PRISM METHODOLOGY

How the read is built.

Prism uses cognitive tasks informed by the Cattell-Horn-Carroll framework, the IPIP-NEO public-domain personality inventory, and deterministic scoring rules. Every step is documented on this page. The full source model is available in the public repository.

Take Full Prism See Study Fit

Cognitive samples

Verbal, numerical, spatial, working memory, and processing speed are sampled directly through original cognitive items and paradigm tasks. Pattern reasoning (Gf) reads with wider uncertainty in this build. Fluid-reasoning items ship next.

Personality items

Sixty IPIP-NEO public-domain items resolve a full Big Five solution, twelve per factor, with BFAS aspect structure underneath.

Deterministic outputs

100%

Every score, archetype assignment, and Application Prompt is assembled from documented rules. Identical inputs always produce identical outputs.

Source model

Built on established cognitive science.

Prism is informed by the Cattell-Horn-Carroll framework, the modern consensus model in cognitive psychology, developed over six decades by Raymond Cattell, John Horn, and John Carroll. The framework organises cognitive ability into broad domains including fluid pattern reasoning, crystallised verbal reasoning, visual-spatial processing, working memory, quantitative reasoning, and processing speed. The personality layer builds on the IPIP-NEO public-domain inventory published by Lewis Goldberg, with aspect scoring drawn from the BFAS tradition developed by Colin DeYoung, Lena Quilty, and Jordan Peterson. Prism writes its item content in-house on these public-domain methodologies.

Cattell-Horn-Carroll cognitive framework

Prism organises cognitive measurement around the Cattell-Horn-Carroll framework, the modern consensus model in cognitive psychology, developed over six decades by Raymond Cattell, John Horn, and John Carroll. Six broad CHC abilities anchor the read: fluid pattern reasoning, crystallised verbal reasoning, visual-spatial processing, working memory, quantitative reasoning, and processing speed.

IPIP-NEO public-domain personality inventory

The personality layer uses 60 items from the International Personality Item Pool, the public-domain Big Five inventory published by Lewis Goldberg and extended by John Johnson. Sixty items resolve a real five-factor solution, not a three-trait proxy.

BFAS aspect structure

Aspect scoring follows the Big Five Aspect Scales tradition developed by Colin DeYoung, Lena Quilty, and Jordan Peterson. Sixty items map to ten aspects, then aggregate to the five broad traits. The aspect layer makes archetype assignment falsifiable rather than vibe-coded.

Original Prism items

Prism writes its own item content on public-domain methodology. Item formats follow established task conventions such as digit span, mental rotation, verbal inference, and letter comparison, while Prism constructs the items themselves in-house. No content is reproduced from commercial assessments.

Scoring

From inputs to a measured profile.

The six cognitive samples contribute to domain-specific scores using internal norms from data/v3/cognitive-norms.json. Personality scoring follows the standard IPIP-NEO procedure with BFAS aspect structure. Archetype assignment uses centroid-based variant matching with bootstrap resampling and confidence interval estimation. Every transformation is documented; the pipeline is deterministic.

Cognitive layer

Six CHC-aligned domains.

Raw task performance converts to standardised scores against internal norms, then combines with g-loadings into a composite cognitive estimate.

Personality layer

IPIP-NEO with BFAS aspects.

Sixty personality items resolve to ten BFAS aspects, then aggregate to the five broad Big Five traits on the same standardised scale.

Archetype assignment

Centroid matching with bootstrap CI.

Variant centroids and a Mahalanobis-style metric produce a softmax probability across archetypes; bootstrap resampling yields a confidence interval.

Deterministic by construction. Identical inputs always produce identical outputs.

Full Prism measurement

How Full Prism is measured.

Full Prism runs on the v3 engine. It combines norm-referenced cognitive scoring, BFAS-structured personality scoring, and centroid-based archetype assignment into a single measured profile, with confidence intervals exposed at every layer. Each component of the pipeline is documented below.

The engine is deterministic and offline. No language model generates an item, a score, or an interpretation at runtime. The same answers always produce the same archetype.

Norm-referenced cognitive scoring

Raw task performance is converted to standardised scores against internal norms held in data/v3/cognitive-norms.json. Each domain produces a point estimate with a confidence interval, not a flat number.

Reliability and confidence intervals

Per-domain reliability is a theoretical estimate based on item count, following the Spearman-Brown tradition. Confidence intervals are exposed on every domain and on the composite. Empirical reliability replaces the theoretical estimate once the calibration sample is large enough to compute it directly.

G-loadings and composite estimate

Domain scores are combined with g-loadings derived from the CHC literature to produce a composite cognitive estimate rather than a simple average. The composite reads through the same standardised scale as the domains.

Age correction on verbal items

Verbal items apply an age correction following the Salthouse 2010 vocabulary curves (Prism methodology v3.5 §C.1). Crystallised verbal reasoning grows through adulthood, and the correction prevents systematic over- or under-reading by age cohort.

ESL correction on verbal items

Verbal items apply an English-as-second-language correction (v3.5 §C.2) so that the verbal score reflects reasoning capability rather than English vocabulary depth for non-native readers.

Quality flags

Each session is scanned for low-effort response patterns, implausibly fast responses, and within-domain inconsistency. Flagged sessions surface a validity warning on the result so the read is interpreted with its quality band.

BFAS-structured aspect scoring

The 60 personality items resolve to 10 BFAS aspects, which aggregate to the five broad Big Five traits. Aspect scores are kept on the same standardised scale as the cognitive layer so cognitive-personality interaction terms remain interpretable.

