DECIDE-CTA: A designer-empowered method for creating structured cognitive task analyses from product instructions

Abstract

Designers are required to perform human factors testing to demonstrate that their product is safe to use before it reaches the market, which includes identification of potential use errors. While this evaluation often focuses on physical tasks, many errors are driven by cognition. Cognitive Task Analysis (CTA) offers a structured method to expose the cognitive processes that shape physical actions, enabling deeper identification of tasks to evaluate. However, existing CTA methods were developed for cognitively trained experts, making them difficult for most design teams to apply. To improve accessibility, the underlying structure of existing CTA methods was combined with (a) embedding of discrete cognitive tasks within psychomotor information-processing sequences, (b) human-machine interaction principles, and (c) syntactic task guidance. This new framework empowers designers to incorporate cognitive depth into their product development considerations and risk assessments. The method, DECIDE-CTA (Designer-Empowered Cognitive Integration for Decomposition and Evaluation), was created by following the steps outlined below:

• •Develop a cognitive-psychomotor unified model, accessible to designers for CTA generation (Phases 1 and 2).
• •Create Task Categories linking designer knowledge to the cognitive-psychomotor unified model (Phase 3).
• •Create rule-based cognitive sentence syntaxes and CTA templates for describing the tasks and ordering the information within a CTA (Phase 4).

Graphical abstract

Specifications table

Subject area	Engineering
More specific subject area	Human Factors, Cognitive Psychology, Safety Science, Risk Assessment
Name of your method	Designer-EmpoweredCognitiveIntegration forDecomposition &Evaluation - Cognitive Task Analysis (DECIDE-CTA)
Name and reference of the original method	Cognitive Task Analysis [1] Bloom’s Taxonomy [2] Harrow’s Taxonomy [3]
Resource availability	https://github.com/jjoyner2/IFU_Python_Code.git

Background

Designers perform human factors testing to assess user performance during product interaction and mitigate use errors that can lead to injuries, death, or harm to the environment [[4], [5], [6]]. Physical activities (e.g., pressing a button), which are easily observable, are commonly the central focus of this assessment [7]. However, many use errors originate from cognitive activities performed before or during the execution of a physical activity (e.g., evaluating the need to press the button) [[8], [9], [10]]. For example, a designer may observe a user performing physical actions in the wrong order without understanding the underlying reason (e.g., confusion with instructional materials, cues not visible) [8,11]. Exclusion of this underlying cognitive information can lead to an incomplete picture of human performance, limiting subsequent risk mitigation strategies [11]. Identifying cognitive use errors during human factors testing requires systematic methods for reliably tracing discrete cognitive activities.

Cognitive Task Analysis (CTA) is a family of human factors research methods aimed at understanding tasks that require significant cognitive activity, such as attention, problem solving, and memory [1]. CTA supports cognitive activity traceability by deconstructing product interactions into subtasks [1,12,13]. Existing methods are grounded in cognitive models, requiring extensive training of analysts in relevant disciplines (cognitive psychology, neuroscience, etc.), which is not commonly found in members of traditional design teams [[14], [15], [16]]. In addition, CTA relies heavily on subjective or analyst-defined input, creating high variability in methodological outcomes [1,17]. For example, a critical step in CTAs is generating a list of cognitive tasks to complete a high-level goal. While cognitive experts have extensive knowledge of these types of tasks, there is no guarantee that experts will consistently generate the same discrete list using the same descriptive task language [18]. CTA methods also do not provide standardized guidance or definitions for task characteristics, which can affect shared understanding of the resulting methodological output [18].

This work seeks to improve the standardization of CTAs by providing a structure for designers to cognitively decompose product tasks, grounded in their knowledge of human-machine interaction (HMI). The resulting method makes cognition more accessible to designers, helping them identify unobservable use errors and ensuring alignment with human factors regulatory requirements when seeking to get their product to market [19,20].

Method details

Overview

The method described below was designed to address several needs associated with existing CTA approaches and create a structured framework that integrates cognitive and psychomotor tasks into a designer-accessible cognitive decomposition process (Fig. 1). Several CTA methods were reviewed to help identify gaps in existing methods. Wei and Salvendy (2004) classified these methods into four “families” based on the type of data sources used to generate a list of tasks and analytical depth [1] (Table 1). Families 1 and 2 require data collection through interviews, observations, or real-time process tracing, such as think-alouds, to capture the product user's cognition in real time. Families 3 and 4 represent this information in structured and formal models. Family 3 uses knowledge representations, including taxonomies and information-processing (IP) models, to systematically describe cognitive demands. Family 3 taxonomies provide a discrete list of tasks, and IP models provide the sequencing of information for cognitive and psychomotor tasks. Family 4 uses information from all the families listed above in formal or computational systems to create product-use simulations. DECIDE-CTA extends these CTA approaches by embedding cognitive and psychomotor taxonomies (Family 3 taxonomies) within an information-processing framework (Family 3 IP models) and standardized CTA output (Family 4 formal models) for consistent use by designers who lack cognitive expertise.

Fig. 1.

Flow of method.

Table 1.

Overview of CTA families.

	Family 1- Observations and Interviews	Family 2- Process Tracking	Family 3- Conceptual Techniques	Family 4- Formal Models
How is the data collected from participants?	Observing task performance and asking the user questions upon task completion	Watching recordings of task performance, asking the user to state what they are doing and why out loud.	Reviewing documents	Reviewing documents
What level of skill is required to use the Family correctly?	Requires proficiency in performing interviews and/or the ability to clearly explain one's thought process, as well as a base level of domain knowledge	Requires proficiency in performing interviews and/or the ability to clearly explain one's thought process, as well as a base level of domain knowledge	Requires a base level of domain knowledge, the ability to review documents, and create structured conceptual models	Requires a significant level of domain knowledge, ability to review documents, and analytical techniques for model creation
Does data collection interfere with task performance?	Yes	Yes	No	No
Output data level	Qualitative data	Qualitative data	Qualitative data	Quantitative data

The analysis of these families identified several needs that DECIDE-CTA will address. There are four primary needs for CTA: (1) a pre-defined discrete list of cognitive tasks, which is necessary for consistently describing the tasks required to complete cognitive and psychomotor activities, (2) a structure that facilitates the sequencing of the information between cognitive and psychomotor tasks, (3) lack of cognitive expertise on traditional design teams, and (4) high variability in CTA generation. These needs will be described in more detail below.

CTA Methodological Gap 1: Design teams need a discrete list of tasks to reference when creating a CTA to ensure a shared description of what is being performed by product users. Family 3 taxonomies partially address this issue. As noted above, some taxonomies provide a discrete list of tasks linked to specific types of cognitive or psychomotor activities, but these lists are independent of one another. They do not reflect the influence of cognitive task performance on psychomotor outcomes. This lack of connection makes it difficult for designers to ensure they include sufficient cognitive depth in their decomposition, and negatively impacts communication about what is necessary to complete an activity.

• •DECIDE-CTA Goal 1: Select and embed cognitive and psychomotor conceptual models to provide a theoretical foundation (a discrete list of tasks) for decomposing product use.

CTA Methodological Gap 2: The discrete list of tasks requires a structure to guide the order of their performance. These tasks are not performed in isolation; therefore, understanding the temporal flow of information will guide how cognitive and psychomotor tasks influence one another in DECIDE-CTA. Family 3 IP models provide a high-level description of information flow between psychomotor and cognitive tasks, but they do not provide a list of tasks to decompose this flow.

• •DECIDE-CTA Goal 2: Create a sequence or order (parallel and/or sequential) for cognitive and psychomotor tasks grounded in IP models.

CTA Methodological Gap 3: As noted above, existing CTA methods assume cognitive expertise that is often lacking in traditional design teams. Family 3 Taxonomies provide lists of cognitive and psychomotor tasks that designers can use when generating a CTA, but designers need guidance on connecting these taxonomies to their HMI knowledge.

• •DECIDE-CTA Goal 3: Embed principles that designers understand into the translation of product design characteristics into CTA generation to make this process more designer-friendly.

CTA Methodological Gap 4: Lastly, there is high variability in CTA generation regardless of the user's level of cognitive expertise [21,22]. Current CTA methods do not guide how to describe the tasks within a CTA or how to create a CTA consistently. Family 4 is used to create product-use simulations based on CTAs using information from Family 3 IP models to guide the simulation flow, but only provides high-level guidance at most for CTA generation (e.g., use if-then statements) and once again assumes the user has the knowledge to describe the decomposed tasks in detail (gap 1: a discrete list of tasks). This can result in important information being omitted or inaccurate task descriptions within a CTA.

• •DECIDE-CTA Goal 4: Define sentence syntax and CTA templates to make the generation of CTAs more consistent.

Phase 1: conceptual model selection

Phase 1 will discuss selecting and embedding cognitive and psychomotor conceptual models to provide a discrete list of domain-specific tasks for DECIDE-CTA (addressing CTA Gap 1: discrete list of tasks). Providing a task list for designers to use will standardize CTA creation and help communicate which processes are required when using a product. Task Analysis methods require a list of tasks to decompose an activity into subtasks. These conceptual models will provide that structure, but they are also domain-specific, meaning they only list cognitive or psychomotor tasks. Embedding these conceptual models within one another will show how information in one model influences the other. For example, decomposing a physical activity such as maneuvering a vehicle based only on psychomotor tasks (e.g., holding the steering wheel, turning the steering wheel, etc.) leaves out the vital cognitive processes that help ensure that the physical activity is being performed correctly (e.g., recalling how the turning of the steering wheel affects the direction that the vehicle will move). The list of cognitive tasks will be selected based on a review of the Family 3 Taxonomies.

Family 3 knowledge representations: taxonomies

Cognitive Taxonomy Review: Cognitive taxonomies in the field of education were assessed for their lay-facing use (students), providing a foundation to inform CTA designer accessibility in DECIDE-CTA. These taxonomies are primarily used to communicate the assignment goals to students and to structure assessments and learning outcomes. In this research, the taxonomies provide structured systems for classifying the types of mental tasks required during product interaction. They classify the nature and complexity of thinking, offering designers a consistent framework for labeling tasks by cognitive demand, rather than modeling the temporal flow of cognition. By classifying these distinctions, taxonomies define whether a task requires simple recall or complex problem-solving.

Several cognitive taxonomies were reviewed for potential use in developing this framework. Taxonomies such as Bloom’s Taxonomy [2], Upton’s “Design for Thinking” [23], Gagne’s Analysis of Instructional Objectives [24], and Guilford’s Structure of Intellect (SOI) [25] all deconstruct knowledge and processing into various levels, ranging from remembering simple facts to generating original ideas. The number of levels provided by each taxonomy also varies in the granularity with which cognition is described. Upton’s “Design for Thinking” provides three levels (sensory input, affective (emotions), and logical (cognitive skills)), Bloom’s Taxonomy provides six hierarchical levels (knowledge, comprehension, application, analysis, synthesis, and evaluation), Gagne’s provides seven levels (response differentiation, associations, identification, sequences, classification, principles/rules, and strategies), and Guilford’s SOI provides 120 levels. In addition to the levels describing cognitive tasks at a high level, a taxonomy might provide verbs to describe the types of tasks that belong to a given level. Bloom’s taxonomy provides a list of verbs for each of the six taxonomy levels (e.g., translation, interpretation, and extrapolation are verbs used to describe tasks at the comprehension level). Although each cognitive taxonomy differs in scope and granularity, they share a common goal: providing a structured way to describe human thought during task performance. Due to the limitations of the above taxonomies, Bloom’s Taxonomy was selected for use in the development of DECIDE-CTA because it provides a more consistent structure, as described below.

