Moving from the "ownership" of knowledge and information to "processing"—how we use them.
Practigence is a next-generation assessment tool that visualizes the process of filtering cognitive noise and making essential decisions in today's rapidly changing world.
Practigence objectively reveals "fluid intelligence" and "metacognitive ability," which traditional written assessments and multiple-choice tests fail to capture, aiming to optimize intellectual resources through placing the right talent in the right place.
Rather than measuring Raymond Cattell's "crystallized intelligence (accumulation of knowledge)," it precisely measures "fluid intelligence"—the capacity to identify dynamics among variables from incomplete information and dynamically update hypotheses under rapidly changing situations.
As information supply becomes effortless due to generative AI, this tool evaluates "information filtering capacity" and "task formulation capacity" to discard cognitive noise and select only essential variables, keeping in mind the limitations of brain working memory (Miller's Law).
It tracks not only final answers but also response latency (thinking time) and modification logs (Process Tracing). Integrating multi-faceted evaluation engines achieves fair assessment unaffected by the cognitive ceilings of evaluators (Ceiling Effect).
Simulate how differences in brain "processing structures" yield different outcomes in real-world problem-solving.
Practigence aims to promote the true placement of the right talent in the right place, optimizing intellectual resources in society. By establishing an environment where individuals can accurately grasp their own objective cognitive structures (CPU characteristics) and make optimal, autonomous decisions, we eliminate inefficiencies in intellectual productivity and enhance society's overall problem-solving capability.
A dynamic, hierarchical cognitive processing capacity that metacognitively evaluates environmental suitability, autonomously selects, constructs, or discards environments to adapt to, extracts and reconstructs invariant logical structures from experience to understand core essences, models invisible causal dynamics as systems, flexibly updates hypotheses (Bayesian updating) even from minimal information, and simultaneously solves future risk prediction/prevention and current complex challenges.
As metacognitive theory in psychology (Flavell, 1979) shows, correct self-awareness of cognitive biases (self-calibration) requires advanced cognitive capability, and correcting biases solely by oneself can be difficult (Dunning-Kruger effect). Furthermore, because self-reported questionnaires trigger the "social desirability bias," leading respondents to unconsciously polish their answers, objective assessment free from subjective intervention becomes essential.
Objective measurement unravels thinking processes that tend to be treated as black boxes:
There is a common discourse in defense of education stating that "mathematics cultivates logical thinking." However, problems in mathematics are mostly closed environments where conditions are explicitly stated and irrelevant information is removed. In contrast, real-world decision-making deals with incomplete variables and varying information freshness.
While saying "because someone is good at math, they must possess logical thinking" is inappropriate, saying "because someone has logical thinking, they will likely be good at math" is appropriate. Mathematics is not logic itself; rather, logic encompasses mathematics.
What humans use to make decisions and communicate in society is vast amounts of "verbal logic" and only a tiny fraction of "mathematical logic." Despite this, overemphasizing mathematical logic just because it is easier to measure in school tests is inappropriate.
Against the correlation fallacy that equates good academic grades with high intelligence, Practigence provides a perspective to look straight at the contents and properly analyze and evaluate psychological biases where correlation fails.
This tool provided by Practigence is designed to evaluate fluid intelligence based on a framework that decomposes and classifies cognitive processing required for real-world problem-solving into functional units.
This tool organizes what cognitive, judgment, or resolution failure occurs when specific functions are lacking, translates each capability into a measurable format, and conducts measurements using that framework.
Based on this definition, we perform functional decomposition to break down capabilities, devise and create methods to draw out, verify, and analyze whether those sub-items are successfully demonstrated, and generate reports from the collected data.
To explain the decision-making process in daily life and business, let's consider a familiar scenario: **"purchasing a sofa."**
When considering a sofa purchase, one typically reviews basic variables such as "dimensions to fit the room," "price," "durability," and "color." However, if the attribute of a "single person who relocates frequently" is added, "ease of disassembly and transport" and "resale value" turn into "essential variables" for the decision. Conversely, for someone with an extremely spacious room, constraint variables regarding dimensions become irrelevant "noise."
