Raw measures (metrics)
Shared counters: exercise completed, score 0-100, accuracy, time used, delay before the first action, hints, errors, corrections, attempts, longest streak, skipped items, score version.
Method
How NeuroTrack computes its indicators.
Formulas, filtering rules and limits of every measure, copied from the application’s reference: score version 7, indicator version 5.
What is recorded
Two objects per exercise, and filtering rules.
At the end of an exercise, each exercise returns counters shared by all exercises and the indicators of its own.
Raw measures (
Score detail (scoreDetails)
The exercise’s own indicators, the score breakdown (accuracy, time, bonus) and the indicator version.
An indicator that cannot be computed is null, never 0: a 0 would read as a measurement. Typical cases: zero denominator, condition absent from the exercise, fewer than 3 trials in a reaction time subset.
An exercise abandoned or stopped by the timer keeps its partial metrics, flagged as not completed, with a placeholder score of 0. Reports and series leave it out: these values say nothing about performance.
Hesitation delays (mean, median, count) come from the telemetry of Anagram, Memory Card, Sentence Reconstruction, Memory List, Simon Says, Stroop Test, Syllable Game, Fill in the Blanks, Mental Math, Find the Intruder, Word Completion, Word Search, Rule Flip, Letter in a Letter, Multi-Criteria Sort, Sense Switch, Spatial Reflex, Symbol Decoder, and Multi-Step Command only.
Reaction time (RT) rule
These exercises apply the same rule, in two steps: filter the trials, then compute.
Filtering trials
- Trials kept for the mean reaction time and the means per condition
- Correct trials only
- Field in the data
avgReactionTimeMs, avgRT*Ms - Validity window
- 100 ms to 5000 ms. Below: anticipation. Above: outlier
- Timed-out trials
- Excluded (pressure bar exceeded at Expert, for the exercises that have one)
- Field in the data
timedOut - Counters
- Valid trials, excluded as anticipations, excluded as outliers, excluded as timed out. No value for exercises without a time limit: Stroop Test and N‑Back Memory
- Field in the data
rtValidTrialCount, rtExcludedAnticipationCount, rtExcludedOutlierCount, rtExcludedTimeoutCount
Computing the means
- Mean
- Rounded integer. No value without a valid trial
- Field in the data
avgReactionTimeMs - Standard deviation
- Population standard deviation, computed around the rounded mean. No value under 2 trials
- Field in the data
stdDevReactionTimeMs - Comparison subset (repeat, rule change, congruent, incongruent, visual, auditory, priority rule, normal rule, rule update, steady state)
- No mean under 3 valid trials. The count of each subset is stored
- Field in the data
rt*TrialCount, MIN_POOL_TRIALS = 3 - Mean time of all trials
- Same window and same exclusion of timed-out trials, incorrect trials included. Used by the impulsivity index
- Field in the data
avgRTAllMs
The denominator of a mean is the number of valid trials after filtering. Two patients who play the same number of trials do not have the same count: check the number of valid trials before comparing two means. N‑Back Memory computes its mean time on correct responses to targets only.
Comparing over time
Two markers track two different things: scoreVersion (7) for the score formula, metricsParadigmVersion (5) for the definition of the indicators.
- Trend per exercise: a row without a marker, or from a version before the switch, does not count for the indicators and exercises concerned.
- Score curve: a session enters it only if all its completed exercises carry score version 7. After a formula update, a patient’s curve can empty until sessions in the current format arrive.
- Score breakdown: the maxima follow the formula version of each exercise.
Indicators compare over time version by version, and the application leaves out what does not compare. Values from before version 7 remain readable in the session report.
The score 0-100
Three components, then a hint penalty.
Each exercise gets a score out of 100. Its profile sets the weight of accuracy, time and bonus.
| Profile | Accuracy | Time | Bonus | Exercises |
|---|---|---|---|---|
rt-pure | 20 | 70 | 10 | Stroop Test, Rule Flip, Letter in a Letter, Multi-Criteria Sort, Sense Switch, Spatial Reflex, Symbol Decoder, N‑Back Memory, Go/No‑Go |
memory | 50 | 30 | 20 | Memory Card, Simon Says |
sequential | 60 | 25 | 15 | Anagram, Sentence Reconstruction, Memory List, Syllable Game, Fill in the Blanks, Mental Math, Find the Intruder, Word Completion, Word Search, Verbal Fluency, Emotion Recognition, Trail Making, Dual Task, Temporal Orientation, Clock, List Recall, Free Fluency, Picture Naming, Spatial Orientation, Digit Span, Multi-Step Command, Figure Copy |
Five exercises are scored on one axis and followed on another: Memory List, N‑Back Memory, List Recall, Picture Naming, and Digit Span. The profile sets the score, the tracking family sets the curves offered.