Original Version of Bloom’s Taxonomy: In contrast to Guilford and Gagne’s Taxonomies, Bloom’s Taxonomy offers a balanced structure for representing cognitive complexity along with consistent labeling for cognitive tasks (verbs) and thus was selected for use in DECIDE-CTA. Its balanced structure stems from an easy-to-follow hierarchy that shows how the cognitive tasks transition in complexity; tasks higher up in the structure are more complex than those beneath it. This is not the case for the other taxonomies reviewed. Guilford presents its 120 levels as a 3-dimensional graph with each axis representing an aspect of learning. This 3D graph is difficult to apply consistently between designers. As mentioned above, Upton’s and Gagné’s both provide levels but do not provide a list of tasks associated with them, making it difficult to apply them consistently across individuals.

Prior Adaptation of Bloom’s Taxonomy: The original form of Bloom’s Taxonomy includes the following levels: Knowledge, Comprehension, Application, Analysis, Synthesis, and Evaluation [2]. Three main changes have been made to the taxonomy over the years to improve usability: levels were renamed, levels were reordered, and a new list of verbs was created. First, all the taxonomy levels were renamed by Anderson and Krathwohl [26] to be in verb form to align with cognition as an active thinking process [27,28]. Second, the order of the last two taxonomy levels, originally named Evaluate and Synthesis, was switched by Anderson and Krathwohl due to creative thinking being more complex than decision-making [27].

Lastly, the list of verbs to describe tasks associated with each taxonomy level was updated by Stanny [29]. The original list provided by Bloom’s vacillates between being a singular verb and being a short description of what to do at a given taxonomy level [2]. For example, associated with the original Knowledge level are short descriptions of activities such as knowledge of criteria and knowledge of methodology. Yet for the original Comprehension level, verbs such as translation and extrapolation are used to describe the types of activities completed at this level. This mix of verbs and descriptions makes it difficult to describe what specifically is performed at each level. The prior adaptation of Bloom’s Taxonomy provides a list consisting only of verbs for each taxonomy level, and over the years, other authors using this taxonomy have associated additional verbs with these levels. A literature review by Stanny [29] was performed to assess what verbs are used when describing activities associated with each level of the taxonomy. In this review, thirty websites that discuss Bloom’s Taxonomy and its verbs were assessed. The lists from each site were combined under each level, resulting in the review of 433 verbs. Verb-level agreement was defined as 33.3% inter-list agreement on the level to which a verb belonged, and was used to consolidate the list to 104 unique verbs.

Current Adaptations to Bloom’s Taxonomy: The three prior modifications made to Bloom’s Taxonomy were used as the baseline model for the current development of DECIDE-CTA. Additionally, given this method's focus, the Apply level was excluded. This level involves using knowledge to complete a task, specifically, product use in this case. The Apply level, when it comes to product-use behaviors, can be better assessed through psychomotor frameworks, as the user's application of information is observable. To understand why the user performed the actions they did, they need to be asked questions at other levels of Bloom’s taxonomy (e.g., about their understanding of the instructions). The remaining levels capture the aspects of cognition central to understanding reasoning in product interactions. The revised structure of Bloom’s Taxonomy, with label names and definitions, along with verbs from Stanny [29], can be found in Table 2.

Table 2.

Adapted version of Bloom's taxonomy and verbs.

Reduced List of Taxonomy Levels(adapted for use in current research)	Anderson and Krathwohl’s Definitions	Example of verbs from Stanny’s Review
Remember	Retrieve relevant information from long-term memory	recognize, identify, recall
Understand	Construct meaning from instructional messages	infer, compare, classify
Analyze	Break material into its constituent parts and determine how the parts relate to one another and to an overall structure or purpose	examine, diagnose, deconstruct
Evaluate	Make judgments based on criteria and standards	decide, judge, determine
Create	Put elements together to form a coherent or functional whole	generate, adapt, develop

Bloom’s Taxonomy will be referred to in the following way from now on:

• •Original Version of Bloom’s Taxonomy- refers to the original version of the taxonomy created by Bloom et al.
• •Prior Adaptation of Bloom’s Taxonomy- refers to the change made by Anderson, Krathwohl, and Stanny
• •Current Adaptation of Bloom’s Taxonomy- refers to the changes made to the prior adaptation for purposes of DECIDE-CTA

Cognitive taxonomies alone offer conceptual tools for identifying the type of thinking that occurs, but not how that thinking is expressed through interaction. To address this issue, these representations must be used alongside psychomotor taxonomies that define perception and physical movement. The next section reviews these psychomotor frameworks, which provide the foundation for linking cognitive reasoning to physical activities.

The names and definitions of the taxonomy levels correspond to the updated names provided by Anderson and Krathwohl [26]. The verbs correspond to the updated list created by Stanny [29]. The Apply level has been omitted to improve methodological focus, generating a reduced list of taxonomy levels.

Psychomotor Taxonomy Review: As with cognitive taxonomies, all the psychomotor taxonomies reviewed originated in the field of education. This once again provides a lay-facing list of tasks for designers to use when creating their CTAs when using the DECIDE-CTA method. Psychomotor taxonomies classify tasks by the types and complexity of the movement involved in performance. These frameworks also account for perceptual abilities that support motor control, such as sensory discrimination and how this information cues movement.

Several psychomotor taxonomies were reviewed using the same criteria used to assess the cognitive taxonomies. The number of levels varies between taxonomies, but the variation is not as great as that observed among cognitive taxonomies. Kibler, Baker, and Miles’ Taxonomy [30] only deconstructs psychomotor tasks into four levels (gross bodily movements, finely coordinated movements, non-verbal communication behavior, speech behavior). Dave’s taxonomy [31] (imitation, manipulation, precision, articulation, naturalization) and Hauenstein’s Taxonomies [32] (perception, imitation, manipulation, performance, and perfection) offer a five-level structure. Harrow’s [3] (reflex, basic-fundamental movement, perceptual, physical abilities, skilled movement, and non-discursive communication) provides six levels. Simpson’s Taxonomy provides seven levels (perception, set, guided response, mechanism, complex overt response, adaptation, origination) [33]. As with cognitive taxonomies, a psychomotor taxonomy can also provide verbs to show the types of tasks associated with a given level. Both Harrow and Simpson provide descriptive verbs (e.g., “manipulate,” “perform,” “adapt”) to describe each level, making them particularly useful for listing tasks in action-oriented language that link perceptual and physical processes. As with Bloom’s Taxonomy, Harrow’s Taxonomy was selected for use in the creation of DECIDE-CTA due to having a consistent structure to build upon.

Original Version of Harrow’s Taxonomy: Harrow’s Taxonomy was selected for its clarity in distinguishing between different types of movement activities and its inclusion of a separate perceptual level. The original version of Harrow’s Taxonomy consists of six levels: reflex movements, fundamental movements, perceptual abilities, physical abilities, skilled movements, and nondiscursive communication (e.g., communicating through body language, gestures) [3]. Harrow also provides a short list of verbs and descriptions of activities that can be performed at each level.

Adaptations to Harrow’s Taxonomy for this Research: As with the current adaptations of Bloom’s Taxonomy, some levels are being omitted for methodological consistency. The Reflex level was removed since these types of tasks occur without conscious cognitive input [3]. The other level to be removed was Non-Discursive Communication. This style of movement is intended for human-to-human communication [3] and involves advanced psychomotor skills that are outside the scope of DECIDE-CTA (a standardized method for designers to use when creating a CTA for human-machine interactions); therefore, it will not be integrated into the discrete task list at this time. The names of the remaining levels, a description of them, and a selection of verbs to describe them can be found in Table 3.

Table 3.

Adapted version of Harrow's Taxonomy.

Reduced List of Taxonomy Levels (adapted for use in current research)	Harrow’s Definitions	Example of Harrow’s Verbs
Basic-Fundamental Movements	Movement patterns that occur in the learner during their first year of life	walking, pushing, pulling
Perceptual	Modalities where stimuli are carried to the brain for interpretation	seeing, hearing, feeling
Physical Abilities	Movements that require strength, endurance, flexibility, or dexterity to complete	endurance, strength, effort
Skilled Movements	The result of the acquisition of a degree of efficiency when performing a complex movement	typing, fine manipulation

The names of the levels, definitions, and verbs are all from the original version of Harrow’s Taxonomy [3]. Changes made to Harrow’s Taxonomy include the removal of two original levels (Reflex, Non-Discursive) for this research.

Harrow’s Taxonomy will be referred to in the following way from now on:

• •Original Version of Harrow’s Taxonomy- refers to the original version of the taxonomy created by Harrow
• •Current Adaptation of Harrow’s Taxonomy- refers to adaptations made to the original taxonomy for this research

Taxonomy Selection Limitations and Outcome: Selecting Bloom’s and Harrow’s Taxonomies provides lists of discrete cognitive and psychomotor tasks for use in the creation of DECIDE-CTA; however, these taxonomies have several limitations. First, Bloom’s Taxonomy simplifies complex human interactions into discrete categories. While Bloom’s Taxonomy is often represented hierarchically, this reflects increasing cognitive complexity rather than a sequence of cognitive processes. Nonetheless, this hierarchical structure provides an easy-to-follow view of cognition that is more approachable by those lacking cognitive expertise. This taxonomy was originally created to help teachers communicate classroom goals with students. This accessibility makes it well-suited for adaptation for those lacking a cognitive background. More complex models offer higher theoretical fidelity, but require expertise to utilize.

Both Bloom’s and Harrow’s Taxonomies are impacted by the context of the task being performed and the culture in which they were created. For example, Bloom’s Taxonomy has been applied to many domains outside of education, such as human factors and system monitoring [12]. The context may also be specific to each individual. Studies over the years have found that novices approach problem-solving differently than experts [[34], [35], [36]]. Individuals also get better at performing physical tasks over time, which can make it difficult to distinguish between a basic-fundamental movement and a skilled movement [3]. Each of these taxonomies was also created by educators in the US to help teachers design assignments. Western views of knowledge are not guaranteed to align with the views around the world [37]. Nonetheless, Bloom’s and Harrow’s Taxonomies serve as structured descriptions of cognition by providing names for different types of cognitive and psychomotor tasks, allowing designers to use these terms as they see fit within their culture.

Another limitation of these taxonomies is that each focuses on its respective task types independently of the others. Cognitive and psychomotor tasks influence one another; the lists must be embedded to reflect this integration. Unfortunately, no existing approaches demonstrate this embedding, so a subject-matter expert review was conducted. This review provided the guidance needed to embed Bloom’s Taxonomy within Harrow’s Taxonomy.

Subject matter expert (SME) review

DECIDE-CTA must provide a cognitive decomposition process for designers that accounts for the interaction of cognitive and psychomotor tasks. Embedding the taxonomies provides a consistent structure that shows which cognitive tasks are required to complete a psychomotor task. To facilitate embedding the current adaptations of Bloom’s Taxonomy within Harrow’s Taxonomy, it was necessary to examine how SMEs have described the embedding of cognitive tasks within psychomotor tasks. The results of this review informed the development of the embedding logic. This ensured that the resulting model reflects empirically grounded relationships between cognitive and psychomotor tasks rather than purely theoretical associations, enabling a traceable representation of HMI.