As Cognitive Load Theory shows, our brain's working memory capacity is limited. If unnecessary information (noise) cannot be discarded from the head, processing resources required for judgment saturate, causing decision freezes or judgment errors.
Practigence evaluates this dynamic **information selection & functional decomposition ability (CD)** to determine "which variables are essential and which are discarded as noise in response to shifting premises."
In the purchase scenario explained above, click on different buyer attributes to visualize which variables become essential and which become unnecessary "noise."
Basic variable to check if it fits the placement area.
Practical variables relative to budget and usage period.
Important constraint condition that arises when relocation is assumed.
Variable to prevent back-end loss when letting go after short-term use.
Layout information of other rooms with completely different shapes is noise that misleads judgment.
Essential in standard placement planning, but its consideration weight drops to zero in extremely large rooms.
<Assumed Situation>
At an unfamiliar place, Train A (20-minute ride) heading to your destination suddenly suspends service due to track intrusion. The departure of the bus you must catch at the destination (absolute deadline) is in 30 minutes. You have only 2 minutes to decide.
Under this tense situation, a person with high dynamic intelligence executes the following "variable classification" and "high-speed computation" in their head instantly.
| Option | Presented Surface Info | Variable Classification & Processing in Brain | Computation Result |
|---|---|---|---|
| Route A: Train (20 mins) | Currently suspended. May resume shortly. | Deems "track intrusion" as a variable parameter with maximum volatility, ranging from minutes to hours depending on police intervention. Excludes immediately due to high risk against the 30-minute absolute deadline. | Immediately Rejected (Cut Loss) |
| Route B: Bus + Train | Bus arrives in 20 seconds, but route is uncertain and slow. | The 20-second window physically fails because "recognition and movement costs" in an unfamiliar place exceed "available time." Also, travel time is long (noise cut). | Immediately Rejected (Noise Filtering) |
| Route C: Subway (25 mins) | Travel time is 25 minutes against 30 minutes remaining. | Determines that the subway, structurally less affected by external ground noise, has 25 minutes of travel time as the "most reliable fixed parameter." Calculates that a 5-minute buffer is adequate to absorb transfer loss. | Confirmed (in 2 mins) |
Recruiter Cognitive Limits and "Process Black-Boxing":
The essence of this decision-making is cool-headed expectation value optimization: choosing the option with the highest probability of fitting within the deadline (30 mins), even if it means discarding the fastest (20 mins) possibility.
Generally, the word "experience" is misunderstood as something whose value increases in proportion to time spent. However, length of experience (accumulation along the timeline) merely proves the physical fact of staying in that environment.
Time spent in uncritical inertia, without changing anything in the system or one's own actions, does not contribute to updating intelligence, no matter how vast it is.
The quality and speed of learning—how much one learns from experience—depends not on time length, but on how much pure logical structure or essential variables one can [extract] from that experience (concrete events).
Accumulates concrete, surface-level events as-is (HDD model). Since context-dependent knowledge is stored as static memory, it cannot be applied when variables or environments change.
Extracts abstract, invariant logical structures from concrete events (CPU model). By generalizing experiences as functions, it allows immediate application and hypothesis updating in unknown, novel tasks.
Data computed after taking the assessment is provided as reports in three grades depending on use and objectives (currently, Tier 1 and Tier 2 are accessible to examinees).
A simplified report describing the specific movements of cognitive processes observed during the test in text. It provides 5 to 7 lines of behavioral observation logs without scores or tags, such as: "tended to rush to conclusions in multiple tasks" or "integrated new information while retaining the initial hypothesis" (analogous to recording "shallow sleep this week" in a health checkup context).
Our core report showing levels, analytical comments, strengths, and weaknesses across "13 cognitive clusters" (such as logical analysis, reasoning, and verbal transmission), along with specific recommended improvement actions like "asking specific questions immediately after daily decision-making" (analogous to functional evaluation and guidance like "Cardiopulmonary: Good / Sleep: Needs Improvement" in a health checkup context).
Detailed data for professional analysis mapping presence/absence for all 80+ high-order cognitive atomic tags per task, including pattern analysis of how cognitive structures collapse under specific question types, and automatically computed tags requiring prioritized improvement (analogous to a complete list of blood test values used for research and detailed analysis).