Formula (version 7)
accuracy = round(wa × completed / total)
time = round(wt × tr)
bonus = clamp(round(wb × bonusRatio), 0, wb)
penalty = min(1, hintsUsed / 3) × 0.25
score = clamp(round((accuracy + time + bonus) × (1 − penalty)), 0, 100)- tr = 1 − timeUsed / timeLimit (floor 0) when timeLimit > 0. When timeLimit = 0: 1 in sequential and memory, reactionTimeRatio in rt-pure.
- In rt-pure, tr is multiplied by min(1, (completed / total) / 0.6): answering fast and at random does not earn the time weight. N‑Back Memory and Go/No‑Go count correct rejections in accuracy: this floor has little effect there. Not validated, decision pending.
- reactionTimeRatio = clamp((2000 − avgRT) / 1500, 0, 1), 0.5 when avgRT is null.
- bonusRatio: the streak is min(1, maxStreak / total) from 3 upwards, otherwise 0. Without an exercise-specific bonus, bonusRatio is the streak. With an exercise-specific bonus g between 0 and 1, bonusRatio = 0.5 × streak + 0.5 × g.
- The difficulty bonus does not enter the score. It remains shown for information.
- The hint penalty reaches 25% at 3 hints, linear below. hintsUsed is the one in metrics.
Exercise-specific bonus
| Exercise | Bonus passed to the score |
|---|---|
| Reaction time exercises (rt-pure profile) | reactionTimeRatio |
| N‑Back Memory | clamp((d′ + 2) / 6, 0, 1), 0.5 without d′ |
| Simon Says | perfectRounds / target × 0.5 + overshoot bonus × 0.5 |
| Temporal Orientation | orientationScore / 100 |
| Figure Copy | angleAccuracy / 100, 0 without a drawing |
| All the others | none: the streak carries the whole bonus weight |
Exercises at an imposed pace (N‑Back Memory and Go/No‑Go) pass timeLimit = 0: the stream’s duration does not depend on the patient, so the speed measured on hits carries the time component. Dual Task does the same, and time is full credit for everyone there: its speed reads in the dual-task costs.
The indicators, one by one
Formula, unit, exercises, reading.
Each card copies the reference: same order of operands, same units, same cases without a value. The reading compares the patient with themselves, never with a threshold.
Shared measures
Accuracy
- Formula
accuracyPercent: shared counter of metrics, in % Exercises with attempts (Anagram, Memory Card, Syllable Game, Fill in the Blanks, Word Completion, and Word Search): successes / (attempts + skips) Memory Card: pairs found / moves × 100- Unit
- %
Accuracy
- Example, fictitious values
- 12 successes, 14 attempts and 2 skipped items: 12 / (14 + 2) = 75%.
- Clinical reading
- Compared with the same patient’s previous sessions, on the same exercise. Since paradigm 5, skipping items lowers accuracy.
- For research
- In exercises with attempts, an item is retried until it succeeds: accuracy is computed on attempts. Skipped items enter the denominator since paradigm 5 (D4). An exercise abandoned or stopped by the timer keeps its partial metrics with completed: false; reports and series leave it out.
Hints
- Formula
hintsUsed: number of hints used during the exercise avgHintsPerWord (Word Search): hints per word penalty = min(1, hintsUsed / 3) × 0.25- Unit
- count
- Exercises
Hints
- Example, fictitious values
- 2 hints: penalty = min(1, 2 / 3) × 0.25 ≈ 0.167. A total of 80 points becomes round(80 × 0.833) = 67.
- Clinical reading
- Help received also shows in autonomy: an exercise passed with a hint counts as “helped”, not “on their own”.
- For research
- The score’s hint penalty reaches 25% at 3 hints, linear below. It uses the hintsUsed of metrics. For List Recall and Picture Naming, hintsUsed is 0 by design: help shows in their own cues.
Longest streak
- Formula
maxStreak: longest run of consecutive correct answers streak = min(1, maxStreak / total) from 3 upwards, otherwise 0- Unit
- consecutive items
Longest streak
- Example, fictitious values
- A streak of 6 out of 12 items: min(1, 6 / 12) = 0.5.
- Clinical reading
- An unbroken run of good answers, followed session by session on the same exercise.
- For research
- Units are not normalised across exercises: consecutive pairs (Memory Card), successful lists capped at 3 (Memory List), faultless rounds (Simon Says), actively given answers (Verbal Fluency), consecutive correct answers elsewhere. In Anagram and Find the Intruder, a skipped item neither breaks nor extends the streak: only a wrong answer resets it.
Corrections
- Formula
correctionCount: shared counter of metrics; in Trail Making, presses on Erase- Unit
- count
- Exercises
Corrections
- Example, fictitious values
- The patient presses Erase 3 times during a trail: correctionCount = 3.