Scope and Screening: This review focused on scholarly work describing how cognitive and psychomotor activities interact during task performance. The review encompassed disciplines including neuroscience, kinesiology, human factors, psychology, linguistics, auditory research, and philosophy. Sources were eligible for inclusion if they explicitly described the link between perception, cognition, and movement during task performance, and were cited by ten or more sources (academic papers, books, or highly referenced book chapters). Google Scholar and ScienceDirect were the primary search engines used. The search terms used to find sources discussing Bloom’s Taxonomy were the verbs provided by Stanny [29], and Harrow’s Taxonomy were verbs provided by Harrow [3].

Data Extraction and Coding: Each record was coded according to three criteria:

• 1.The levels of the current adaptations of Harrow’s Taxonomy referenced (e.g., Perceptual, Physical, Skilled Movements) based on the use of verbs and context clues describing the taxonomy level
• 2.The levels of the current adaptations of Bloom’s Taxonomy referenced (Remember, Understand, Analyze, Evaluate, Create) based on the use of verbs and context clues describing the taxonomy level
• 3.Verbatim text excerpts supporting these designations.

This coding scheme enabled the systematic mapping of how different domains conceptualize the interaction among perception, cognition, and movement through a hybrid deductive-inductive thematic analysis.

Deductive Thematic Analysis: This type of analysis involves evaluating information with a predetermined theme in mind. In the case of the SME review, the predetermined themes were the Current Adaptations to Bloom’s and Harrow’s Taxonomies, and the analysis followed a procedure similar to those described by Fereday and Muir-Cochrane [38] and Braun and Clarke [39]. Deductive coding was performed by first reviewing each source and classifying them according to the highest cognitive level necessary to complete the described psychomotor task. Coding was achieved by comparing the use of verbs in the SMEs' descriptions and context clues, in the case of synonyms, against the verbs used to describe the current adaptations of Bloom’s and Harrow’s Taxonomies (provided by Stanny and Harrow, respectively). For instance, an article by an SME would be reviewed to determine if it described movements using the same verbs provided by Stanny and Harrow, or if it mapped onto a specific level of the current adaptation of Harrow’s Taxonomy (criterion 1). This iterative coding process was refined as additional articles were reviewed. The coding classifications were re-evaluated until stable themes were evident. Next, the article was assessed to determine how it described the use of cognitive tasks to aid movement (criterion 2).

Inductive Thematic Analysis: The inductive coding process was performed after the deductive classification. Inductive thematic analysis involves identifying patterns in the data to capture connections among data points and creating themes based on the materials [39,40]. Direct quotes were pulled from the article to support the embedding of the current adaptations of Bloom’s and Harrow’s Taxonomies (criterion 3). Repeating terminology and synonyms used to describe cognitive and psychomotor task completion were noted. For instance, terms such as "movement" and "top-down" appeared in several SME articles about the performance of physical activities. Synonymous terms were grouped according to the list of verbs used to describe the various levels within Bloom’s and Harrow’s Taxonomies. This process helped identify recurring patterns in how SMEs describe cognitive task performance in reference to physical tasks and informed the development of the cognitive embedding [39,41].

Coding Challenges: Several challenges emerged during the coding process. SMEs did not describe the boundaries between the different types of movements described within Harrow’s Taxonomy. The authors described physical movements in vague terms, such as "movement" or "motor-cognition". To overcome this issue, context clues were utilized to ensure SMEs were describing the same underlying psychomotor task. Another challenge was determining the correct cognitive activity level that was being described. As mentioned above, some verbs used to describe tasks were associated with multiple levels of Bloom’s Taxonomy. This meant that context clues had to be relied upon to determine the appropriate taxonomy level being discussed.

Evidence for Level Merging and Simplification in the Current Adaptation of Bloom’s Taxonomy: A separate analysis from the SME review was performed to assess any additional changes, such as merging levels of Bloom’s Taxonomy to aid in its use within DECIDE-CTA. The hierarchical nature of the taxonomy is meant to separate cognitive levels by increasing complexity, but it is not always clear when one level ends and another begins. The prior adaptations of Bloom’s Taxonomy provided verbs to describe the activities that take place at the different levels, and additional words have been added to that list by other authors over the years. Still, a literature review by Stanny revealed that many verbs are used across multiple levels. For instance, identify was in lists for Remember, Understand, and the Apply level, and compare was in lists for Understand, Analyze, and Evaluate [29]. As described above, Stanny consolidated the list to improve communication at the taxonomy-level descriptions. However, verb agreement alone is not enough to describe a taxonomy level. This overlap in verbs made it difficult to distinguish the performance of one level from that of another. In addition, prior research that conducted causal analyses to determine non-hierarchical links among taxonomy levels found that the Understand and Analyze levels were functionally interdependent [42]. This meant that these levels cannot be performed separately. This relationship led to the two levels being merged into one to reduce confusion when distinguishing between them when using DECIDE-CTA. This merging includes using lists of verbs from either level to describe an Understand and Analyze task. The remaining levels of the current adaptation of Bloom’s Taxonomy are as follows: Remember, Understand, Evaluate, and Create (Fig. 2).

Fig. 2.

Current adaptation of Bloom's taxonomy. (a) shows the separate understand and analyze levels. (b) Shows the merged understand level.

SME Review: The SME review resulted in 22 sources across 10 disciplinary fields. The majority of publications (n = 13, 59%) were peer-reviewed journal articles, and the rest were from books (n = 9, 41%). Neuroscience and related biological sciences were the most represented fields (n = 12, 55%), followed by kinesiology (n = 2, 9%), health (n = 1, 5%), human factors (n = 1, 5%), marketing (n = 1, 5%), otorhinolaryngology (n = 1, 5%), philosophy (n = 1, 5%), psychology (n = 1, 5%), and social psychology (n = 1, 5%). Deductively, each article was sorted according to the Current Adaptation of Bloom’s and Harrow’s Taxonomies to understand how SMEs describe cognitive and psychomotor processes. Inductively, recurring descriptions were compared across sources to identify thematic saturation.

Eleven out of the twenty-two sources examined perception as an active information-processing function that enables humans to interpret, predict, and regulate their environment. The list of articles referenced can be found in Appendix Table 1, where column one lists the citation, column two lists the necessary level(s) of the current adaptation of Bloom’s Taxonomy, and column three lists a quote from the paper connecting to those taxonomy levels. Eleven out of the twenty-two reviewed sources described motor control as a cognitive-psychomotor system that integrates planning and feedback during movement execution (Appendix Table 2).

Across the reviewed literature, two themes are of note. First, the Create level of the Current Adaptation of Bloom’s Taxonomy was considered to be necessary for the completion of physical and perceptual activities. These themes were identified across descriptions of completing both perceptual and physical activities, showing that higher-order cognition is not restricted to cognitive activities only. Higher-order cognition is necessary for completing novel perceptual and physical activities. Second, there was no distinction in classifying physical activities based on skill required. While Harrow’s Taxonomy describes physical activities across three main skill types (Basic fundamental, Physical, and Skilled Movements), SMEs describe it as a continuum of movement execution supported by cognitive activities.

The list of articles referenced can be found in Appendix Table 2, where column one lists the citation, column two lists the necessary level(s) of the current adaptation of Harrow’s Taxonomy, and column three lists a quote from the paper connecting to those taxonomy levels.

Evidence for Level Merging in the Current Adaptation of Harrow’s Taxonomy: The dataset analysis also revealed a lack of distinction in SMEs' descriptions of physical activities. The current adaptation of Harrow’s Basic-Fundamental Movement, Physical, and Skilled Movements levels were absent across all sources, with authors describing physical actions as “movement,” “action,” or “motor control.” The lack of distinction among these three levels in the current adaptation of Harrow’s Taxonomy resulted in their collapse into a single level named Physical for use within DECIDE-CTA to help reduce possible confusion by designers who struggle to determine the difference between these three levels (Fig. 3). This leaves Perceptual and Physical as the remaining levels of the current adaptation of Harrow’s Taxonomy.

Fig. 3.

Merging levels in the current adaptation of Harrow's Taxonomy. (a) Shows the initial remaining levels of Harrow's Taxonomy. (b) Shows the merged levels based on SME feedback.

1. Evidence of Cognitive-Perceptual Embedding: The embedding of cognitive tasks within perceptual task performance was guided by the results of the hybrid deductive-inductive thematic analysis of the SME literature. A coding convergence criterion was established whereby a cognitive-perceptual relationship was documented if it was identified by at least half of the SMEs (n ≥ 11 of 22 sources). This threshold helped to ensure that embedded relationships reflected recurring themes across disciplines rather than isolated observations. All cognitive tasks mentioned by an SME were documented (e.g., they stated that Evaluate and Create are necessary to complete perceptual tasks). When discussing perception in general, the literature consistently describes this as a multi-stage process that integrates signal detection with higher-order cognitive tasks. Roscoe and Davis defined perception in three stages (detection, comparison, and recognition) [43], starting with sensing the stimuli (Harrow’s Perceptual level), moving to memory retrieval of prior experiences with that stimulus (Bloom’s Remember level), and ending with interpreting what the stimuli mean (Bloom’s Analyze and Evaluate levels). Similarly, Furey [44], Toppino and Long [45], and [46] emphasize the use of bottom-up and top-down processing of stimuli. Bottom-up mechanisms focus on the initial detection of perceptual information (Harrow’s Perceptual level). In contrast, top-down mechanisms draw on cognition to process stimuli (the Analyze, Evaluate, and Create levels of Bloom’s). While not listing stages specifically, Pinker, Capo-Aponte et al., Kraus, and Eggermont and Ponton, also agree that perception is linked to the completion of the aforementioned higher-order cognitive tasks [[47], [48], [49], [50]].

For discrete cognitive tasks required to complete perceptual activities, half of the sources (n = 11) stated that the Analyze, Evaluate, and Create levels are used. Only one paper explicitly referenced the Remember level, noting that a stimulus must be identified as requiring attention. While the SMEs did not mention the Understand level, its inclusion remains necessary to maintain the flow of information within the taxonomy. Understanding the meaning of a sensory cue (e.g., what color a light changes to) is a prerequisite for the Analyze, Evaluate, and Create levels. This interpretation is consistent with the hierarchical structure of the prior adaptations of Bloom’s Taxonomy. These foundational levels are necessary for completing the higher levels of the taxonomy. While the Create level is necessary for Perceptual level tasks, it will be intentionally excluded from the cognitive decomposition of the Perceptual level since generative cognitive processing occurs automatically at this level and cannot be consciously evaluated [51] (Fig. 4).

Fig. 4.

Embedding of the current adaptation of Bloom's taxonomy into the perceptual level of Harrow's Taxonomy. (a) Embedding according to SME review. (b) Refinement of the embedding through conscious cognitive effort.

2. Evidence of Cognitive-Physical Embedding: Mulder et al. (2003) characterized Create as the motivation and goals to achieve movement [52]. Magill and Anderson (2014) and McLaughlin et al. (2010) described closed-loop and feedforward control systems that compare sensory input against stored motor schemas to guide movement [53,54]. Leisman et al. (2016) introduced the concept of motor cognition, emphasizing mental processes such as prediction and anticipation that guide self-directed actions [55]. In a similar vein, Haggard (2005), Jeannerod (2006), and Roscoe and Davis (2004) discuss anticipatory control, linking goals and planning to the performance of voluntary movement plans [43,56,57]. This reflects the current adaptation of Bloom’s Create level, in which a person makes a plan and adapts its execution. For cognitive tasks utilized to complete psychomotor activities, all but one source indicated that the Create level of the current adaptation of Bloom’s Taxonomy is the highest level necessary, with one source indicating that the Evaluate level is the highest [43]. While SMEs did not mention the other levels, they will still be included in the structural cognitive embedding. Bloom's work indicates that the lower levels of the taxonomy are necessary for achieving higher levels [2] (Fig. 5).