Practigence comprehensively decomposes thinking processes from the moment a human "recognizes" things to "judgment/decision-making," defining and measuring them in domains A to N, transcendent ability M, and holistic integration P.
Because traditional checklist evaluations measure superficial behaviors (What) rather than individual processing mechanisms, they run the risk of misidentifying highly intelligent individuals as having "developmental disorders." Click tabs to compare.
Fails to sustain attention on specific tasks due to difficulties in working memory control, unintentionally looking at other stimuli. Appears unable to maintain attention when facing boring tasks.
Because the brain's processing speed is too fast, simple tasks are completed instantly, leaving working memory mostly vacant. Consequently, to maintain the brain's optimal arousal level, the individual voluntarily starts separate, highly demanding thoughts or intellectual exploration (mental multitasking). This is not "inattention," but self-protective optimization to prevent brain stall. Shows extreme focus on difficult tasks.
Difficulty deciphering other people's intentions or unwritten social contexts (like reading the room). Does not follow irrational rules. Thought processes skip steps and appear odd. Ignores implicit rules.
Because premises, levels of abstraction, and logic processing speed differ too much from counterparts, the individual's sound reasoning or essential arguments are not understood by peers. Non-compliance with rules is because the "uselessness of compliance" is too logically clear. Skip-step reasoning makes conclusions appear "abrupt and odd" to those around them.
As the Law of Requisite Variety (Ashby, 1956) in cybernetics shows, evaluating and controlling a complex system requires the evaluating side to possess equal or greater variety (complexity of states). If there is a massive gap between the evaluator's (manager, teacher) cognitive traits and the examinee's, the evaluator may fail to recognize the examinee's advanced processing or meta-perspectives, causing hiring or management mismatches.
Thus, actions that are actually "voluntary exploration to avoid boredom" or "custom variable design" are superficially mislabeled as "inattention" or "lack of cooperation" due to the structural limitations of conventional systems.
Need for Multidimensional Intelligence Models:
The dominance of knowledge (crystallized intelligence) in teaching and the dominance of processing (fluid intelligence) in solving unknown tasks lie on different dimensions. Rather than evaluating superficial behavior (What) like traditional tests, modeling behavior as an interaction (function) of "internal resources (CPU)" and "external task load (premises)" excludes misclassifications from cognitive limits, capturing the individual's essential cognitive traits objectively.
Directly classifies the single output of surfaced "behavior," ignoring underlying contexts and variables.
Rather than directly classifying superficial behavior, it properly factorizes it as an interaction (function) between "internal resources" and "external environment."
When facing a situation, which variables or noises are first caught intuitively. This is the starting point of all cognitive processing. Failures in recognition at this level directly lead to subsequent analysis or judgment errors.
Fine-Grained thought Process Visualization
Multiple AI evaluation engines analyze your answers independently. Instead of simple pass/fail or correctness, they map and visualize the micro-level thought processes, including where logic leaps or hypotheses are revised.
Pure Measurement of Thinking Capacity
Unlike typical tests that require memorized facts, it evaluates your capacity to identify variables and infer logic when facing unfamiliar problems. This ensures no bias based on academic background or major.
Sophisticated Scoring Weights
Rather than scoring easy mistakes and hard successes equally, scoring is weighted according to task complexity and variable density. This accurately evaluates the examinee's true processing boundary.
Countering Bias and Skewed Judgments
To capture outstanding talent without oversight, prompt parameters and sequences are structured to avoid evaluation bias. This guarantees an objective, fair, and highly accurate assessment.
Dynamic Item Variations
Key variables change dynamically with each session, making memorized answers or test-prep techniques completely useless. The test measures real-time processing rather than static memory.
No Personal Data Mapping & Cooldowns
Test results are not linked to identifiable personal information to secure privacy. Additionally, a strict cooldown interval is enforced before retakes to eliminate simple practice effects.
Self-understanding of cognitive traits and guidelines for cultivating future intelligence
Enables objective calibration of one's own thinking habits, such as: "in what situations are judgment biases likely to occur" and "which information failed to be processed as noise." Going beyond merely revealing current strengths and weaknesses, it provides concrete guidelines on how to foster irreplaceable high-order cognitive skills unique to humans in the AI coexistence era.