- Clinical reading
- Shows where the patient goes back on an answer.
- For research
- The gesture counted depends on the exercise: the counter does not compare across exercises.
Skipped items
- Formula
skipCount: items skipped with the “Skip” button- Unit
- count
Skipped items
- Example, fictitious values
- The patient skips 2 items out of 10: skipCount = 2, and these 2 items enter the accuracy denominator.
- Clinical reading
- A skipped item enters the accuracy denominator.
- For research
- Simon Says, Syllable Game, Fill in the Blanks, and Word Search emit skipCount since fix C21, with no version change. In assessments, the “Skip” button is hidden.
Speed and variability
Reaction time
- Formula
avgReactionTimeMs: mean of the valid RTs of correct trials, ms- Unit
- ms
Reaction time
- Example, fictitious values
- Correct trials at 640, 720, 80 and 5300 ms: 80 ms (anticipation) and 5300 ms (outlier) are excluded; mean = (640 + 720) / 2 = 680 ms.
- Clinical reading
- Processing speed. A rise followed over several sessions matters more than a single value.
- For research
- Validity window from 100 ms to 5000 ms, timedOut trials excluded, rounded integer, null without a valid trial. The denominator is the number of valid trials after filtering: check rtValidTrialCount before comparing two means. N‑Back Memory averages hits only.
Reaction time variability
- Formula
stdDevReactionTimeMs: population standard deviation of valid RTs, ms attentionalVariability = stdDevReactionTimeMs / avgReactionTimeMs × 100- Unit
- ms; % for the coefficient of variation
Reaction time variability
- Example, fictitious values
- Valid RTs of 600, 700 and 800 ms: mean 700 ms, standard deviation ≈ 82 ms, coefficient of variation = 82 / 700 × 100 ≈ 11.7%.
- Clinical reading
- Spread of RTs. Read it with the mean.
- For research
- The standard deviation is computed around the rounded mean, null under 2 trials. The coefficient of variation is computed on the fly, null if an operand is missing. No threshold is shown: it is a raw ratio.
Time per item
- Formula
avgTimePerItemMs: useful time divided by the number of items passed, ms- Unit
- ms
Time per item
- Example, fictitious values
- 180 s of useful time for 12 items passed: 180,000 / 12 = 15,000 ms.
- Clinical reading
- Followed session by session on the same exercise.
- For research
- The denominator depends on the exercise: items passed (Anagram, Sentence Reconstruction, Syllable Game, Fill in the Blanks, Find the Intruder, Word Completion, and Word Search, and Mental Math since paradigm 5), all trails (Trail Making, with the 15 s reveal per error removed from the numerator), encoded words (List Recall, encoding phase removed), first production before a cue (Picture Naming). Two exercises do not compare on this key.
Duration per item shown
- Formula
avgStepDurationMs: total time divided by all items, ms- Unit
- ms
Duration per item shown
- Example, fictitious values
- 200 s of total time for 16 items shown, passed or not: 200,000 / 16 = 12,500 ms.
- Clinical reading
- Includes abandoned items.
- For research
- Not to be confused with time per item, whose denominator counts passed items only.
Items per minute
- Formula
itemsPerMinute: correct answers actively given per minute- Unit
- answers per minute
Items per minute
- Example, fictitious values
- 18 correct answers given in 90 s: 18 / 1.5 = 12 answers per minute.
- Clinical reading
- Output rate, followed session by session on the same exercise.
- For research
- Each exercise has its own definition. Verbal Fluency is a classification task at an imposed pace: its rate credits actively given answers only and does not compare with the fluency literature.
Inhibition
Congruency effect
- Formula
congruencyEffectMs = avgRTIncongruentMs − avgRTCongruentMs, ms inhibitionIndex = congruencyEffectMs / avgRTCongruentMs × 100 (Stroop Test)- Unit
- ms; % for the inhibition index
- Exercises
Congruency effect
- Example, fictitious values
- Incongruent RT 1036 ms, congruent RT 842 ms: 1036 − 842 = 194 ms; inhibition index = 194 / 842 × 100 ≈ 23%.
- Clinical reading
- The extra time when two pieces of information clash, for this patient, from one session to the next. Never read the effect without its construct.
- For research
- null if either mean is null; each mean needs 3 valid trials. The construct differs by exercise (congruencyConstruct): semantic for Stroop Test, hierarchical for Letter in a Letter. The inhibition index has no meaning for Navon’s hierarchical construct and is left out of the trend catalogue of Letter in a Letter. Accuracy per condition: congruentAccuracyPercent, incongruentAccuracyPercent. References: MacLeod (1991), Navon (1977).