Fig. 5.

Embedding of the current adaptation of Bloom's Taxonomy in the physical level of Harrow's Taxonomy.

Complementary perspectives from neurophysiology highlight the need for the Perceptual level in Harrow’s Taxonomy and the Create level in Bloom’s Taxonomy to complete physical activities. Magill and Anderson (2014) discuss how open-loop (pre-planned instructions) and closed-loop (perceptual feedback) systems work together to guide the completion of movement [53]. McLaughlin et al. (2010) and Leisman et al. (2016) describe the perception-action cycle, in which movement intentions are constantly updated by sensory feedback [54,55]. Lastly, studies confirm that anticipating, simulating, and interpreting actions require the same cognitive effort to generate them, whether the user makes them or watches the performance of an action [[58], [59], [60], [61]].

SME Review Outcome: The above embeddings of Bloom’s and Harrow’s Taxonomies provide a reliable foundation for designers to use when describing cognitive and psychomotor tasks. The SME review supported several taxonomy refinements: (1) consolidating the current levels of Harrow’s Taxonomy into a single Physical level, (2) merging the Understand and Analyze levels of Bloom’s Taxonomy, and (3) outlining what levels are necessary for the completion of psychomotor activities. While the SME Review embedding provided a clear cognitive decomposition of perceptual and physical tasks, it still lacks guidance on the sequencing of information, which is necessary to structure DECIDE-CTA. The embedded task list was integrated with IP models to illustrate the temporal flow of information among perceptual, cognitive, and physical activities.

Phase 2: cognitive-psychomotor unified model

Phase 2 describes how the Family 3 IP models and the Family 4 models were combined with the cognitive-psychomotor embedding from Phase 1 to create a unified model for use in DECIDE-CTA (addressing CTA Gap 2: lack of sequencing structure). Phase 1 provides a discrete list of tasks to decompose cognitive and psychomotor activities, but does not provide guidance on embedding them within a temporal flow. Family 3 IP models provide a high-level overview of how information flows between cognitive and psychomotor activities. Likewise, Family 4 models provide a high-level overview of how to use IP models, along with a discrete list of tasks, to generate a model of human behavior. This resulted in combining Families 3 and 4 to provide a discrete list of tasks for use in a temporal flow to describe task completion. The review of models associated with Family 3 IP Models and Family 4 models will be discussed in the following subsections.

Family 3 knowledge representations: IP models

The structure of IP models provides a basis for the flow of information within DECIDE-CTA. IP models were reviewed to guide the flow of information between discrete cognitive and psychomotor tasks as defined in the embeddings of Bloom’s and Harrow’s Taxonomies. IP models represent cognition and psychomotor tasks as a sequence of functional stages through which information flows during task performance. These models do not provide a decomposition of the types of cognitive tasks that occur at a given stage, leaving it to designers to generate a list of tasks to describe what is taking place in more detail. Unlike taxonomies, which classify the cognitive and psychomotor activities, IP models describe the progression of information over time [62].

Several IP models were reviewed for DECIDE-CTA to determine whether there are major differences in how IP is discussed across models. These models were found to have similar structures and processes for describing the sequencing of information between cognitive and psychomotor tasks. Welford’s [63] and Endsley’s Situation Awareness [64] models portray human performance as a cyclical loop of perception, mental processing, and movement response, highlighting the importance of feedback in shaping ongoing activity. Wickens’ Multiple Resource Theory (MRT) also uses a cyclical loop, but also includes parallel processing channels, such as visual versus auditory inputs, that can operate independently or interfere with one another (e.g., users may fail two tasks if both rely on visual input) [65]. Card, Moran, and Newell’s Model Human Processor (MHP) describes perceptual, cognitive, and motor processors in a cyclical loop, but specifies capacities and processing times, enabling predictions of workload and task duration [66]. All of these IP models describe the temporal flow and interaction of information as follows: perceive something from the environment, perform a series of cognitive activities on the sensory cue, and use that information to guide a physical response. The flow reveals where breakdowns or interference may occur. These structures can be used to develop a set of rules for designers to follow when creating a CTA.

Family 4 models

Family 4 models build upon the information provided by Family 3 methods and were reviewed to provide high-level guidance on structuring output for DECIDE-CTA. Family 4 models extend CTA beyond descriptive representations by transforming IP models in Family 3 into formalized structures that can be applied to predict performance. In practice, most Family 4 methods follow a common analytical sequence that builds directly upon the conceptual work of Family 3. Designers are instructed to create a list of relevant tasks but are not provided with a predefined task list, as in cognitive or psychomotor taxonomies. This list of tasks is then sequenced to capture execution order, decision points, and feedback loops, forming a temporal flow of information. For instance, consider the following step: “Press the power button and wait for the status light to turn green.” A designer using a Family 4 model first needs to determine the underlying cognitive tasks (e.g., recognizing the power button, monitoring the status indicator). These tasks need to be organized in a specific order before they can be used in a predictive model. However, Family 4 does not provide a list of cognitive or psychomotor tasks for designers to use; designers must develop their own list of tasks. The designer-derived list is then integrated into the formal logic or computational architecture of the chosen model, enabling it to simulate task performance. Family 4 methods assume the designer knows how to generate an appropriate list of tasks. These methods also assume that designers can describe tasks without a predefined syntax, ensuring the description targets the desired task. This assumption may result in inaccurate predictions because task lists can vary significantly. Since this workflow assumes that the decomposition and syntax have already been created elsewhere, much of the responsibility for ensuring validity, completeness, and consistency lies with the analyst rather than the model itself.

Family 4 models use the aforementioned analyst-designed list of tasks and their descriptions to generate various types of simulated output, ranging from task timing calculations to predictive performance rules. The GOMS model [66] quantifies task performance by mapping user goals to the specific actions and decision rules required to achieve them, enabling predictions of task time and efficiency. CWA [67] models how users adapt their behavior to system constraints, providing a structured means of analyzing work domains, control strategies, and decision contexts in complex environments. Cognitive architectures such as Adaptive Control of Thought-Rational (ACT-R) [68] and State, Operator, and Result (SOAR) [69] simulate the internal mechanisms of human cognition, such as memory retrieval, learning, and motor execution, through computational modules and production rules. Continuing with the “Press power button […]” example, GOMS requires the designer to break down this step into goals (turning on the device), operators (e.g., visual search for status indicator), methods (list of tasks needed to complete the goal), and selection rules (determining if the goal was completed). Predictions of goal completion in GOMS rely heavily on task completion times for the tasks identified by the designer.

Collectively, although these Family 4 models differ in focus, they share a common foundation: each relies on the IP models in Family 3 to define the flow of information that enables modeling. Yet their reliance on designers to create and describe the task list, without guidance, limits standardization. This means that while Family 4 models advance predictive rigor, they do not eliminate the subjectivity of task decomposition, making consistency across analysts and contexts a persistent challenge [21]. Returning to the previous example, one designer might describe the instructions as having a single goal (turn on the device). In contrast, another designer might describe them in terms of multiple cognitive and psychomotor tasks (e.g., locate the power button, interpret the indicator status, verify that the device is on). This would result in two different predictive models of user performance. Therefore, Family 3 Taxonomies and IP models need to be combined with Family 4 models to provide a discrete list of tasks with a temporal flow structure that guides the creation of CTAs and improves predictive performance.

Combining Families 3 and 4: The adapted Family 3 taxonomies from the previous section were integrated into the IP model structures for Families 3 and 4. As discussed above, Family 3 IP models provide a standardized temporal flow of information between tasks in a list. In contrast, Family 4 provides a structure for using task lists to add granularity and aid in CTA generation. Across models such as Welford’s IP model, Wickens’ Multiple Resource Theory, and GOMS [43,65,66], a consistent structure emerges: perceptual information is received through sensory channels, evaluated by cognitive activities, and then used to guide a physical response. Each model emphasizes this sequential flow (perception → cognition → action) as the foundation for how information is processed and actions formed. This shared representational logic informed the development of the sequencing for the structural cognitive embedding.

Based on this sequential flow of information, the structural cognitive embedding was organized into three corresponding blocks: Perceptual Standalone (PS), Perceptual Embedded in Physical (PEP), and Physical (PHY). Each block contained predefined tasks derived from the current adaptations of Bloom’s and Harrow’s Taxonomies. The three-block structure formalizes the flow of information described by these IP models, defining how perceptual cues feed into cognitive evaluation and, in turn, direct motor execution. Embedding validated taxonomies within this structured format ensures consistency while maintaining accessibility across designers who use DECIDE-CTA.

Perceptual Standalone (PS) Block: The Perceptual (PS) Block represents passive perception of sensory cues, without the intent of performing or in service of a physical action, meaning the perception occurs in a “standalone” IP environment (Fig. 6). IP models all emphasize that task performance begins with the detection and interpretation of a sensory cue and ends with sensory confirmation of task completion. This cyclical flow provided the structural foundation for the PS block.

Fig. 6.

Perceptual standalone block.

The structure of the PS block also draws directly from the adapted taxonomies identified through the SME-informed review. Within the PS block, four structured components define the sequence of perceptual tasks:

• 1.Sensory Task Goal – The instructions for use (IFU) specify the sensory task goal. It is the cue the user is attempting to detect (e.g., locating a warning light, listening for an alarm).
• 2.Sensory Activity – the mode of perception used to obtain information (seeing, hearing, or feeling).
• 3.Cognitive Activity – the mental processing applied to interpret sensory cues, encompassing Remember, Understand, and Evaluate.
• 4.Output – the result of the perceptual process, typically an interpreted state or conclusion that informs subsequent action or verifies task completion.

The solid black arrow represents the flow of the task goal. This goal is achieved through a specific sensory activity the user performs, followed by cognitive activities that interpret the sensory information. The output is the completion of interpreting the meaning of the sensory task goal and feedback on the goal if continuous monitoring is needed, represented by the dashed arrow. By structuring perceptual tasks in this manner, the PS block defines the perceptual tasks described in Family 4 IP models while including the categorical rigor of the validated taxonomies. It ensures that perception is represented not as a passive event but as an active, cognitively integrated process. For example, the PS block can be used to monitor a semi-autonomous system, in which the user only needs to observe its output passively. The user first observes feedback from the system (e.g., a notification on the screen) as part of the sensory activity, then 1) they must remember what the notification is trying to convey (Remember), 2) understand what the notification on the screen means about the system (Understand + Analyze), and 3) evaluate the system status based on the notification (Evaluate).

Perceptual Embedded in Physical (PEP) Block: Unlike the PS Block, the Perceptual Embedded in Physical (PEP) Block is always performed in the service of completing a physical task. It guides or regulates ongoing physical movement represented by the Physical (PHY) Block (e.g., applying more pressure to a brake pedal to slow down based on visual feedback) (Fig. 7). Temporally, this block occurs only while a user is engaged in a motor control activity, when sensory feedback is continuously interpreted to adjust or maintain performance. This block cannot be performed in the absence of motor control. This use of discrete tasks and flow of information aligns with findings from Family 4 models, which frame perception and cognition as interdependent regulators of physical activities.

Fig. 7.

PEP and PHY blocks. The user is processing feedback from the environment and their performance (PEP) to help inform how to perform a physical action (PHY).

The PEP block, therefore, serves as a formal mechanism for representing the dynamic relationship among perception, cognition, and motor control. By explicitly defining this intermediate layer, the framework addresses a persistent limitation of traditional CTA approaches: the lack of guidance for consistently describing interactions among cognitive, perceptual, and physical tasks.