Measuring processing capacity (CPU) adaptable to the AI era and preventing hiring mismatches
Evaluates pure thinking and adaptability (CPU) to tackle unknown tasks, rather than volume of memorized knowledge (HDD). Because the structure makes it difficult to fake or prepare in advance, recruiters can objectively grasp "true cognitive processing capacity"—which is hard to evaluate through interviews alone—drastically reducing organizational costs associated with hiring and placement mismatches.
Upgrading evaluations from knowledge-overemphasis to cognitive processing, and promoting the right talent in the right place
Upgrading the conventional evaluation infrastructure—which only favored those who remembered a lot of past knowledge (crystallized intelligence)—to center on the "high-order fluid intelligence" required in the highly volatile and unpredictable AI era. We promote the right placement of cognitive resources according to each individual's diverse cognitive structure, encouraging a collaborative society where individuals leverage strengths to complement one another autonomously.
This test is not a "trivia quiz to guess the correct answer." We look at **how you processed information and made judgments**. Therefore, rather than trying to write a polished answer, it is most important to **express your thought process in words as it is**. Reading the following before starting will help you demonstrate your capabilities to the fullest.
There is a Consent Gate prior to testing. The test has a 2-part structure (Part 1 and Part 2) and can be safely paused and resumed at any time.
Part 2 must be taken within a certain period (48 hours) after completing Part 1. The expiration time is clearly indicated on the screen.
Rather than fixed memorization items, tasks are designed to extract your thought process. Some items change dynamically per session, rendering test-prep ineffective.
Provides your level from L0 (base unreached) to L11 (flawless), profiles by field, and a breakdown of "what is working / what is not."
Results are indicated as levels from L0 (base unreached) to L11 (flawless). This level is not based on a single sharp skill, but is integrated such that you only advance when Perception, Analysis, Thinking, Design, Objectivity, and Verbalization all function simultaneously. At the highest tiers, processing speed is also included in the conditions.
The report clearly distinguishes between "active skills" (currently functioning stable strength) and "latent potential/talent" (inspiration or possibility that is not yet stable) using graded labels. Please use this not as a label of superiority but as a guide for your cognitive evolution and future learning.
This page describes the construct concepts of the measurement, measurement models, scoring architecture, reliability assurance, and the **arguments and limitations of validity at this point** without exaggeration. We distinguish between what is achieved and what remains unestablished in a verifiable manner.
The primary measurement target is **fluid intelligence**—the ability to identify relationships and structures from novel, incomplete information and dynamically update hypotheses independent of acquired knowledge. Along with this, we evaluate **self-objectivity (calibration) based on metacognition**. The theoretical background of these constructs, the mapping of each field to research constructs, and references are aggregated in the "Theoretical & Research Basis" page.
Cognitive processing required for real-world decision-making is **decomposed into functional units**, systemized so that "which cognitive, judgment, or resolution failure occurs when specific functions are lacking" can be backward-computed. Decomposed functions are aggregated into cognitive clusters, and attainment is expressed in progressive levels. The targets of observation are assigned not based on "given question formats" but starting from **what can be observed in each task (observability)**; the mapping between target constructs and tasks is traceable by design.
Scoring is not done per question; instead, evaluation criteria are judged **across all responses**, suppressing localized judgment volatility. Judgments are made by **consensus among multiple independent large language models**, and only instances where models disagree are arbitrated by a higher-order model. This architecture aligns with findings in evaluation research showing that ensemble (consensus) and reasoning-explicit designs yield higher consistency and transparency than single models.
Each evaluation criterion is assessed across all responses, applying reasoning and confidence levels.
Evaluates qualities and flaws as a whole that cannot be reduced to individual criteria.
Multiple independent models score and cross-reference results.
Higher-order model arbitrates only where judgments diverge.
* Designed to route tasks by degree of disagreement: areas resolved by consensus are processed lightly, while only split-decision points are forwarded to higher-order arbitration (balancing cost and accuracy).
**Content Validity:** Each task has evaluation criteria assigned based on observability; design-wise, the mapping between target constructs and tasks is traceable. **Construct Validity:** The hierarchy of functional decomposition -> clusters -> levels is structured to correspond to the theoretical frameworks of fluid intelligence and metacognition. It intends to capture the "quality of processing" that existing metrics (which measure only outcomes) miss, by observing processes. For details on correspondence with theory and research, see the "Theoretical & Research Basis" page.