Rule compatibility effect
- Formula
simonEffectMs = avgRTIncongruentMs − avgRTCongruentMs, ms- Unit
- ms
- Exercises
Rule compatibility effect
- Example, fictitious values
- Opposite-side trials 910 ms, same-side trials 780 ms: 910 − 780 = 130 ms.
- Clinical reading
- Sets the same-side rule against the opposite-side rule, for this patient, from one session to the next.
- For research
- It is a rule compatibility effect, not the classic Simon effect (Simon, 1969): the key name is historical. Means are null under 3 valid trials. Accuracy per condition is emitted too.
False alarms
- Formula
falseAlarmRate: share of commissions among trials that required inhibition, % falsePositiveCount: distractors ticked (Memory List), responses to a distractor (Go/No‑Go)- Unit
- %; count
False alarms
- Example, fictitious values
- 3 responses on 20 trials that required holding back: 3 / 20 = 15%.
- Clinical reading
- For Memory List, N‑Back Memory, and List Recall: recognition distractors or non-targets, not the inhibition of a gesture.
- For research
- The session report’s attentional profile leaves out Memory List, N‑Back Memory, and List Recall for this reason. On Spatial Reflex, falseAlarmRate equals impulsivityRate: only the latter is shown.
Impulsivity
- Formula
impulsivityIndex = (avgReactionTimeMs − avgRTAllMs) / avgReactionTimeMs × 100 impulsivityRate (Spatial Reflex): share of errors on incongruent trials, %- Unit
- %
Impulsivity
- Example, fictitious values
- avgReactionTimeMs = 700 ms, avgRTAllMs = 672 ms: (700 − 672) / 700 × 100 = 4%.
- Clinical reading
- Positive when errors are faster than correct answers.
- For research
- avgRTAllMs: mean RT of all trials (correct or not) within the 100-5000 ms window. The index is computed on the fly, null if an operand is missing, with no threshold shown. Since paradigm 4, avgRTAllMs is filtered on Rule Flip, Multi-Criteria Sort, and Sense Switch: the index changed meaning there.
Omissions
- Formula
omissionCount: targets left without a response- Unit
- count
- Exercises
Omissions
- Example, fictitious values
- Out of 40 targets, 3 get no response: omissionCount = 3.
- Clinical reading
- Targets left unanswered during the stream.
- For research
- omissionCount and falsePositiveCount point in opposite directions (inattention, impulsivity): do not add them up.
Flexibility
Switch cost
- Formula
switchCostMs = avgRTSwitchMs − avgRTRepeatMs, ms switchCostRatio = switchCostMs / avgRTRepeatMs × 100, %- Unit
- ms; %
Switch cost
- Example, fictitious values
- RT with a rule change 1150 ms, without 900 ms: 1150 − 900 = 250 ms; 250 / 900 × 100 ≈ 27.8%.
- Clinical reading
- A negative cost is a real measurement: the app keeps it.
- For research
- avgRTSwitchMs, avgRTRepeatMs: mean RT of trials with and without a rule change, null under 3 valid trials. The first trial is excluded, as it has no previous trial. For Symbol Decoder, switchCostMs is the difference between ruleUpdateLatencyMs and avgRTSteadyState: it does not compare with that of Rule Flip, Letter in a Letter, and Multi-Criteria Sort. For Sense Switch, the cost shows in switchCostRatio and modalShiftCostMs.
Perseveration errors
- Formula
perseverationErrorCount: trials where the answer matches the old rule- Unit
- count
- Exercises
Perseveration errors
- Example, fictitious values
- After a rule change, 2 answers still follow the old rule: perseverationErrorCount = 2.
- Clinical reading
- Counts the answers that still follow the previous rule.
- For research
- null for Rule Flip and Multi-Criteria Sort (disjoint response sets). Two separate definitions: never average Letter in a Letter and Symbol Decoder, which the progress view plots as two series. On Symbol Decoder, at difficulties 1 and 2, two categories make every error after a rule change a perseveration; at Expert, a third category makes the measure selective, and the two values do not compare. Not to be confused with the repetitions of List Recall and Free Fluency.
Rule update latency
- Formula
ruleUpdateLatencyMs: mean RT over the 2-trial window after a rule change avgRTSteadyState: mean RT in steady state- Unit
- ms
- Exercises
Rule update latency
- Example, fictitious values
- Mean RT of the 2 trials after a change 1250 ms, steady state 950 ms: the switch cost is 1250 − 950 = 300 ms.
- Clinical reading
- The time it takes to apply a new rule.
- For research
- Means are null under 3 valid trials. Their difference gives the switchCostMs of Symbol Decoder.
