable delay of 900–2400 milliseconds, a red target balloon appears alongside a growing number of blue distractors. The measurement is the interval between the stimulus being painted on your screen and your pointer or key landing on the target.

• Structure. One unscored practice round, then five scored rounds with 1, 1, 2, 3 and 4 distractors respectively.
• Timing. We start the clock after two animation frames have elapsed following the render, which is the closest a browser lets us get to “the moment you could see it”, and we read the pointer-down event rather than the click.
• Headline number. The median of your valid rounds, not the mean — a single lapse should not define the result.
• Exclusions. Responses under 120 ms are treated as anticipation rather than perception; responses over 10 seconds are treated as a break in attention. Both are shown to you and excluded from the median.
• Search slope. We fit a least-squares line of response time against distractor count. The slope estimates what each extra distractor cost you.

The largest limitation is your equipment. Display refresh rate, touchscreen sampling rate, browser, and whether you used a finger, trackpad or mouse can move this number by tens of milliseconds without any change in you. A single 30-second run is one noisy sample of a state, not a measurement of a trait.

A percentile is a claim about a population. Publishing one before you have a population is fabrication, and it is the single most common dishonesty on sites in this category.

MindScore therefore withholds percentiles until a comparison group contains at least 500 valid sessions. Comparison groups are defined by device category first — phones, tablets and desktops are never pooled — and by age bracket only where the participant volunteered one. Until a group reaches that threshold, your result page says so and shows you the raw numbers instead.

Only complete, fully valid runs enter the reference distribution. Incomplete runs, runs with implausible timings and repeat runs are stored but excluded, because a reference distribution polluted by practice effects is worse than none.

Performance across five reasoning formats: pattern recognition, logical reasoning, numerical reasoning, verbal reasoning and spatial reasoning. It describes how you handled the 18 items you were shown on this occasion, not a general or fixed ability.

• Length. 18 questions, about 5 minutes, drawn from a pool of 36 so no two attempts are the same test. Answers save as you go.
• Where the items come from. All 36 items in the pool were written for MindScore, and each attempt draws 18 of them. None are drawn from a published or proprietary intelligence battery, and several are deliberately built so that the first plausible reading is wrong.
• How it is scored. Each attempt draws four pattern, four logical, four numerical, three verbal and three spatial items from the pool, in a random order with options in a random order. Each item scores 0 or 1. A domain score is credit earned over credit available within that domain, expressed on a 0-100 scale. The overall score is the weighted mean of the five domain scores, and fixed thresholds map it to one of four descriptive bands.
• Versioning. The item bank (v1) and the scoring rules (v1) both carry version numbers. Changing an item or a key requires a new version, so a stored result can always be reproduced against the rules it was scored under.
• Server-side scoring. Answer keys, item weights and reverse-scoring flags never reach your browser. Your answers are stored as you give them and scored on our servers.
• Speed versus accuracy. Median time per item is described alongside your result so you can see the trade-off you made. It is descriptive only, and it does not affect your score.

Validation status. Not independently validated. Item difficulty and discrimination statistics have not yet been collected, and no norming sample exists.

What it cannot tell you.

• 18 items over roughly five minutes is a short sample; sleep, distraction and familiarity with the item formats all move the result.
• This is not an IQ test and produces no IQ-like number.
• It is not designed or validated for employment selection, academic placement or any other consequential decision.
• Bands describe performance on these items, not standing in a population.

Accuracy across five formats that general-ability tests commonly use: completing visual matrices, extending numeric and letter series, verbal relationships and meaning, quantitative reasoning, and deductive logic. It describes how you handled the items you were shown on this occasion. It does not measure, and does not claim to measure, intelligence as a fixed trait.

• Length. 25 questions, about 12 minutes, drawn from a pool of 50 so no two attempts are the same test. Answers save as you go.
• Where the items come from. All 50 items in the pool were written for MindScore. The matrices are generated from stated rules, so each key is provably the rule’s own output. None of the items is taken from, adapted from or modelled on a published or proprietary intelligence battery.
• How it is scored. Each attempt draws 5 items from each of the five domain pools, in a random order with the options in a random order. Each item scores 0 or 1. A domain score is items correct over items shown in that domain, on a 0–100 scale, and the overall score is the mean of the five domain scores. Fixed thresholds map it to a descriptive band. A percentile is shown only once the comparison group for your device category holds the configured minimum number of valid results.
• Versioning. The item bank (v1) and the scoring rules (v1) both carry version numbers. Changing an item or a key requires a new version, so a stored result can always be reproduced against the rules it was scored under.
• Server-side scoring. Answer keys, item weights and reverse-scoring flags never reach your browser. Your answers are stored as you give them and scored on our servers.
• Speed versus accuracy. Median time per item is described alongside your result so you can see the trade-off you made. It is descriptive only, and it does not affect your score.

Validation status. Not independently validated. Item difficulty and discrimination statistics have not yet been collected, no norming sample exists, and no percentile is shown until the minimum sample gate is met.

What it cannot tell you.