The system's design was established first, and verification/research comparison was conducted post-hoc to support and confirm its correctness. This evaluation is based on observing the thought process, measuring fluid intelligence (dynamic, CPU-like intellect), and does not define a person's human value. With these premises in mind, we honestly disclose the following limitations:
This page aggregates the theoretical and research foundations with which Practigence's **intelligence framework (functional decomposition of abilities) and test design** align, demonstrating how each field maps to established research constructs and how alignment with the references was verified.
**Design came first; verification, validation, and refutation matching were performed afterwards.** The research, findings, and refutation studies mentioned on this page were all **used to assist and verify the accuracy of the designer's own conceptualization, design, and hypotheses**. Rather than using external findings as a starting point or adopting them as conclusions, they were referenced post-hoc as comparison materials to confirm alignment between already constructed frameworks/test designs and hypotheses. Therefore, references on this page are not intended as replication, replacement, or borrowing authority from each study.
The primary measurement targets are **fluid intelligence**—the ability to identify relationships and structures from novel, incomplete information and dynamically update hypotheses independent of acquired knowledge—and **metacognition (self-objectivity/calibration)** to monitor and control one's own cognition. These have been confirmed to align with the following lineages of research:
Each field in this framework is not an arbitrary division, but corresponds to constructs repeatedly identified in cognitive science and intelligence research. The validity of the framework is supported by these correspondences.
| Framework Field | Corresponding Research Construct | Primary Basis |
|---|---|---|
| A Recognition / CD Selection | Selective Attention, Relevance Filtering, Cognitive Load Management | Cognitive Load Theory (Sweller) / Attention Research |
| B Analysis / C Understanding | Relational Inference, Structure Mapping, Inductive Reasoning (Core Gf) | Structure Mapping Theory (Gentner) / Cattell–Horn–Carroll |
| D Memory / Ds Selective Memory | Working Memory, Abstraction & Analogical Transfer | Baddeley & Hitch (1974) / Analogical Transfer (Gick & Holyoak) |
| E Thought | Fluid Reasoning, Hypothesis Verification, Bayesian Belief Updating | Cattell (1963) / Bayesian Cognition (Tenenbaum et al.) |
| F Design/Solution / G Social Implementation | Ill-Defined Problem Solving, Practical Intelligence | Problem Solving Research / Sternberg's Practical Intelligence |
| K Self-Objectivity / L Holistic Objectivity / I Env Selection | Metacognition, Self-Regulation, Executive Function | Flavell (1979) / Nelson & Narens |
| J Cognitive Endurance | Sustained Attention, Cognitive Durability | Sustained Attention Research / Recent Empirical Evidence on Cognitive Endurance |
| M Fusion / P Holistic | General Intelligence Factor g, Integration of Abilities, Successful Intelligence | Spearman's g / Carroll (1993) / Sternberg's Successful Intelligence |
**The validity of centering on "fluid intelligence × metacognition"** is also reinforced by recent intelligence research. Fluid reasoning is stably identified as the core of the CHC hierarchy, and metacognition/executive functions are positioned as independent predictors of learning outcomes and adaptive judgments. * This indicates correspondences at the field level.
Following the widespread adoption of generative AI, empirical studies show a decline in the relative value of knowledge-reproducing tasks and an **increased demand for high-order cognitive skills**. Jobs that assume the use of generative AI tools exhibit significantly higher cognitive skill requirements (Hampole et al., 2025, arXiv:2503.09212), and analyses by the IMF and McKinsey also highlight impacts on knowledge work and a **supply shortage of critical thinking, problem structuring, and complex information processing** (IMF Staff Discussion Note, 2024 / McKinsey Global Institute). The constructs measured by this tool respond directly to this supply shortage area.