Centroid and Mahalanobis variant matching

Archetype assignment uses centroid-based matching against the 16 variant centroids in data/v3/variant-centroids.json. Distance is computed in a Mahalanobis-style metric that respects within-trait covariance, then resolved into a softmax probability across archetypes.

Bootstrap resampling for confidence

Confidence in the assigned archetype is estimated by bootstrap resampling the score vector under its measurement error and re-running variant matching. The headline confidence number is the proportion of bootstrap iterations that return the modal archetype.

Spike pattern detection

Before archetype assignment, the engine checks the cognitive profile for a true spike: a domain or domain-pair whose elevation exceeds the rest of the profile beyond a t-statistic gate. Spike detection separates a Master-tier read from a balanced Prism.

Confidence

How confidence is computed.

Full Prism treats archetype assignment as a classification problem. The scoring engine estimates evidence for all sixteen archetypes from the cognitive layer, the personality layer, and validity signals, then resolves the evidence into a posterior probability distribution via centroid matching and bootstrap resampling. The headline percentage is the posterior probability of the most likely archetype.

Agreement between the cognitive and personality streams narrows the bootstrap distribution and raises confidence. Streams that point toward neighbouring archetypes widen the band, and the page says so plainly. If two adjacent archetypes sit within a small margin, Prism reports a tie rather than picking a winner by rounding.

Reliability is currently a theoretical estimate based on item count, in the Spearman-Brown tradition. Empirical reliability is published once the calibration sample is large enough to compute it directly.

Confidence intervals are exposed on every domain and on the composite. They surface measurement uncertainty rather than hiding it inside a headline number.

Item parameters refine as the user base grows. The calibration plan is tracked against the first 500 completed reads.

Study Fit measurement

How Study Fit is measured.

Study Fit is a brief screening tool whose construct alignment sits in the learning-strategies tradition: the Motivated Strategies for Learning Questionnaire (MSLQ; Pintrich) and the Survey of Study Habits and Attitudes (SSHA; Brown & Holtzman). Prism writes its own items rather than copying those instruments. The forced-choice format suits screening and resists acquiescence bias.

Behavioural items carry roughly seventy percent of the scoring weight on the recommendation. Cognitive samples contribute roughly twenty percent. Context and qualification fill the remaining ten percent. The weighting is documented in the public source model and is identical across all reads.

Behavioural items (≈70%)

Eighteen forced-choice behavioural items observe self-reported revision habits across planning, monitoring, checking, avoiding, recovering, and testing. Behavioural responses carry the majority of weight on the recommendation because behavioural patterns predict revision outcomes more reliably than short-form cognitive samples.

Cognitive samples (≈20%)

Four short cognitive samples give an independent signal on how the user handles information under light time pressure: digit span for working memory, mental rotation for spatial transformation, verbal inference for crystallised reasoning, and letter comparison for processing speed.

Context and qualification (≈10%)

One context screen (exam timing, subject, stated weak points) and two bounded free-text moments adjust the priority of the first recommendation and anchor the read in concrete examples. They calibrate specificity. They do not alter the underlying weighting.

Forced-choice format

Behavioural items use forced-choice phrasing rather than agree-scale Likert items. Forced-choice format is appropriate for a screening tool: it reduces acquiescence bias and produces cleaner signal in under ten minutes.

Research lineage

Reference instruments.

Study Fit's construct map sits inside an established research tradition: cognitive strategy use, metacognitive self-regulation, time and study environment management, effort regulation, delay avoidance, and work methods. The instruments below shape that map. Prism writes the items in-house.

Motivated Strategies for Learning Questionnaire

The MSLQ (Pintrich and colleagues), the canonical learning-strategy instrument in educational psychology, informs Study Fit's behavioural construct map. Prism does not reproduce, copy, or norm against the MSLQ. It draws from the broader construct language around motivation, strategy use, monitoring, and regulation.

Survey of Study Habits and Attitudes

The SSHA (Brown and Holtzman) anchors the longer study-skills assessment tradition. Study Fit's behavioural items are construct-aligned with SSHA domains such as delay avoidance, work methods, and study attitudes, using original Prism phrasing throughout.

Classical cognitive task formats

Digit span, mental rotation, verbal inference, and letter comparison are conventions established in classical psychometrics and cognitive psychology. They are public-domain task formats rather than proprietary content. Prism writes its own items inside these formats.

Study Fit limits

What Study Fit does not claim.

Study Fit is a screening tool, not a clinical assessment.

It does not diagnose dyslexia, ADHD, autism, learning disabilities, anxiety, or any medical condition.

It does not predict examination outcomes deterministically. It names a likely revision bottleneck and a first move to test.

It does not replace deeper instruments like the full MSLQ, a full psychometric battery, or professional educational assessment.

Our principles

Four rules we hold ourselves to.

We use public-domain methodology and cite the academic tradition we draw from.

We write our own items. We do not use copyrighted test content from commercial assessments.

We name our limits openly on every page of the product.

We do not pretend a short online read is a clinical assessment.

Scope and limits

What Prism does not claim to do.

Prism is not a clinical assessment, medical diagnosis, IQ ranking, or hiring instrument.

Short-form cognitive tasks produce signal, not a full psychometric battery.

The recommendation is a working hypothesis to be tested against real outcomes.

The Application Prompts use deterministic prompt assembly, not runtime AI psychological inference.

Source note

This page is the public mirror of the repository methodology document. The full source model lives in METHODOLOGY.md alongside the engine code, including item-level psychometric notes, calibration plans, and version history.