• This is not an IQ test. An IQ score is a position relative to a large, representative, normed sample, and MindScore does not have one. The result describes accuracy on these items, not a standing in the population.
• 25 items in about twelve minutes is a short sample. Sleep, distraction, the device you used, and familiarity with these item formats all move the result.
• It is not designed or validated for employment selection, academic placement, clinical assessment or any other consequential decision, and it must not be used for one.
• Items differ between attempts by design, so two attempts are not directly comparable item for item, only by domain and overall band.

Six facets of everyday emotional skill: emotional recognition, self-awareness, emotional regulation, empathy, social judgement and conflict response. It measures the choices you make in written scenarios, which is not the same as what you do under real emotional load.

• Length. 36 questions, about 8 minutes. Answers save as you go.
• Where the items come from. All 36 scenarios and their response options were written for MindScore. No items are taken from a published or proprietary emotional-intelligence inventory. Options are graded by partial credit rather than keyed right or wrong, and were written so that no option is an obviously poor choice.
• How it is scored. Each option carries a credit between 0 and 1. A domain score is credit earned over credit available within that domain, expressed on a 0-100 scale. The overall score is the unweighted mean of the six domains, and fixed thresholds map it to one of four descriptive bands.
• Versioning. The item bank (v1) and the scoring rules (v1) both carry version numbers. Changing an item or a key requires a new version, so a stored result can always be reproduced against the rules it was scored under.
• Server-side scoring. Answer keys, item weights and reverse-scoring flags never reach your browser. Your answers are stored as you give them and scored on our servers.
• Speed versus accuracy. Median time per item is described alongside your result so you can see the trade-off you made. It is descriptive only, and it does not affect your score.

Validation status. Not independently validated. No norming sample exists, no item-level statistics have been collected, and no convergent validity against established measures has been established. Scores describe performance on these items only.

What it cannot tell you.

• Choosing a response in a written scenario is easier than producing it in the moment, so this measures judgement rather than behaviour.
• The credits reflect MindScore’s reasoning about what each response attends to. Reasonable people disagree about individual items, and cultural norms about directness and emotional expression vary widely.
• This is not a clinical instrument. It does not screen for alexithymia, autism, personality disorders or any other condition, and a low score is not evidence of any of them.
• It is not designed or validated for hiring, promotion or team selection.

Six interest dimensions framed on the RIASEC model, eight working-style dimensions, four work values, a stated priority ordering and an environment preference. It describes preferences you reported, not aptitude, and not what you would be good at.

• Length. 70 questions, about 12 minutes. Answers save as you go.
• Where the items come from. All 71 items were written for MindScore. The interest section is organised on Holland’s RIASEC framework, which is a published theoretical model rather than a proprietary instrument; no items are taken from any published RIASEC inventory. Career families and example roles are written by MindScore in plain language.
• How it is scored. Each Likert item contributes a normalised 0-1 value to its dimension, with reverse-scored items mirrored. Domain scores are the weighted mean of their items on a 0-100 scale. Career families are matched by comparing your dimension profile against each family’s published requirement profile, with a documented weighting between interest fit, working-style fit and values fit. The match score is a fit percentage, explicitly not a percentile.
• Versioning. The item bank (v1) and the scoring rules (v1) both carry version numbers. Changing an item or a key requires a new version, so a stored result can always be reproduced against the rules it was scored under.
• Server-side scoring. Answer keys, item weights and reverse-scoring flags never reach your browser. Your answers are stored as you give them and scored on our servers.
• Speed versus accuracy. Median time per item is described alongside your result so you can see the trade-off you made. It is descriptive only, and it does not affect your score.

Validation status. Not validated against career outcomes. No norming sample exists and no predictive validity has been established. The RIASEC framing is drawn from published theory; this implementation of it is untested.

What it cannot tell you.

• This measures stated preference, not ability. Enjoying investigative work is not evidence that you would be good at it, and the reverse is also true.
• Career fit depends heavily on the specific team, manager and organisation. A well-matched family can contain roles you would dislike.
• The matching rules are our own, written down and testable, but they have not been validated against real career outcomes.
• No salary, demand or growth figures are given anywhere, because we have no sourced, dated and geographically specific data to give.
• It is not a psychometric instrument and must not be used for hiring, selection or placement.

Attribution.

• Interest dimensions are organised on the RIASEC model described in the vocational-interest literature (Holland). The model is used as a public theoretical framework; no proprietary inventory items are reproduced.

Two broad dimensions from the adult-attachment literature — anxiety (how much attention goes to where you stand) and avoidance (how much of yourself you keep back) — plus secure-base behaviour and six narrower dimensions: comfort with closeness, need for reassurance, need for independence, conflict approach, emotional communication and response to uncertainty. It describes self-reported behaviour in close relationships, not a personality type and not a condition.

