About our test.

The short explanation

You answer an adaptive 25–35 item session sampled from a 120-item reviewed bank across pattern, numeric, verbal, spatial and working-memory tasks. Selection responds to earlier answers inside fixed coverage constraints, and retakes prefer unseen items on the same browser. The report shows observed accuracy, a proprietary AurorIQ reasoning index and a descriptive profile. The index is not an IQ score, percentile or population norm.

Maintained by Muhammad Ali Rao. Explore the technical details, references, change record or report a problem.

How the reasoning assessment works

AurorIQ uses a content-balanced adaptive reasoning engine over a 120-item reviewed bank. Each item has a developer-seeded difficulty band and discrimination weight used for routing. After each response, the engine updates a bounded Bayesian/EAP-like internal estimate over a fixed theta grid and selects among eligible reviewed items near that modeled level while enforcing domain coverage, retake unseen-item preference and lightweight exposure diversity.

The internal theta grid is bounded from −3 to +3 with a normal prior. Posterior spread is retained as an uncertainty diagnostic. These mechanics are designed for stable browser behavior with short sessions; they are not evidence that the seeded item values are empirically calibrated or that the assessment is a validated computerized adaptive test.

Scoring breakdown, step by step

What the five domains represent

The assessment samples pattern recognition, numerical reasoning, verbal reasoning, spatial reasoning, and working memory. Those areas are relevant to broad traditions in cognitive measurement, but AurorIQ's five-domain design is its own practical online battery. It should not be described as equivalent to a WAIS, Stanford–Binet, Woodcock–Johnson, or any other proprietary, norm-referenced clinical test.

The overall estimate uses the full response set. Each domain receives at least five and at most seven items under the current blueprint. Domain estimates are calculated from the responses within that domain and remain cautious within-session indicators. Five to seven seeded items do not justify precise subscale or population claims.

Validity: what evidence exists today

In psychometrics, validity is not a badge a test earns once; it is an evidence-based argument that a particular interpretation of scores is appropriate for a particular use. The Standards for Educational and Psychological Testing, produced jointly by AERA, APA, and NCME, emphasizes validity, reliability/precision, and fairness when tests are developed and interpreted. AurorIQ uses those ideas as an evaluation framework, but we do not claim formal compliance, certification, or endorsement by those organizations.

Accordingly, the strongest defensible interpretation is: AurorIQ provides a model-based educational estimate of performance on this item bank under the conditions of this sitting. It does not provide a clinical diagnosis, educational eligibility decision, employment credential, legal finding, or official IQ classification.

Calibration and norming status

The difficulty bands and discrimination weights in the present item bank are developer-selected starting values. They have not been estimated from a suitable participant response sample. The AurorIQ reasoning index is therefore a product model index, not a norm-referenced transformation.

A stronger future calibration program would require enough quality-controlled responses to estimate item parameters, inspect item fit and local dependence, detect differential item functioning across relevant groups, evaluate test information across the ability range, cross-validate on a holdout sample, and publish uncertainty and sample limitations. Until that work exists, we deliberately avoid calling the test “clinically validated” or “professionally normed.”

Reliability, precision, and repeat scores

The engine retains posterior spread and recent estimate movement as internal uncertainty/stability diagnostics under its developer-seeded model. These do not establish empirical uncertainty coverage or test–retest reliability. No population confidence interval is shown. Repeated and independent measurement studies would be needed to support stronger precision claims.

Real users can score differently across attempts because of sleep, stress, distraction, guessing, learning the item format, prior exposure to the same questions, device conditions, and performance-responsive item order. A retake should therefore be treated as another measurement, not proof that intelligence suddenly changed.

Fairness and interpretation standards

Good testing practice asks whether score interpretations remain appropriate across the populations who take the test. AurorIQ is currently an English-language online instrument, so verbal items can be affected by language exposure, education, culture, reading experience, and familiarity with online test formats. Numerical items can reflect schooling as well as reasoning. Timed tasks can be affected by hardware, input devices, browser performance, motor speed, and accessibility needs.

Those influences are not reasons to hide results; they are reasons to interpret them carefully. We do not currently claim measurement invariance or equal predictive meaning across languages, cultures, disability groups, ages, or educational backgrounds. If future data show meaningful group-specific item behavior, the correct response is to revise the item bank or interpretation—not to explain away the discrepancy.

Descriptive archetypes

Archetype names summarize the relative pattern of your results within AurorIQ. They are descriptive labels, not validated psychological types. Product tiers also summarize model output, not population standing. No simulated rarity is displayed.

How test data is handled in this codebase

The assessment engine is implemented client-side. The current JavaScript test engine does not contain fetch, XMLHttpRequest, sendBeacon, WebSocket, or similar network calls for transmitting answer-by-answer test data. In-progress state and recent results are handled through browser storage so the experience can work without a user account. Theme and consent preferences are also stored locally.