Physical (PHY) Block: The PHY block contains tasks where motion is the primary activity. This block uses all the levels of the current adaptation of Bloom’s Taxonomy. Motor control cannot be completed without perceptual input to guide its performance, as shown by the IP models; therefore, the PEP block is included in Fig. 7. Although not shown in this block, the PS block must always be performed first to initiate the PEP and PHY blocks. For example, in the prior semi-autonomous scenario, a system can first beep to indicate that a subsequent action is required (PS block). For example, the user must perceive the system beep via the PS block. Next, the user will perform the PEP and PHY blocks simultaneously. They will use the PEP block to determine when to start and stop the PHY block (e.g., by assessing changes in the system's pressure). Within the PEP block, the user first observes the current pressure in the system, then 1) they must remember what the value on the screen is conveying (Remember), 2) understand what the changing pressure means in relation to interacting with the system (e.g., turning the pressure valve to the left lowers the pressure) (Understand + Analyze), and 3) evaluate the pressure in the system as the pressure valve is turned (Evaluate). The PHY block will adjust the effort used to complete the motor control activity (e.g., how quickly to turn the valve). In the PHY block, the user must 1) remember what the amount of force needed to turn the pressure valve feels like (Remember), 2) understand the amount of force needed to turn the pressure valve (Understand + Analyze), 3) evaluate the amount of force being used to turn the pressure valve (Evaluate), and 4) adjust the amount of force applied to the pressure valve based on feedback from the system (Create).

Embedding the psychomotor dimension within this IP sequence demonstrates how these activities interact during task execution, ensuring that motion is not an isolated behavioral endpoint. This approach provides a repeatable and designer-accessible way to document the completion of physical activities within a cognitively grounded CTA framework.

CTA Families 3 and 4 Integration Outcome: The review of Families 3 and 4 provided a theoretical structure for DECIDE-CTA, grounded in educational representations of cognition and psychomotor tasks, refined by expert insights, and embedded within IP models to illustrate the flow of information. However, this basis is not guaranteed to align with designer perceptions of cognition.

IFUs were reviewed next to bridge designer knowledge, and this cognitive-psychomotor unified model. IFUs are documents created by designers to explain what the user should perceive and how to act upon that information (decision-making or physical response). IFUs externalize the assumptions designers make about the cognitive and psychomotor tasks required to interact with their products. The basic structure of an IFU lends itself well to CTA generation since it provides a high-level decomposition of human-machine interactions. This can serve as a basis for understanding how designers approach cognitive and psychomotor tasks.

Phase 3: IFU review

Phase 3 will discuss the process of grounding DECIDE-CTA in designer knowledge, based on HMI research and standards, and how that information is applied within IFUs (addressing CTA Gap 3: lack of cognitive expertise on traditional design teams). Incorporating HMI principles provides designers with a familiar knowledge base. It can be connected to aspects of CTA generation (e.g., linking the required cognitive tasks to a product's visual feedback). Department of Defense Military Standard 1472H (MIL-STD-1472H) was reviewed to extract explicit design characteristics relevant to DECIDE-CTA (e.g., visual alert characteristics). This standard was selected because it was designed explicitly for HMI integration into product design, provided the most comprehensive list of design characteristics, example recommendations for when to use design characteristics (e.g., no more than two flashing light indicators should be used at a time), and actual values or measurements to use when integrating these characteristics into a product (e.g., the difference between two flashing light rates should be greater than 2 hertz). IFUs were reviewed to gain insight into how designers apply HMI principles and design characteristics when describing user interaction tasks, which subsequently informed the creation of DECIDE-CTA Task Categories. The Task Categories enhance the method's accessibility for designers and connect the cognitive-psychomotor unified model to real-world applications.

HMI principles

Four key HMI principles were identified in the literature and in MIL-STD-1472H as relevant to the creation of DECIDE-CTA: effort, control type (human- or machine-initiated actions), interaction state (initial-priming, active, steady, and final), and timing (duration of task completion). Each of these characteristics captures a distinct feature of product interaction that has been emphasized in both task analysis and human reliability research. Understanding the effect each has on product interaction will guide the integration of these elements into the task categories (discussed later in this section) and into sentence syntax and CTA templates (discussed in Phase 4).

Effort is necessary to complete tasks that require physical exertion [70]. MIL-STD-1472H details various aspects of physical activity that can be considered effort-intensive (e.g., strength, precision) [70]. For example, there is guidance on the level of force that is required to push a vehicle pedal relative to the user’s knee angle [70]. Information such as this can help designers avoid creating products that a person is unable to physically use safely or effectively. In addition, understanding if a task involves physical exertion helps designers determine if completing the PEP and PHY blocks is necessary. For instance, a task that requires no physical effort will require only the PS block, thereby narrowing down for designers which parts of the cognitive-psychomotor unified model are required.

Control type, whether actions are human- or machine-initiated, is a defining feature of system interaction [71]. Research by Parasuraman et al. on human–automation interaction shows that machine-initiated behaviors change the cognitive demands on the user: instead of planning an action, the user must interpret cues, monitor the product's state, and verify task completion. Their research breaks down task automation into 10 levels that affect HMI, with the lowest level utilizing no machine assistance, and the highest level relying on the machine to make every decision and ignore human input. This is necessary for determining which of the three blocks (PS, PEP, and PHY) to use based on the interaction the user is expected to have with the product. Products or steps during product use that require the user to view output only need the PS block, while those that require a physical response also need the PEP and PHY blocks.

Interaction state refers to when and how users respond to system cues. Four states, identified in HMI studies, were used to guide the classification of steps pulled from the IFUs for review: initial priming, active, steady, and final. These states were created to provide a cognitively grounded way of segmenting discrete HMI phases over time. Traditional Task Analysis and CTA methods fail to capture how cognitive demand fluctuates as users transition through different types of interactions (e.g., changing demand between perceiving cues and verifying their meaning). IP models demonstrate that HMI occurs across several cognitive phases. Several studies were reviewed to define each of the above interaction states.

• •The initial-priming state corresponds to the recognition and interpretation of design characteristics, when the user first perceives a product’s starting condition and forms a mental image [64,72]. This interaction state parallels the early stages seen in IP models that emphasize perception and the conversion of the initial cue for use in decision-making [73]. Errors made during the initial-priming state can propagate to later decisions; the accuracy of the information gathered in this state is crucial.
• •The active state represents decision-making while using a product [64,71]. The user is translating their understanding of the current information into actions. During this state, attention and cognitive effort are devoted to choosing, executing, and adjusting responses in real time [64,71]. Use errors can occur during rapid responses required when the user has less time to make a quality decision.
• •The steady state captures periods of sustained observation and monitoring when system variables remain constant [74]. This state is intended only for activities where things should remain unchanged, requiring only monitoring stability and the ability to detect deviations by the user [74]. Failure to detect a change can lead to delayed interventions.
• •The final state represents the verification and closure phase of interaction, when users confirm task completion [63,64,73,75]. Cognitively, this phase aligns with feedback evaluation loops described in IP models [63,73,75], as well as projecting future results [64]. This state ensures that earlier decisions have yielded the correct outcome. Use errors can occur during the final state if the user does not recognize that a task is complete [76].

Timing is another critical factor because it shapes whether a user can adjust performance during execution [77,78]. A study by Hülsdünker and Mierau was reviewed to ground the relationship between visual perception and physical response. In contrast, work by Tchakouté et al. was reviewed to ground the relationship between response time and vibrotactile feedback. For example, pushing a button takes little time to complete. The user would have enough time to determine the button's starting position (PS), press the button (PHY), and examine its final position (PS). There is not enough time for perceptual feedback (PEP) to alter how the button is pressed. Studies of simple visuomotor tasks have shown that the latency from stimulus to motor execution takes place within hundreds of milliseconds [78]. This means there is not enough time for PEP to be able to alter how a button is pressed in a meaningful way once the action has started. IFUs frequently state these temporal expectations by instructing users to act “quickly,” “hold until,” or “continue until.” Therefore, capturing timing provides a way to model whether tasks are discrete and immediate or extended and subject to real-time feedback.

Altogether, these characteristics provide a structured framework for interpreting IFU sentences beyond their surface-level wording. Rather than treating instructions as simple action statements, considering effort, control type, interaction state, and timing allows for linking them to cognitive and reliability-relevant aspects of product use.

HMI design characteristics review

Designers can use a wide range of design characteristics to communicate with their product users. Since these characteristics directly influence how users perceive, interpret, and physically interact with a product, standards have been created to provide guidelines for their use. MIL-STD-1472H was used to ground the design characteristics reviewed [70]. As mentioned above, this standard provided a comprehensive list of design characteristics and was created with HMI in mind. This standard specifies the psychomotor parameters that must be considered to ensure safe HMI.

These categories focus on distinct communication or interaction modalities between the user and the product.

• •Visual cues rely on the use of displays, indicators, and color-coded controls (Color, Shape, Brightness (light only), Contrast, Location, Size, Movement).
• •Auditory cues support performance in multimodal or time-critical contexts through the use of sound modalities (Volume, Duration, Pattern, Comprehensibility, Sound (type, signal, or style))
• •Tactile cues provide direct feedback through physical interaction, such as button resistance or vibrotactile feedback. (Texture, Feedback, Intensity, Depression)
• •Physical characteristics describe human performance constraints, focusing on required effort, strength, posture, or precision. (Response time (latency, reaction time), Sensitivity, Effort, Strength, Posture, Precision, Endurance)

IFU analysis

A review of IFUs was conducted to guide the development of Task Categories for use in DECIDE-CTA, linking current design documentation practices and designers' knowledge of human-machine interaction to the aforementioned cognitive blocks. The review aimed to understand (1) the ways cognitive and psychomotor taxonomies were referenced in the IFUs, (2) the expected forms of user interaction with product characteristics, and (3) the specific design characteristics referenced within IFUs. This step served as a bridge, enabling designers to translate applied knowledge into theoretical insights. IFUs are structured, designer-authored documents that outline the steps users must follow to use a product. Since they represent the designers' externalized expectations of user behavior, IFUs provide a practical foundation for identifying how perceptual, cognitive, and physical activities are described in natural language.

IFU Sample: IFUs from a diverse set of hardware and software products across multiple industries were analyzed. One hundred twenty-one IFUs were randomly selected for this review. This ensured that the sentence syntax was not biased toward a single product type. IFUs were found online through a database called ManualsLib [79]; eligibility required English text instructions (not pictogram-only) to enable sentence parsing. These findings were then mapped onto the structural cognitive embedding to create sentence syntax grounded in cognitive and psychomotor tasks. Of the 121 IFUs that were analyzed, 31 were from highly regulated fields (e.g., healthcare), and 62 were from less regulated fields (e.g., consumer goods), and 28 were from multiple industries, covering a diverse set of products across industries (Table 4). Systematic sampling was used to select IFUs for review based on their availability in the ManualsLib database. An industry field was selected (e.g., healthcare), and then a random number was drawn to indicate which IFU to select. If the IFU met the above criteria, then it was downloaded for additional review. Analyzing IFUs from highly regulated and less regulated fields provided insights into how instruction details varied between the two groups. The sample was evenly divided between hardware-based and integrated hardware-software systems.

Table 4.

Summary of instructions for use (IFUs) reviewed.