Cross-modal switch cost
- Formula
modalShiftCostMs = avgRTSwitchMs − avgRTRepeatMs between modalities- Unit
- ms
- Exercises
Cross-modal switch cost
- Example, fictitious values
- RT after a change of modality 980 ms, without change 860 ms: 980 − 860 = 120 ms.
- Clinical reading
- What moving from one sense to the other costs.
- For research
- Means are null under 3 valid trials. At Expert, the pressure bar of Sense Switch is 2000 ms, against 3000 or 4000 ms elsewhere: its exclusion rate does not compare with other exercises.
Sensory dominance
- Formula
sensoryDominanceRatio = avgRTVisualMs / avgRTAuditoryMs, 2 decimals- Unit
- unitless ratio
- Exercises
Sensory dominance
- Example, fictitious values
- Visual RT 720 ms, auditory RT 800 ms: 720 / 800 = 0.90.
- Clinical reading
- Above 1, vision is slower than hearing.
- For research
- null if either mean is. avgRTVisualMs, avgRTAuditoryMs: RT per modality.
Priority rule time
- Formula
overrideRuleRTMs, normalRuleRTMs: RT on priority-rule trials and on normal trials, trials without a criterion change only- Unit
- ms
- Exercises
Priority rule time
- Example, fictitious values
- Priority-rule trials 1240 ms, normal trials 1010 ms: the two values read side by side.
- Clinical reading
- Compares the time of trials where a priority rule applies with that of normal trials.
- For research
- overrideRuleRTMs is null at difficulty 1 (no trial). Both means cover trials without a switch since paradigm 5 (D7).
Memory
Attempts per item
- Formula
avgAttemptsPerItem: attempts per item; for Memory List, selections per word shown- Unit
- attempts per item
Attempts per item
- Example, fictitious values
- 20 attempts for 16 items: 20 / 16 = 1.25 attempts per item.
- Clinical reading
- How many tries it took to pass an item.
- For research
- For Memory List, the value can drop below 1: incomplete recall. Left out of the trend catalogue for Sentence Reconstruction, Mental Math, and Find the Intruder, where it always stays at or below 1.
Level reached
- Formula
levelReached: rounds completed (0 to the target) perfectRounds: faultless rounds- Unit
- rounds
- Exercises
Level reached
- Example, fictitious values
- Target of 8 rounds, 5 completed including 3 faultless: levelReached = 5, perfectRounds = 3.
- Clinical reading
- Sequence length held, from one session to the next.
- For research
- Since paradigm 5, the round in progress is never counted (D3). levelReached is not a span.
Sensitivity d′
- Formula
dPrimeScore = z(hits) − z(false alarms), rates corrected (k + 0.5) / (n + 1) nLevel: configured N level- Unit
- unitless
- Exercises
Sensitivity d′
- Example, fictitious values
- 18 hits on 20 targets, 4 false alarms on 40 non-targets: corrected rates 18.5 / 21 ≈ 0.881 and 4.5 / 41 ≈ 0.110; d′ ≈ 1.18 − (−1.23) = 2.41.
- Clinical reading
- Tells a target from a distractor, regardless of the tendency to answer “yes” or “no”.
- For research
- null under 2 targets and under 2 non-targets. nLevel is an exercise setting, not a performance. Rate correction: Hautus (1995), Snodgrass and Corwin (1988).
Cue sensitivity
- Formula
cueingSensitivityIndex = (cued − free) / (encoded − free) freeRecallCount, cuedRecallCount, recognitionCount, encodedCount: words recalled freely, after a cue, recognised, encoded primacyEffect, recencyEffect: share of recalled words from the start, from the end of the list- Unit
- ratio; number of words
- Exercises
Cue sensitivity
- Example, fictitious values
- 16 words encoded, 8 recalled freely, 13 after a cue: (13 − 8) / (16 − 8) = 0.625.
- Clinical reading
- The share of recall that the category cue retrieves, from one session to the next.
- For research
- encodedCount is the denominator of every count. The index is null when free recall is complete; it is a badge, never a curve. primacyEffect and recencyEffect are null under 4 words or without free recall. Paradigm: Grober and Buschke (1987).
Repetitions
- Formula
perseverations: repetitions of an answer already given- Unit
- count
- Exercises
Repetitions
- Example, fictitious values
- The patient writes “apple” twice: perseverations = 1.
- Clinical reading
- A word given again within the same task.
- For research
- A repetition requires strict equality of a distinct word (D8, paradigm 5). The key is called perseverations, but it is not the rule perseveration of Letter in a Letter and Symbol Decoder: the two do not compare.
Intrusions
- Formula
intrusions: answers outside the material- Unit
- count
Intrusions
- Example, fictitious values
- The patient gives a word that was not in the learned list: intrusions = 1.
- Clinical reading
- Answers that do not belong to the material shown.