The design philosophy of using "consensus among multiple models" and "requiring explicit reasoning" in scoring aligns with recent LLM evaluation research. It has been reported that ensemble (consensus/majority vote) and explanation-requiring designs yield higher agreement and transparency than single models (Zheng et al., 2023 / Research on Reliability of Multilingual Evaluation, 2025). At the same time, it has been demonstrated that in judgments requiring specialized expertise, the agreement rate between humans and AI drops compared to general tasks (falling to 64–68% in specialized domains compared to ~90% human-AI agreement in general tasks, which is lower than the inter-expert agreement of ~72–75%), serving as the basis for this tool's recognition of its own limitations.
| Reference | Application/Correspondence in Framework & Test |
|---|---|
| Cattell (1963) / Carroll (1993, CHC) | Basis for positioning fluid intelligence as the primary target of measurement. Conceptual background of level hierarchies. |
| Flavell (1979) / Nelson & Narens | Measurement basis for metacognition and self-objectivity (Domains K/L). |
| Dunning & Kruger (1999) | Basis for the design measuring gaps between self-declaration (pre/post-test) and actual measurements. |
| Miller (1956) / Baddeley & Hitch (1974) | Cognitive load design for memory (D) tasks, noise filtering, and information overload tasks. |
| Sweller (Cognitive Load Theory) | Design of cognitive load in long-form information processing questions. |
| Ashby (1956, Requisite Variety) | Theoretical explanation of the evaluator-ceiling problem, motivation for multi-model consensus. |
| Kahneman (2011, Dual Process) | Perspective for capturing validation of intuition and deliberate processing. |
| Gentner (Structure Mapping) / Gick & Holyoak | Design basis for analysis/understanding (B/C) and analogical transfer (Ds / transfer problems). |
| Sternberg (Practical/Successful Intelligence) | Construct background for social implementation (G) and ability integration (P). |
| Schulte-Mecklenbeck et al. (2011) | Methodological basis for process tracing (response latency, edit logs). |
| Zheng et al. (2023, LLM-as-a-Judge) | Alignment with consensus and reasoning-explicit scoring designs. |
| Multilingual LLM-as-a-Judge (2025, arXiv:2505.12201) | Supporting evidence for lower agreement rates in specialized domains = recognition of limitations. |
| Hampole et al. (2025, arXiv:2503.09212) | Supporting evidence for increased demand for high-order cognitive skills = societal necessity. |
| IMF SDN (2024) / McKinsey MGI | Impact on knowledge work and supply shortage of critical thinking & problem structuring. |
* The above references are intended to position constructs, methods, and limitations; they do not assert that this tool is a replication or replacement of each study.
These Terms apply to all relationships between the provider of the Service (hereinafter "the Provider") and examinees. Examinees must review and agree to these Terms and Consent Items before starting the assessment. If you do not agree, you cannot take the assessment.
The validity of measurement is compromised when examinees familiarize themselves with question formats and tasks (measurement contamination due to learning/practice effects). To prevent this, a specific cooldown period is established for retakes.
This Service measures "raw cognitive processing"; the following actions are prohibited. If violations are confirmed, measures such as invalidation of the assessment, non-provision of results, or suspension of use may be taken.
* Tab switching, paste attempts, and screenshot operations during the test are recorded and may be used for fraud detection. Collection of question data and reverse engineering are strictly defined in Article 3-2.
Questions, sub-questions, options, model answers, scoring logic, evaluation frameworks, tags/judgment criteria, judgment methods, model configurations, user interfaces, and all other components of the Service are intellectual property and trade secrets belonging to the Provider and legitimate rights holders. Examinees must not perform any of the following acts:
In response to violations, the Provider may take legal action including injunction claims, damages claims, and criminal charges, in addition to invalidating the assessment and suspending Service use.
The validity and enforceability of specific provisions vary based on applicable law, jurisdiction, and individual circumstances. These constitute contractual agreements between the Provider and examinees; their finalization and operation require verification by an attorney. These Terms do not constitute legal advice.
* This Article defines the Provider's handling policy. Details regarding specific retention periods and management frameworks follow a separately defined Privacy Notice.
* Specific amounts and plan details will be reflected in this Article once finalized.
Examinees shall confirm and agree to each of the following matters before starting the assessment (consent is obtained for each item on the pre-test screen):
The Provider may revise these Terms as necessary. Important changes will be announced on this page. Taking the test after revision constitutes agreement to the revised Terms.