		Total n (%)
Product Type	Software	7 (6)
	Hardware	58 (48)
	Both	56 (46)
Industry	Healthcare	11 (9)
	Transportation	18 (15)
	Energy	2 (2)
	Electronics	28 (23)
	Communication	5 (4)
	Other	19 (16)
	Food	10 (8)
	Multiple Industries	28 (23)

IFU characteristics reviewed

The IFU review focused on understanding the expectations designers express in product documentation. IFUs were collected from products across multiple fields to ensure generalizability. A Python script [80] was developed to convert the IFU documents into text files and extract instruction-relevant sentences using a curated list of verbs (Appendix Table 3 and Appendix Table 4), provided by Stanny for Bloom’s and Harrow’s original work for their taxonomy [3,29]. This approach ensured that analysis remained consistent with the current adaptations of the cognitive and psychomotor taxonomies. Sentences were grouped by verb and then randomly sampled for manual assessment. By grounding the review in validated taxonomies, the analysis produced sentence sets that can be embedded in the method consistently and repeatably, reducing reliance on designer interpretation.

After the IFUs were converted to text files, sixty-nine random sentences were pulled and deconstructed for review. Steps in IFUs are commonly written in one of two ways. Simple sentences instruct the user to perform a single task, and compound sentences describe multiple tasks to be completed either sequentially or concurrently. For example, a simple sentence might read, Press the power button, while a compound sentence might state, Hold the button while turning the knob. For DECIDE-CTA, compound sentences were identified and deconstructed into simple sentences. Standardizing the representation in this way improves consistency across the analysis process. This resulted in the classification of 116 deconstructed IFU steps. These sentences were individually reviewed to determine the taxonomy level to which they belonged and to determine if they could be grouped by characteristics described in HMI literature and guidance documents.

Referencing of Cognitive and Psychomotor Taxonomies in IFUs: One hundred and twelve of the 116 sentences were human-initiated. These sentences were found to belong to the Physical level (n = 63) of the current adaptation of Harrow’s Taxonomy, with Evaluate (n = 15), Understand + Analyze (n = 11), Create (n = 9), and Perceptual levels (n = 14) (Table 5). Only human-initiated sentences were sorted by taxonomy levels. This imbalance suggests that IFUs overlook the cognitive processes that often precede physical movement. For example, instructions may specify “Press the power button” but omit the tasks of identifying the correct button or confirming the device’s state before acting. By leaving out the cognitive tasks, IFUs obscure possible causes of use error, a finding that mirrors a limitation identified in traditional task analysis methods [81]. This disparity underscores the need for standardized guidelines on cognitive fidelity, ensuring that IFUs serve not only as user guides but also as reliable inputs for CTAs. These findings highlight a major limitation of current IFU writing practices. The omission of cognitive tasks obscures the reasoning behind user actions and restricts the ability to trace potential error pathways [82]. This disparity shows the need for standardized guidelines for cognitive fidelity, ensuring that IFUs can serve not only as user guides but also as reliable inputs for CTAs.

Table 5.

Summary of IFU sentence review.

Category	n
Taxonomy Levels
Understand	11
Evaluate	15
Create	9
Physical	63
Perceptual	14
IFU Characteristics
Effort
Cognitive	52
Physical	60
Control Type
Human	112
Machine	4
Interaction State
Initial-priming	30
Active	76
Final	9
Steady	1
Timing
Long	65
Short	51

Effort: While there was a disparity in how the cognitive and psychomotor tasks were referenced in IFUs, analysis of effort demands revealed a near balance between cognitive tasks (n = 52) and physical tasks (n = 60) (Table 5). Only human-initiated sentences were sorted by effort. This near balance indicates that cognitive effort is important for IFU step completion but is not always emphasized, as described above. Effort is a crucial indicator for creating Task Categories, as it can quickly divide steps into groups that use different aspects of the unified cognitive-psychomotor model. For example, an IFU step that requires only cognitive effort uses only the PS block.

Control type: The majority of IFU steps reviewed were human-initiated (n = 112) with a few machine-initiated steps (n = 4) (Table 5). While most steps reviewed were human-initiated, Task Categories still need to distinguish between them. As mentioned above, users will approach task completion differently depending on who is responsible for each step; therefore, Task Categories need to specify who is responsible for each step.

Interaction state: The interaction state distribution further highlighted the need for Task Category distinctions (Table 5). The active state (n = 76) had the highest representation, compared with initial-priming (n = 30), final (n = 9), and steady (n = 1), suggesting that IFUs overwhelmingly emphasize real-time engagement. As noted previously, errors can occur during transitions between these states. The Task Categories were created with this in mind to help designers identify where errors are occurring during product use.

Timing: Analysis of timing demands in IFUs revealed a near balance between short (discrete effort) tasks (n = 51) and long (continuous effort) tasks (n = 65) (Table 5). The split between short-duration steps supported the subdivision of human-initiated physical actions into two distinct Task Categories. These timings vary in the amount of conscious cognitive effort a user can apply to changing their task performance [56].

IFU Characteristic Analysis Outcome: The IFU characteristics reviewed above provided insight into the aspects of HMI principles that designers find important to communicate to users. Designers tend to focus on describing the physical tasks and aspects of product interaction, further underscoring the need for DECIDE-CTA to provide support to designers in contextualizing cognitive tasks. The creation of Task Categories was deemed necessary to provide designers with a structured way to link their HMI knowledge with the cognitive-psychomotor unified model.

Task categories

Task Categories were created to serve as a structured bridge between the cognitive-psychomotor unified model and product-specific documentation by linking the above IFU patterns to the structural cognitive embedding, IP model sequencing, design characteristics (buttons, lights, tones), and sentence syntax templates. These Task Categories were designed to facilitate the DECIDE-CTA creation process.

To account for these differences in the number of human- versus machine-initiated sentences, Task Categories 1a and 1b were created. 1a tasks capture situations where the machine initiates the process, and the user’s role is limited to passive perceptual confirmation. 1b tasks capture situations where the user initiated a task in a previous step and is now waiting for the machine to complete it.

The even balance of the effort demands led to the need for Task Categories that address perceptual–cognitive tasks (Task Categories 1a, 1b, and 2) and psychomotor tasks (Task Category 3). Additionally, the time required to complete active-state HMI tasks led to splitting Task Category 3 into 3a and 3b. Task Category 3a captures tasks that allow for continuous perceptual monitoring and time to make conscious adjustments during performance. In contrast, Task Category 3b captures rapid, discrete tasks completed before conscious corrective perceptual monitoring feedback can be applied. Table 6 provides a shorthand describing the Task Categories and an example of an IFU step belonging to that Task Category.

Table 6.

Task categories and IFU examples.

Task Category	Description	IFU Example
1a	Machine-initiated, passive human perception	Listen for three beeps when testing is complete
1b	Human-initiated on the prior step, machine-completed	View the progress bar as the software is downloaded
2	Human-initiated, passive human perception	Look at the shape of a component.
3a	Human-initiated, physical effort, long time-lapse to complete motion	Press the brake pedal to slow down
3b	Human-initiated, physical effort, short time-lapse to complete motion	Push the power button to turn on the device

Table 7 provides an overview of how the five Task Categories compare in terms of model and taxonomy alignment, interaction state, the action's initiator and completer, and the effort required. The comparison shows that Task Categories 1a and 1b cluster around perceptual, no-effort tasks, while Task Category 2 is distinguished by steady-state monitoring. In contrast, Task Categories 3a and 3b are defined by physical effort and differ primarily in whether the action unfolds over time or is completed instantaneously.

Table 7.

Attributes of task categories derived from IFU analysis.

		Task Category					Selection Justification
		1a	1b	2	3a	3b
Model-Taxonomy	PS-REM	x	x	x	x	x	The PS block is required for the performance of all Task Categories.
	PS-UND	x	x	x	x	x
	PS-EVA	x	x	x	x	x
	PEP-REM				x		Task Category 3a is the only human-initiated category with a completion time that allows for conscious evaluation of the PEP block.
	PEP-UND				x
	PEP-EVA				x
	PHY-REM				x	x	Task Category 1a, 1b, and 2 do not require physical activity to complete them.
	PHY-UND				x	x
	PHY-EVA				x		These pairings are not available for Task Category 3b tasks since there is not enough time to consciously evaluate or adapt one's performance of human-initiated tasks.
	PHY-CRE				x
Interaction States	Initial Priming		x		x	x	1a tasks are machine-initiated; the human can only observe the outcome.
	Active		x		x	x
	Final	x	x		x	x
	Steady			x			Task Category 2 is the only category concerned with observing something that will not change, no state change is needed.
Action Initiation	Human		x	x	x	x	Task Category 1a tasks are the only ones that are machine-initiated.
	Machine	x
Action Completion	Human			x	x	x	Task Categories 1a and 1b are the only ones in which the machine has to complete the action.
	Machine	x	x
Effort	No Physical Effort	x	x	x			Task Categories 3a and 3b are the only ones that require physical activity; they require effort to complete.
	Physical Effort				x	x

IFU Review Outcome: The IFU review provided insight into how designers view and utilize cognitive and psychomotor tasks when describing product use. These insights were combined with the cognitive-psychomotor unified model to create Task Categories for designers to use when generating CTAs. The Task Categories allow steps within IFUs to be sorted and provide designers with a discrete list of tasks to include in their task decomposition. However, these Task Categories do not guide how to describe or order tasks within a CTA; therefore, sentence syntax and CTA templates were created and are discussed in more detail below.

Phase 4: sentence syntax and CTA template development

Phase 4 focused on standardizing CTA generation using cognitive sentence syntax and templates (addressing CTA Gap 4: lack of a standardized method). Family 4 models described, at a high level, how to create a CTA but did not provide strict guidelines on how to structure the information or what to include in a CTA. Providing sentence syntax and templates for designers to use will enable them to apply DECIDE-CTA consistently.

Cognitive sentence syntax development

Developing cognitive sentence syntax was critical for translating the embedded structure into a format that designers can apply to describe tasks in their CTAs. Where the structural cognitive embedding defines tasks, the sentence syntax enables the use of those tasks in standardized linguistic templates. Analysis of the sentences within IFUs revealed consistent structural and content-related patterns across taxonomy levels and embedded model types. This ensured that tasks are documented in ways consistent with how designers already write IFUs, while making cognition and psychomotor activities explicit, repeatable, and directly linked to product documentation.

Syntax Contextual Cues: The CTA sentences need to communicate which cognitive task is being performed reliably. The use of verbs associated with a particular level alone is not enough to communicate the taxonomy level. As noted previously, some verbs were found to be used across multiple taxonomy levels; therefore, the verb alone may not be interpreted the same way by designers. Cognitive contextual cues were added to help ground the sentences to a particular level.

The definitions of the four levels from the current adaptation of Bloom’s Taxonomy were used to generate the cognitive contextual cues incorporated into each sentence's syntax. Each level was translated into a rule-based linguistic element that linked cognitive function within a product interaction context. The contextual cues are all based on the definitions of Anderson and Krathwohl’s revision of Bloom’s Taxonomy (Table 2, Table 8).

Table 8.

Syntax Contextual Cues.

Taxonomy Level(adapted for use in current research)	Anderson and Krathwohl’s Level Definitions [26]	Contextual Syntax Cue (developed for current research)
Remember	Retrieve relevant information from long-term memory	based on seeing, hearing, or feeling its similarity to another device
Understand	Construct meaning from instructional messages	given recent training, understand how to…
Evaluate	Make judgments based on criteria and standards	based on seeing, hearing, or feeling the decision criteria….
Create	Put elements together to form a coherent or functional whole	based on/given adjustment criteria … adjust the …

Syntax Creation: The syntax creation process was informed by the current adaptation of Bloom’s and Harrow’s Taxonomies, as well as the IFU analysis. The creation of the cognitive sentence syntax provides a cognitive grammar for designers to use when creating a CTA. Cognitive grammar enables designers to think of CTA generation as a language system with syntax, semantics, and pragmatics. Patterns observed in IFU sentences, such as consistent references to design characteristics and temporal markers, were incorporated into syntax rules to preserve familiarity. At the same time, contextual cues were added to give the sentence-syntax templates a cognitive-task focus. By adding the validated taxonomies into these templates, the syntax ensures that task descriptions remain theoretically rigorous while accessible to designers.