- For research
- Emitted by Verbal Fluency, List Recall, and Free Fluency, each on its own material.
Forward and backward span
- Formula
forwardSpan, backwardSpan: longest step passed, forward and backward order spanDelta: their gap- Unit
- digits
- Exercises
Forward and backward span
- Example, fictitious values
- Longest step passed: 6 digits forward, 4 backward. forwardSpan = 6, backwardSpan = 4, spanDelta = 2.
- Clinical reading
- The gap between both spans separates holding information from manipulating it.
- For research
- null under the starting step, never 0. The gap is a badge, never a curve. Digit Span is not classified for autonomy: its accuracy covers administered trials that the adaptive procedure varies.
Language
Clusters and switches
- Formula
clusterSize: Troyer cluster size (n−1 per cluster) clusteringSwitchCount: number of switches, article removed fluencySlices: output per 15 s slice (4 for 60 s)- Unit
- words; switches
- Exercises
Clusters and switches
- Example, fictitious values
- Three clusters of 3, 2 and 1 words: Troyer sizes 2, 1 and 0; 2 switches between clusters.
- Clinical reading
- Shows how the patient organises the search: clustered words and switches.
- For research
- null in semantic mode. Troyer approximation in phonemic mode. The slices add up to the total output. Verbal Fluency emits no switches: no sequence produced by the patient exists. Reference: Troyer, Moscovitch and Winocur (1997).
Naming
- Formula
namingAccuracy: share of pictures named without a cue phonemicCueBenefit, semanticCueBenefit: share of cued items then found phonemicParaphasiaCount: answers close to the target in form, edit threshold of one third of the length (3 at most)- Unit
- %; count
- Exercises
Naming
- Example, fictitious values
- 20 pictures, 14 named without a cue: namingAccuracy = 70%. 4 get a sound cue and 3 are then found: phonemicCueBenefit = 3 / 4 = 75%.
- Clinical reading
- Read with accuracyPercent, which counts successes after a cue.
- For research
- Benefit denominator: cued items (phonemicCueCount, semanticCueCount), never the total. Under 6 letters, nothing stands apart from another word. Semantic paraphasias and circumlocutions are not detected: they belong to the clinician’s review. Time per item stops at the first production before a cue.
Orientation, construction, praxis
Temporal orientation score
- Formula
orientationScore = earnedWeight / totalWeight × 100. Weights: year 1, month 1, date 1, day 2, period 2, hour 3 temporalGranularityError: mean gap in days over wrong answers to “day of the month” consistencyScore: 100 or 0: does the weekday implied by the “day of the month” answer match the “day of the week” answer?- Unit
- weighted score out of 100, indicative; days
- Exercises
Temporal orientation score
- Example, fictitious values
- Year, month, day and period right, date and hour wrong: (1 + 1 + 2 + 2) / (1 + 1 + 1 + 2 + 2 + 3) × 100 = 60.
- Clinical reading
- Weighted score, indicative. The raw verdicts per question (orientationItems) come first.
- For research
- Weighting not clinically validated: orientationScore follows the code, not an MMSE or MoCA scoring. temporalGranularityError is null without a day error; consistencyScore is null at difficulty 1. The exercise asks 3, 6 or 12 questions depending on difficulty.
Spatial orientation
- Formula
spatialOrientationScore: score over the levels asked (room, floor, city, region, country) spatialGranularityError: widest wrong level spatialLevelsAsked: levels asked- Unit
- score; level
- Exercises
Spatial orientation
- Example, fictitious values
- Four levels asked (room, floor, city, region), region wrong: spatialGranularityError points to the region.
- Clinical reading
- Spots the widest wrong level, from the closest to the widest.
- For research
- A level left blank is not asked. No answer text is stored.
Clock errors
- Formula
clockAbsoluteErrorMin: mean of minute gap + hour gap, in clock minutes clockHandReversalCount: trials where swapping the hands would make the answer right clockStimulusBoundErrorCount: trials where the minute hand points at the number stated clockMinuteHandAccuracyPercent, clockHourHandAccuracyPercent: accuracy per hand, excluding trials where the cue showed the hand- Unit
- clock minutes; count; %
- Exercises
Clock errors
- Example, fictitious values
- For 3:00, hour hand on 12 and minute hand on 3: swapping the hands would make the answer right, so the trial counts in clockHandReversalCount.
- Clinical reading
- Grades failure where accuracy is binary.
- For research
- clockHandReversalCount is under-detected at times when the hands overlap: a 0 does not prove absence. clockStimulusBoundErrorCount is null when the exercise could not produce the error. Accuracy per hand is null if no trial enters the denominator (cue: step 2 for minutes, from step 1 for hours).