Analytics and advertising are separate from the scoring engine. When enabled, Google Analytics and our advertising partner Monetag can process site-usage, device, cookie, and advertising data according to the user's choices and each provider's policies. The details and user choices are documented in our Privacy Policy and Advertising Disclosure.

Research and standards used as reference points

The Testing Standards (2014) supply a framework for considering interpretation and appropriate use. Embretson and Reise (2000) is a scholarly reference on IRT concepts. Neither certifies this implementation. Evaluate the code and missing empirical evidence separately.

Inspect the current implementation

Items and answer keys are delivered to the browser in the question bank. Inspect adaptive estimation, scoring and domain definitions. Each reviewed item has a stable versioned ID, domain/subdomain, original rationale, developer-seeded difficulty band and discrimination weight, exposure group, review status and accessibility metadata. There is no claim that a published paper supplied or empirically calibrated these exact items.

The response model uses a bounded logistic likelihood over theta values from −3 to +3 with a normal prior. The EAP-like estimate is the posterior mean and posterior spread is used only as an internal uncertainty signal. Item routing favors developer-seeded difficulty near the current estimate while respecting the five-domain blueprint and retake exposure controls. These are engineering choices, not empirical IRT calibration.

Raw accuracy is correct answers divided by answered items. The displayed AurorIQ reasoning index maps the bounded model estimate to a proprietary 20–80 index and is not an IQ score or percentile. Domain estimates use only their own five to seven responses and are withheld if minimum coverage is not met. The descriptive archetype summarizes relative within-session model patterns rather than a validated psychological type.

The source includes a scoring regression script for maintainers. Correct software arithmetic does not establish psychometric validity. Calibration, test–retest reliability, fairness and external comparisons remain work that would be needed, not completed studies or a promised research schedule.

References

1. AERA, APA & NCME (2014). Standards for Educational and Psychological Testing. American Educational Research Association.
2. Embretson, S. E., & Reise, S. P. (2000). Item Response Theory for Psychologists. Lawrence Erlbaum Associates. Link identifies the later digital edition.
3. Hausknecht, J. P., et al. (2007). Retesting in Selection: A Meta-Analysis of Coaching and Practice Effects for Tests of Cognitive Ability. Journal of Applied Psychology, 92, 373–385.

Methodology change record

These entries describe package changes. They do not claim when a release became live.

Date / version	Change	Reason
V6.30 — prior package	Removed population-percentile, age-adjustment, rarity and displayed numeric confidence-interval claims; retained raw counts and a labeled model estimate.	The bank has no empirical calibration or representative norms. Historical publication date is not independently verified.
2026-10-05 · V6.31	Canonical-host and index.html redirects, sitemap and metadata cleanup.	Make public URL identity consistent; no scoring change.
2026-10-05 · V6.32	Named founder responsibility; added claim-specific sources, privacy/share disclosures and corrections route; corrected percentile explanation.	Make evidence and accountability inspectable. No engine, item or scoring-weight changes.
2026-10-06 · V6.33	Added evidence-led reports, optional goal framing, practical experiments and full exports. Raw ties and small model-index gaps receive cautious interpretation.	Make results more useful without changing scoring, item parameters or validation status.
2026-10-07 · V9.00	Expanded the reasoning bank to 120 reviewed versioned items; added content-balanced adaptive subset selection, bounded Bayesian/EAP-like estimation, variable 25–35 item stopping, local unseen-item preference, exposure controls, simulation QA and a non-IQ proprietary result index.	Improve session diversity, retake protection, uncertainty handling and future calibration readiness without claiming validation or norms.

Limits you should keep in mind

• Not a clinical instrument. Do not use AurorIQ for diagnosis, disability evaluation, educational accommodations, legal decisions, hiring, credentialing, or high-IQ society admission.
• No representative AurorIQ norms yet. The reasoning score is a model transform. No age correction, population rank or rarity percentage is reported.
• Item parameters are developer-selected. They still need empirical calibration and cross-validation.
• Per-domain estimates are especially uncertain. Roughly five items per domain is too little evidence for precise subscale interpretation.
• Language and schooling matter. Verbal and numerical performance can reflect experience as well as reasoning.
• Device and state matter. Timing, attention, fatigue, stress, and hardware can change results.
• Practice effects are real. Retesting research supports caution about repeated scores; it does not quantify practice gains for AurorIQ.

Who built and maintains AurorIQ

AurorIQ is an independent project created and maintained by Muhammad Ali Rao for exploring cognitive tasks and self-reflection tools. The project is intentionally explicit about the difference between using psychometric methods and having a clinically validated instrument.

Muhammad Ali Rao is responsible for maintaining the product and published explanations. AI assistance is disclosed in the editorial policy; no independent professional review is claimed. Advertising does not determine scoring or editorial conclusions. Corrections are welcome at contact@auroriq.com or through the Contact page.

Frequently asked questions