Each sentence template was designed to include five mandatory components:

• 1.Cognitive activity (e.g., recall, evaluate, create)
• 2.Perceptual or psychomotor task (e.g., seeing a cue, effort used)
• 3.Design characteristic (e.g., button, display, alarm)
• 4.Temporal context (e.g., before, after, during)
• 5.Cognitive or psychomotor contextual cues (e.g., recalling prior experience)

These syntax components ensure that each sentence created captures not only what the user does, but also why it is done, when it occurs, and which product feature it relies on in a standardized way. Table 9 shows the process for creating the sentence templates. Column 1 contains the steps followed, column 2 describes what is taking place at that step, and column 3 provides an example or outcome of that step.

Table 9.

Creation of cognitive sentence syntax.

Step	Description	Example / Outcome
1. Define Task Categories	Task Categories were defined using the current adaptation of Bloom’s and Harrow’s taxonomies, pairing each level with representative verbs.	Example: “Remember” level → verbs like recognize, identify
2. Align with IFU phrasing	Template structures were cross-checked against sentence patterns in IFUs to ensure that phrasing matched the way designers naturally communicate tasks.	IFUs commonly follow this pattern: Perform [verb] on/based on [part of product] to achieve [goal] within [time frame].
3. Encode timing for selection of states	The time frame element (e.g., “before use,” “during operation”) was used to indicate when the task occurs	Example sentence: “Identify that the initial gray color of the power button means the device is OFF.”
4. Add cognitive contextual cue	A cognitive cue was appended to clarify the underlying reasoning or mental process, reducing the risk of misclassification across taxonomy levels.	Updated example: “Identify that the initial gray color of the power button means the device is OFF, based on seeing its similarity to power buttons on other devices.”
5. Iterative revision	Templates were iteratively refined to ensure syntactic consistency and alignment with the refined taxonomies while maintaining natural readability for designers.	Final template maintained structural consistency across all taxonomy levels.

Syntax for Each Taxonomy-Block Combination: The standardized syntaxes developed using the steps above for each level demonstrated how cognitive processes can be represented in a uniform linguistic format. Each syntax contains placeholders in brackets that guide the replacement of necessary linguistic elements: verbs representing cognitive activity, nouns describing the interaction state, product design characteristics, sensory cues, and contextual cues. These brackets represent necessary parts of the syntax and places where the designer needs to provide input about their product. Optional terms, indicated by strikethrough, allow flexibility where perceptual overlap occurs (e.g., the word "click" can refer to auditory or tactile feedback). The resulting syntaxes strike a balance between standardization and adaptability, while ensuring that cognitive tasks are represented with both precision and contextual relevance. Task Categories will also assist with the standardized use of the cognitive sentence syntaxes, since not every syntax can be used with every Task Category (Table 7). Below are descriptions of the sentence syntaxes created following the steps above.

The Remember syntax, as shown below, contains the following elements: begin with a verb describing the action being performed (recall or a synonym for recall), followed by an indication for the interaction state (is it being performed before, during, or after some indication), what characteristic of the device is being observed (e.g., color, placement, sound), what part of the device that characteristic is coming from (e.g., screen, speakers), and the contextual cue relating that information to a prior experience with another device (a similar product to what they are currently using). These elements of the syntax only alter slightly between the blocks.

• •PS syntax: {Recognize/identify} that the {initial/final/steady state} {specific characteristic} of the {device} {device component}'s shows/indicates that {state}, based on its similarity to {device} {device component} in/on other devices
• •PEP syntax: {Recognize/identify} that the {device} {device component} {active state (in progress)}{specific characteristic} while {performing physical action}, based on its similarity to the {device} {device component} in/on other devices
• •PHY syntax: {Recognize/Identify} the {characteristic} needed to {perform physical action on}{device}{device component}, based on its similarity to the {characteristic} used on {device component}

The Understand syntax contains the same elements as the Remember syntax, except it begins with the cognitive contextual cue about training to help distinguish it from a Remember task, since they can share verbs that describe each activity.

• •PS syntax: Given/based on recent training, {understand/infer} that the {initial/final/steady state}{sub characteristic} of the {device} {device component} means that {state}
• •PEP syntax: Given/based on recent training, {understand/infer} that the {state} {sub characteristic} of the {device}{device component}, while {you are performing a physical action}, means {active state}
• •PHY syntax:{Based on/given} recent training, {understand/infer} that the {amount of effort applied} to the {device}{device component} which means/controls/influences/sets how the {device component} {responds to that effort}

The Evaluate syntax differs in two ways from the syntaxes above. It begins with an indication that a judgment is being made about a human- or machine-performed action and ends with a cognitive contextual cue focused on certain decision criteria (e.g., the state of the device).

• •PS syntax: Before/While/After {you/machine} {performs action}, {evaluate/determine} if the {initial/final/steady specific characteristic state} {given/based on} {seeing/hearing/feeling} the {device component} {decision criteria}/{decision criteria} of the {device component} on the {device}
• •PEP syntax: While {you perform action} on the {device}{device component}, {evaluate/determine} if the {active specific characteristic state} {given/based on} {seeing/hearing/feeling} the {decision criteria} of the {device component} on the {device}
• •PHY syntax: While {using device component}, {evaluate/determine} {your ability to apply sufficient effort} on the {device}{device component to complete a motor-control activity (in progress use)}{given/based on} the {completion decision criteria}

The Create syntax follows a similar structure to Understand syntaxes. It begins with a cognitive contextual cue focused on the criteria that require the creation of new information.

• •PHY syntax: Based on/given {adjustment decision criteria}of the {device} {device component}, {adjust/vary} the {characteristic} while {performing physical action} on {device} {device component}

The structure of the cognitive sentence syntax enables the automation of CTA generation. Software can be trained on task categories and cognitive sentence syntax. The task categories can guide the software through the customizable parts of the syntax for each IFU step (e.g., interaction state). Designers can integrate design characteristics into the software, allowing it to recognize them within IFUs. This software can apply the design characteristics provided within the selected syntax. Using software designed around DECIDE-CTA would allow designers to quickly generate CTAs for longer IFUs, where they might start satisficing to speed up the process, thereby undermining the method's effectiveness.

Syntax Creation Outcome: Creating the cognitive sentence syntax provided a standardized method for designers to describe the tasks within their CTAs. The CTA templates were created to guide designers on which sentence syntaxes to use and presentation order.

Structured CTA template

Despite the variety of formal CTA methods discussed previously, the CTAs produced by these methods share the same output structure. They start with a goal and outline the steps necessary to achieve it. The outline can be set up with a series of if-then statements that show the hierarchical and categorical performance of these steps [17,[83], [84], [85]]. This structure was translated into a template for use in DECIDE-CTA that builds upon the information in IFUs and the structure of the cognitive-psychomotor unified model. All CTAs begin with a goal, which in this case is the IFU step being decomposed. At a high level, the IFU step explains what needs to be done and the expected result. From there, the IFU step is decomposed into the three models (PS, PEP, and PHY). These models specify the type of action the user needs to take while completing the IFU step. Within each model are the taxonomy levels from the current adaptation of Bloom's Taxonomy. The designer will need to write sentences describing what is happening at each taxonomy level. The execution of these models is categorical, where all taxonomy levels must be performed before moving on to performing another model, as seen in Figs. 6 and 7.

The template can help designers guide the CTA generation process. The basic form of a CTA is as follows:

• •
  Goal: Individual IFU Step

  • ·
    PS

    • ▪
      Remember
    • ▪
      Understand
    • ▪
      Evaluate
  • ·
    PEP

    • ▪
      Remember
    • ▪
      Understand
    • ▪
      Evaluate
  • ·
    PHY

    • ▪
      Remember
    • ▪
      Understand
    • ▪
      Evaluate
    • ▪
      Create

This basic structure requires expert knowledge to fill in the outline with the necessary detail for task decomposition, as well as how to write sentences describing the decomposition. This template was converted into a designer-accessible format by incorporating the Task Categories. Task Categories will be used to determine the interaction state of an IFU step, which affects which models are needed and omits taxonomy levels based on the type of activity occurring in the step. The cognitive sentence syntax will then be used to build sentences for each taxonomy level. This reduces the subjectivity of determining what is needed to complete an IFU step, therefore increasing consistency in CTA generation. The integration of the Task Categories and cognitive sentence syntax enabled the creation of a fully articulated CTA template that serves as a standardized hierarchy (Appendix List A).

Selecting a Task Category provides the customization needed to make this a designer-facing template. Task Categories determine the interaction state, model, taxonomy, cognitive sentence syntax, and possible designer characteristics. By selecting a Task Category, designers are presented with an outline that they fill in with HMI knowledge. The template’s structure and Task Category selection guide the performance order by hierarchy and Task Categories. This means that the Task Category specifies the order of the model's performance based on the interaction states (e.g., PS initial-priming tasks are performed first, and the final state is performed last). The taxonomy levels in each model are performed in the order specified by the current adaptation of Bloom’s Taxonomy. A fully fleshed out CTA for a Task Category 1a IFU step can be found in Appendix List B.

In addition, flowcharts have been created that show how to use DECIDE-CTA (Appendix Figs. 1 through 8). These flowcharts show the decisions the designer must make when using this method (e.g., specifying design characteristics) and the decisions the method makes (e.g., determining which syntax the designer must use). Two examples that follow the flowcharts have also been provided. Each part of the example follows the necessary flowchart for completing it.

DECIDE-CTA Outcomes: The goal of the DECIDE-CTA method is to enable designers to systematically embed cognition, perception, and psychomotor tasks into task decompositions. Existing task analysis methods do not provide designers with guidance on integrating cognitive tasks into a CTA in an accessible and repeatable manner. This lack of guidance weakens risk mitigation strategies and leaves possible causes of use errors unaddressed, reducing product safety.

The method developed here addressed these gaps through several innovations:

• •Embedded cognitive and psychomotor taxonomies. By embedding Bloom’s and Harrow’s taxonomies, the method provides structure for how these types of tasks influence one another.
• •Sequencing representation of cognition–perception–action. The structural cognitive embedding was combined with IP models to create three blocks (PS, PEP, PHY). These blocks provide an order to how cognitive tasks flow within psychomotor activities, ensuring tasks are sequenced in ways that mirror real-world product use. This reduces analyst subjectivity in task ordering and strengthens use error tracing.
• •Task Categories and sentence syntax. Task Categories (1a, 1b, 2, 3a,3b) capture variations in effort, initiation, and interaction state. These Task Categories help designers determine what tasks are needed for their CTA. Cognitive sentence syntax provides a structure for designers to follow when describing the cognitive tasks required for product use.

Overall, DECIDE-CTA extends the existing CTA families by integrating the strengths of Family 3 taxonomies and IP models, as well as Family 4 formal modeling approaches, into a designer-accessible framework. Unlike traditional CTA methods, which require designers to independently determine task decomposition, cognitive structure, and task descriptions, DECIDE-CTA provides predefined task categories, embedded cognitive-psychomotor relationships, and rule-based syntax for CTA generation. This shifts part of the analytical burden from the individual designers to the framework, improving consistency, repeatability, and cognitive traceability. Consequently, DECIDE-CTA serves as a translation layer between cognitive theory and practical design activities, enabling designers without a formal cognitive background to generate structured CTAs suitable for human factors evaluation, risk assessment, and future predictive modeling applications.