Trails A and B
- Formula
trailATimeMs, trailBTimeMs: mean time of the successful ascending and descending trails, ms trailBtoARatio: their ratio- Unit
- ms; ratio
- Exercises
Trails A and B
- Example, fictitious values
- Trails A passed in 40 s on average, trails B in 70 s: 70 / 40 = 1.75.
- Clinical reading
- Compares, for this patient, the time of both trail types from one session to the next.
- For research
- Contents A and B are drawn at random: the ratio is not a normed B/A TMT (Reitan, 1958). Since paradigm 5, time covers successful trails, without the reveal (D8).
Instruction steps
- Formula
stepsCompleted, stepsExpected: steps done, expected sequenceErrorCount: steps out of order (steps done minus longest increasing subsequence) sequenceAccuracyPercent: sequence accuracy commandIntrusionCount: actions that complete no step- Unit
- steps; %
- Exercises
Instruction steps
- Example, fictitious values
- A 4-step instruction, 3 steps done in the order 1, 3, 2: stepsCompleted = 3, longest increasing subsequence 2, sequenceErrorCount = 3 − 2 = 1.
- Clinical reading
- Tells a forgotten step from a step done in the wrong order.
- For research
- stepsCompleted and sequenceErrorCount are independent. The order error is a badge, never a curve.
Figure copy
- Formula
angleAccuracy: angular fidelity (0-100) closureError: worst closure gap (% of the diagonal) overlapCorrect: pentagon overlap rotationDeg: signed rotation (value of largest amplitude) cubeEdgeCount: cube edges- Unit
- 0-100; % of the diagonal; degrees; edges
- Exercises
Figure copy
- Example, fictitious values
- A cube copied with 11 of its 12 edges: cubeEdgeCount = 11; the stored drawing lets the clinician check the verdict.
- Clinical reading
- Indicative geometric scoring: the clinician sees the patient’s drawing and can check every verdict.
- For research
- Criteria calibrated on synthetic drawings, not on scored drawings: no equivalence with a clinical scoring. A mirror image shows as a rotation.
Social cognition and divided attention
Negative bias
- Formula
negativeBias: share of trials with a non-negative target where the patient answers a negative emotion (anger, fear, sadness), % mostConfusedPair: most frequent target→answer- Unit
- %
- Exercises
Negative bias
- Example, fictitious values
- On 20 trials with a non-negative target, the patient answers anger, fear or sadness 3 times: 3 / 20 = 15%.
- Clinical reading
- No alert threshold shown: the value reads from one session to the next.
- For research
- null under 4 trials with a non-negative target. mostConfusedPair is absent when no confusion dominates.
Dual-task cost
- Formula
dualTaskCostPrimary = (dual RT − single RT) / single RT × 100, % dualTaskCostSecondary = dual error rate − single error rate, percentage points secondaryAvgRTMs: mean RT on secondary targets in dual task- Unit
- %; percentage points; ms
- Exercises
Dual-task cost
- Example, fictitious values
- Single RT 600 ms, dual-task RT 750 ms: (750 − 600) / 600 × 100 = 25%. Error rate 5% alone, 12% in dual task: 7 points.
- Clinical reading
- What running two tasks at once costs.
- For research
- dualTaskCostSecondary is null without a target in a phase. Since paradigm 5, two rates are compared; prioritizationBias no longer exists (D5). A pause during the inter-stimulus interval resumes on a fresh stimulus: an isolated RT of up to about 5.4 s can result.
Composite indices
Inverse efficiency
- Formula
inverseEfficiencyScore = round(avgReactionTimeMs / (accuracyPercent / 100)), ms- Unit
- ms
Inverse efficiency
- Example, fictitious values
- Mean RT 700 ms, accuracy 80%: round(700 / 0.8) = 875 ms.
- Clinical reading
- Rises when the patient slows down and when they make mistakes.
- For research
- null at 0% accuracy. Computed on the fly, with no threshold shown: it is a raw ratio. Reference: Bruyer and Brysbaert (2011).
The session report
What the clinician reads after a session.
The “See the results” report sums up the session, then details each exercise.
- Average score: rounded mean of the scores of completed exercises. Abandoned exercises and those stopped by the timer are left out.
- Trend of an exercise: relative gap between the score and the patient’s past mean on this exercise, classed as up, stable or down by a fixed marker of the application. No trend under 3 runs.
- Score breakdown: accuracy, time and bonus, with the maxima of the profile and of the formula version; hint penalty recomputed from the number of hints.
- Planned versus felt: planned difficulty (1 to 3) against the patient’s rating, from 1 to 10 brought back to 1 to 3.
- Attentional profile: mean, over reaction time and flexibility exercises, of speed, vigilance slope, switch cost and false alarm rate, from 2 exercises. Memory List, N‑Back Memory, and List Recall are left out, as is the switch cost of Symbol Decoder. The false alarm alert follows a fixed marker of the application, indicative, with no normative basis.