• Length. 30 questions, about 7 minutes. Answers save as you go.
• Where the items come from. All 30 items were written for MindScore. No items are taken from the ECR, ECR-R, AAI, RQ or any other published or proprietary attachment instrument. The two-dimensional anxiety–avoidance framing is a published theoretical model used here as a public framework; this implementation of it is entirely original and entirely untested.
• How it is scored. Each item contributes a normalised 0-1 value to one dimension, with reverse-scored items mirrored. A dimension score is the mean of its items on a 0-100 scale. Eight of the nine dimensions are bipolar: a high score is a position, not an achievement, and only secure-base behaviour contributes to the headline band. The four familiar style labels are read off the anxiety and avoidance scores as regions of a two-dimensional space, and the report says so explicitly rather than assigning you to a category.
• Versioning. The item bank (v1) and the scoring rules (v1) both carry version numbers. Changing an item or a key requires a new version, so a stored result can always be reproduced against the rules it was scored under.
• Server-side scoring. Answer keys, item weights and reverse-scoring flags never reach your browser. Your answers are stored as you give them and scored on our servers.
• Speed versus accuracy. Median time per item is described alongside your result so you can see the trade-off you made. It is descriptive only, and it does not affect your score.

Validation status. Not validated. No norming sample exists, no reliability or factor-structure statistics have been collected, and no relationship between these scores and any real-world outcome has been established. The dimensional framing is drawn from published theory; these items and this scoring are ours and are untested.

What it cannot tell you.

• This is a self-report questionnaire about how you believe you behave. It cannot see what you actually do, and self-report on this subject is known to be flattering in some directions and harsh in others.
• Attachment positions are not fixed traits. They move with the relationship you are in, with what is happening in your life, and over years. A result is a snapshot of how you answered today.
• It describes one person. Relationships are two people, and a pattern that causes friction with one partner can be unremarkable with another.
• The four familiar style names are convenient labels for regions of a continuous space. Very few people sit squarely in one, and treating them as categories is the single most common misuse of this framework.
• It is not a clinical or diagnostic instrument, it does not screen for any condition, and it is not therapy or a substitute for it.
• It cannot predict whether a relationship will last, and nothing in the report attempts to.

Attribution.

• The anxiety–avoidance dimensional framing is drawn from the published adult-attachment literature, where it is a public theoretical model. No items from any published attachment inventory are reproduced or adapted.

Six areas of a close relationship as each participant describes themselves: communication directness, conflict and repair behaviour, trust and reliability, ambition and the place of work, lifestyle and pace, and long-term priorities. It describes two self-reports, not a relationship.

• Length. 36 questions, about 9 minutes. Answers save as you go.
• Where the items come from. All 36 items were written for MindScore. Nothing is taken from a published or proprietary relationship inventory, and the comparison rules are our own.
• How it is scored. Each participant answers independently and is scored exactly like any other MindScore questionnaire: normalised item values, reverse items mirrored, a 0-100 score per area. The two sets of area scores are then compared using a different rule per area, chosen in advance and printed in the report — a floor where both people need to clear a bar, a gap rule where the distance between two people is the cost, a tolerant similarity rule where a difference is negotiable, and a strict alignment rule where a large gap genuinely matters. It is deliberately not an average similarity score, because the areas do not work the same way.
• Versioning. The item bank (v1) and the scoring rules (v1) both carry version numbers. Changing an item or a key requires a new version, so a stored result can always be reproduced against the rules it was scored under.
• Server-side scoring. Answer keys, item weights and reverse-scoring flags never reach your browser. Your answers are stored as you give them and scored on our servers.
• Speed versus accuracy. Median time per item is described alongside your result so you can see the trade-off you made. It is descriptive only, and it does not affect your score.

Validation status. Not validated. No norming sample exists, no reliability statistics have been collected, and no relationship between these scores and any real-world outcome has been established or tested.

What it cannot tell you.

• This compares two self-reports. It has no access to how either of you actually behaves, and self-description in relationships is unreliable in both directions.
• A high score is not a prediction that a relationship will work, and a low one is not a prediction that it will not. Nothing here predicts anything.
• The comparison rules and their weights are our own judgement. They are written down and testable, and they have not been validated against any relationship outcome.
• Six areas is not a relationship. Attraction, history, circumstance, timing, children, health and money all matter enormously and none of them are measured here.
• It is not a clinical or diagnostic instrument, it is not couples therapy, and it is not a substitute for either.
• It cannot detect coercion, dishonesty or harm, and it must not be used to decide whether someone is safe to be with.

Reports are deterministic. Score bands select passages from version-controlled content, so the same answers always produce the same report, and we can tell you exactly why a sentence appeared.

The architecture leaves room for an AI-written narrative layer in future. If that is ever added it will present the same numbers in different words and will be labelled as such. An AI layer will never be permitted to change a score, because the score is the part you are trusting us with.

• We do not claim clinical or diagnostic validity, because we have not established it.
• We do not claim predictive validity for job performance, study outcomes or life outcomes.
• We do not publish rarity claims such as “top 1%”.
• We do not publish testimonials, user counts or average scores, because we would have to invent them.
• We do not describe any MindScore assessment as scientifically validated. “Research-informed” means the task formats draw on published paradigms; it does not mean this implementation has been independently validated.

If you believe an item is wrong, ambiguous or badly keyed, tell us at hello@getmindscore.com. Item quality is a standing work item and corrected items ship as a new version rather than a silent edit.