These contributions establish a designer-accessible CTA method that improves repeatability, reduces designer subjectivity, and provides a structural cognitive embedding that ties directly into product design characteristics. The findings point to several next steps for advancing this work in device design practice:

• •Validation study of the cognitive-psychomotor unified model. This model can be validated by recruiting designers to categorize sentences created using DECIDE-CTA. The CTA templates, syntax, and task categories would guide sentence creation, with a focus on cognition, given its existing underrepresentation in IFUs. Then, the subjects will categorize them. The results of the study will show how the designers group cognitive tasks to provide insight into whether or not each level is viewed as an independent task.
• •Formative and Summative method studies. The cognitive-psychomotor unified model can be paired with human factors tools. For example, regulatory agencies (e.g., the Food and Drug Administration, U.S. Nuclear Regulatory Commission, National Highway Traffic Safety Administration, Department of Defense) and standards bodies (e.g., ISO, IEEE, ANSI) have issued human factors guidance documents (e.g., FDA- Human Factors Guidance document [86], MIL-STD-1472 [70]).

As part of these documents, designers are asked to perform a task analysis. DECIDE-CTA would enable designers to create a task decomposition that captures cognitive performance at every step of a human-machine interaction. Designers are also asked to use self-assessment tools to gain feedback from product users. These surveys are used in usability testing to gather user feedback on their experience with the product. Current self-assessment tools and guidance lack cognitive depth. DECIDE-CTA will aid in the design of assessment tools by providing a list of tasks for designers to develop questions around. The cognitive sentence syntax will serve as the basis for the survey questions to ensure they target the desired task. These surveys will link the self-assessment questions to the cognitive-psychomotor unified model, helping improve the accuracy of risk mitigation.

• •Assessment of CTA Framework scalability. The CTA Framework was developed with insights from IFUs that were lay-user-friendly, simple, and relatively short. Short IFUs contain only a few steps and do not result in a long CTA. A study could assess the scalability of this method for longer or more complex IFUs. Longer, more complex IFUs might have multiple sections' worth of steps to deconstruct. This could result in designers satisficing to avoid creating a full CTA for these documents. Designers might want to focus on steps they know users have the most trouble with or those that pose the greatest risks. The framework can be combined with techniques for assessing the likelihood and severity of use errors to narrow the focus to the steps for creating CTAs. This allows designers to create CTAs that target the most dangerous steps in longer or more complex IFUs in an accessible manner.
• •Automation and digital tooling. The Task Categories and cognitive sentence syntax are rule-based; therefore, computational tools, such as generative AI (GenAI), specifically large language models (LLMs), can aid the CTA generation process. The structure of DECIDE-CTA provides a foundation for automating parts of CTA generation. LLM-based systems can be trained to identify device characteristics and components, as well as task categories, rather than having designers do this manually. Multiple implementation approaches can be utilized. A simple approach could use prompts, in which the LLM is provided with the rules of DECIDE-CTA and guided through the process step by step. Domain-specific models can be developed to streamline CTA generation and enhance consistency across products and industries. LLMs also improve DECIDE-CTA's scalability. Designers could rapidly generate a CTA for longer, more complex IFUs. Future studies are needed to evaluate the consistency and accuracy of such LLMs.

GenAI tools can be trained to translate IFUs into CTAs using cognitive sentence syntax, which can aid designers in creating initial CTA drafts and ease method use.

Method validation

A preliminary validation process of DECIDE-CTA was embedded within each sequential methodological phase (introduced previously) rather than conducted as a separate post-development evaluation. Table 10 summarizes how preliminary validation was integrated across each phase, followed by descriptions of the activities. Preliminary theoretical validation was assessed by ensuring DECIDE-CTA aligned with existing theories; structural validation centered on ensuring the method was built upon valid frameworks; ecological validation ensured that DECIDE-CTA was grounded in real-world concepts designers use; and reproducibility validation focused on providing guidance to standardize its use.

Table 10.

Validation activities across DECIDE-CTA methodological phases.

Phase	Validation Type	What Was Validated
Phase 1	Theoretical + Preliminary SME Validation	Taxonomy selection, level merging, cognitive–psychomotor embedding logic
Phase 2	Structural + Model Preliminary Validation	Temporal sequencing, IP model alignment, unified block structure
Phase 3	Preliminary Empirical + Ecological Validation	Alignment with real-world IFUs, HMI grounding, and task categorization
Phase 4	Preliminary Standardization + Reproducibility Validation	Sentence syntax, contextual cues, and variability reduction

Phase 1: conceptual model selection (Theoretical and preliminary SME validation)

Theoretical validation in Phase 1 occurred through systematic taxonomy review and SME analysis. Multiple cognitive taxonomies (Bloom, Gagne, Guilford, Upton) and psychomotor taxonomies (Harrow, Dave, Simpson, Kibler) were comparatively evaluated to determine their suitability for designer accessibility, task discreteness, and cross-user consistency. Bloom’s and Harrow’s taxonomies were selected because they provided structured, action-oriented task lists with sufficient granularity while avoiding excessive complexity or conceptual ambiguity. Beyond selection, preliminary validation was embedded through a structured SME review spanning 22 scholarly sources across 10 disciplines. Each source was coded to identify how perception, cognition, and movement were described and mapped to the adapted taxonomy levels. This empirical coding process provided preliminary validation of how cognitive tasks are embedded within perceptual and physical activity, ensuring that the final embedding logic reflected interdisciplinary consensus rather than theoretical assumption. The SME review further informed structural refinements. Bloom’s Understand and Analyze levels were merged due to demonstrated functional interdependence, and Harrow’s Basic-Fundamental, Physical, and Skilled levels were collapsed into a unified Physical level due to a lack of distinction in expert descriptions. These refinements reduced ambiguity and improved usability for non-cognitive experts.

Phase 2: cognitive–psychomotor unified model (structural and model preliminary validation)

Phase 2 involved preliminary validation of the sequencing and structural organization of the embedded taxonomies by aligning them with established IP models and Family 4 modeling approaches. A review of Family 3 IP models, including Welford’s model, Endsley’s Situation Awareness model, Wickens’ Multiple Resource Theory, and the Model Human Processor, revealed consistent representation of human performance as a cyclical flow of perception, cognition, and action. This cross-model convergence validated the decision to organize DECIDE-CTA into three structured blocks: PS, PEP, and PHY. These blocks formalize the temporal and feedback relationships described in IP theory, ensuring that the cognitive–psychomotor embedding aligns with established representations of human information flow rather than inventing a new sequence structure. Family 4 models (e.g., GOMS, CWA, ACT-R, SOAR) were reviewed to assess how task lists are typically operationalized in predictive systems. This review provided preliminary validation of a critical methodological gap: while these models assume a structured task list and sequence, they do not standardize task decomposition or linguistic description. DECIDE-CTA addresses this limitation by embedding predefined taxonomies into an IP-based flow structure, thereby reducing analyst subjectivity prior to modeling.

Phase 3: IFU review (preliminary empirical and ecological validation)

Phase 3 embedded preliminary validation by grounding the method in real-world designer documentation practices. A total of 121 IFUs were reviewed across multiple industries, and 116 deconstructed instruction steps were analyzed using taxonomy-aligned verb extraction. This empirical dataset provided preliminary ecological validation, demonstrating that DECIDE-CTA aligns with how designers naturally describe product interactions. The IFU analysis revealed a strong emphasis on physical tasks and underrepresentation of cognitive activities, confirming the need for structured cognitive embedding. Additional coding of effort, control type, interaction state, and timing provided quantitative evidence that IFUs implicitly reflect cognitive and psychomotor demands even when not explicitly labeled. These findings offered preliminary validation of the relevance of the cognitive–psychomotor unified model and justified the development of Task Categories.

Task Categories (1a, 1b, 2, 3a, 3b) were derived from observed IFU patterns in conjunction with HMI principles and taxonomy alignment. Their creation was therefore empirically informed rather than theoretically imposed.

Phase 4: sentence syntax and CTA template development (preliminary standardization and reproducibility validation)

Phase 4 preliminary validation activities at this stage focused on ensuring consistency, clarity, and reproducibility of task descriptions. Cognitive contextual cues were derived directly from Anderson and Krathwohl’s definitions of the levels of Bloom’s taxonomy to reduce ambiguity in interpreting verbs. This process ensured that taxonomy assignment relied on definitional grounding rather than isolated verb usage. Sentence templates were structured to include five mandatory components: cognitive activity, perceptual or psychomotor task, design characteristic, temporal context, and contextual cue. By requiring these components, the syntax enforces completeness and reduces omission of critical cognitive information.

Limitations

The development of this CTA framework was limited by ambiguity in the boundaries of the PS-PEP-PHY blocks and by the lack of human-subject empirical validation of the derived Task Categories.

Ambiguity in Block Boundaries: One limitation of the cognitive-psychomotor unified model is the potential ambiguity in defining boundaries between the PS, PEP, and PHY blocks. Distinguishing when a perceptual cue should be treated as a standalone activity and when it is necessary to guide motor control is not always straightforward. Research has found that perceptual and motor processes often occur in overlapping or iterative cycles rather than as discrete phases [73,74]. This raises the possibility that a perceptual task could be classified differently depending on interpretation, thereby reducing the method's consistency and repeatability. This limitation was addressed by creating the cognitive sentence syntax templates and the method flowchart. These templates provide additional information for the designer to consider when determining what block to use. The templates relate user actions to the design characteristics of the device, helping to guide the designer's thinking about when a perceptual task is being used to guide an action or to observe. The method flowchart shows that DECIDE-CTA will provide the necessary cognitive sentence syntax template for the designer to fill out, so they do not have to guess whether to include the PEP block for a given IFU step.

Human Subject Empirical Validation of DECIDE-CTA: Empirical validation of CTA methods is needed to ensure the methods effectively capture the desired information and provide insight into practical applications [11,15]. Future work will focus on evaluating how designers perceive the latent cognitive structure of DECIDE-CTA and the interrater reliability of the CTAs generated with this method. This form of validation must be performed for the methodological elements developed as part of DECIDE-CTA to determine how designers interpret and apply the information in practice. For example, Task Categories 3a (continuous effortful action) and 3b (discrete effortful action) could be viewed as closely related and may lead to comprehension issues among designers. Ensuring designers understand the DECIDE-CTA elements helps to ensure correct and consistent application. While the current stage of methodological development did not include empirical validation in the designer population, as noted in the methodological validation section, several key steps were undertaken to provide a methodological foundation and preliminary validation for subsequent testing. First, Task Categories were grounded in existing HMI principles observed in IFU sentence patterns. Second, the cognitive-psychomotor unified model is grounded in educational knowledge representations. These educational knowledge representations were designed to communicate cognitive and psychomotor processes without a background in cognitive science [2,3]. Third, these knowledge representations were further refined with input from SMEs, which helped to remove potentially redundant taxonomy levels.

CRediT author statement

Janell Joyner: Conceptualization, Methodology, Software, Investigation, Writing- Original draft, Writing-Reviewing and Editing. Monifa Vaughn-Cooke: Conceptualization, Supervision, Writing- Reviewing and Editing

Ethics statements

The methodological development referenced in this manuscript did not include human subjects, animal experiments, or data collected from social media platforms.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

I have shared a link to my code in the manuscript.