Detail of an exercise: accuracy, time, then the exercise’s own measures. Each measure shows its definition when hovering the ⓘ icon.
Header
Exercise score, planned difficulty and the patient’s felt difficulty. See point 4
Score bar
Accuracy, time and bonus, each against its maximum. See point 3
Accuracy and time
Measures shared by every exercise, shown when the exercise recorded them.
Exercise measures
Specific to each exercise. Here, for the Stroop Test: congruency effect, accuracy by condition, false alarms…
The patient’s progress
One value per session, over time.
The “Progress” window plots the indicators session after session, over 3 months, 6 months, a year or the whole period.
- One value per completed session: the mean of the completed exercises concerned. Sessions without a value are left out.
- Score: rolling mean over 7 sessions. The curve keeps only sessions whose completed exercises all carry score version 7.
- Change over a period: mean of the last n values minus mean of the first n, n = min(10, ⌊N/2⌋), no value under 6 values.
- Summary indicators: sessions per week, average score, median reaction time per session, share of autonomy, mean actual duration, mean felt difficulty, most frequent planned level.
- Autonomy: each completed exercise counts as “on their own”, “helped” or “failed”, depending on its accuracy, compared with a fixed marker, and the help received. Help reads in hints, except for List Recall and Picture Naming, which have their own cues. Digit Span is not classified.
- Reaction time: Go/No‑Go, Spatial Reflex, Stroop Test, and N‑Back Memory. The time of switching or of the orchestra does not mean the same thing.
- Flexibility: Rule Flip, Letter in a Letter, and Multi-Criteria Sort. Modality: Sense Switch. Rule update: Symbol Decoder. Perseveration: Letter in a Letter and Symbol Decoder, as two series, never averaged.
- Impulsivity: rate of Spatial Reflex alone. The chart’s “watch” zone is a fixed marker of the application, indicative, with no normative basis.
Exercise by exercise
One curve per indicator the exercise emits.
The “By exercise” tab follows an indicator on one exercise or one type of exercise.
- One value per session, median or mean of the exercise’s runs. A curve appears only if the exercise emits the indicator.
- Training and assessment form two separate series. Assessments get no difficulty adjustment.
- The comparison by difficulty brings each value back to the patient’s baseline for the same difficulty. Under 4 sessions, the badge says “few points”.
- Values from an earlier indicator version are left out key by key.
- Within-session trend: the server computes the slope of reaction times over trial rank (ms per trial), after a warm-up, and the goodness of fit. Its thresholds remain to be confirmed: indicative, not validated.
- Statistics per domain: 56-day window (at least 3 sessions, 6 as a fallback), mean score per domain and its change. A domain becomes “to strengthen” when it falls away from the patient’s mean by a fixed marker, or when its trend goes down.
Limits and proper use
What the numbers do not say.
NeuroTrack supports cognitive rehabilitation under the responsibility of a health professional and does not diagnose.
- No norm: no z-score, no percentile, no population threshold. Each patient is their own reference.
- The application’s display markers (alerts, “watch” zones, autonomy, trend) are indicative, with no normative basis.
- Temporal orientation: weighting not clinically validated. The score follows the code, not an MMSE or MoCA scoring; the raw verdicts per question come first.
- Figure copy: indicative geometric scoring, calibrated on synthetic drawings. A mirror image shows as a rotation.
- Trail making: contents A and B drawn at random. The B / A ratio is not a normed TMT.
- Verbal Fluency: a classification task. Its rate does not compare with the fluency literature.
- Picture Naming: semantic paraphasias and circumlocutions are not detected. They belong to the clinician’s review.
- Clock: hand reversal is under-detected at times when the hands overlap.
- Score accuracy floor: not validated for N‑Back Memory and Go/No‑Go, which count correct rejections.
- Within-session trend: server thresholds to be confirmed, indicative.
- The screenshots on this page show fictitious data. This site promises no raw data export.
References
- MacLeod (1991), on the Stroop effect
- Simon (1969)
- Navon (1977)
- Hautus (1995); Snodgrass and Corwin (1988), for the correction of d′ rates
- Troyer, Moscovitch and Winocur (1997), for clusters and switches
- Grober and Buschke (1987), for cued recall
- Reitan (1958), for the Trail Making Test
- Bruyer and Brysbaert (2011), for the inverse efficiency score
Request a demo
Discover NeuroTrack with our team.
A video demonstration, tailored to your practice and your organisation.
- The documentation of the calculations, for a research project or a thesis
- Assessment protocols and how to read the indicators
- Your questions about HDS hosting, roles and deployment
Already a customer? Sign in with your Microsoft work account.
Mapping Minds, Tracking Progress