Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Situation Awareness in Dynamic Human-System Control

Abstract. Situation awareness in dynamic human-system control concerns how operators perceive changing system states, comprehend their operational significance, and project consequences before decisions become irreversible. This fundamental research monograph traces the historical development of situation-awareness science, distinguishes the three-level cognitive model from ecological and distributed approaches, and investigates the roles of attention, mental models, workload, expertise, mode awareness, and automation. Its critical narrative synthesis examines primary theoretical and experimental research alongside systematic reviews and meta-analyses, including direct probe measurement, competing psychometric instruments, the relation between awareness and performance, and the routine-versus-failure tradeoff in supervisory automation. Application-focused chapters analyze aviation, industrial networks, healthcare teams, transportation, and remotely controlled systems, with particular attention to the evidence needed for causal claims. The final chapters establish research questions for interface design, training, temporal measurement, machine transparency, and human–artificial control in the Artificial Era. Findings support situation awareness as a task-dependent explanatory construct with measurable associations to performance, substantial heterogeneity, and continuing challenges in operational definition, instrument validity, and systems-level generalization. This work is a critical narrative scholarly synthesis, not a preregistered systematic review, new experiment, or independently conducted meta-analysis.

Keywords: situation awareness; human factors; engineering psychology; dynamic systems; human-system control; perception; comprehension; projection; SAGAT; SPAM; psychometrics; mode awareness; automation surprise; supervisory control; mental models; ecological interface design; distributed cognition; team situation awareness; adaptive automation; human–artificial coordination; Artificial Era

Situation Awareness as a Scientific Object

Situation awareness is an operator's task-relevant understanding of a changing environment and the probable near-term consequences of that environment for action. In engineered systems, the environment includes physical processes, interface representations, machine modes, other actors, and the evolving relation between current conditions and operational goals. The classical formulation in Endsley (1995) separates perception of elements, comprehension of their meaning, and projection of future states. Its scientific importance lies in treating awareness as a temporally situated state that can be investigated through specific questions about system variables, relationships, and likely developments. A controller who notices a pressure indicator, correctly understands an emerging instability, and anticipates its direction demonstrates different kinds of knowledge than an operator who simply looks at the display. Measurement must therefore specify which information was relevant, when it was relevant, and what was known at the decision point.

The object of this monograph is situation awareness in dynamic human-system control: the formation, updating, distribution, and use of task-relevant representations during regulation of systems whose states continue to change. Such systems include aircraft, air traffic, transportation networks, clinical monitoring arrangements, industrial processes, and human-supervised automated agents. The key research questions concern how operators acquire information, organize it into an explanatory model, predict system trajectories, coordinate across humans and machines, and recover when their picture diverges from reality. An engineering outcome such as safe recovery is consequential evidence, but it is not interchangeable with knowledge of state. An operator can understand a hazardous situation and still lack authority or time to correct it; favorable outcomes can also arise despite deficient understanding. The distinction allows psychological processes and system affordances to be investigated jointly Sarter & Woods (1991) Bakdash et al. (2022).

This work is a critical narrative synthesis of foundational theory, original human-factors experiments, methodological comparisons, systems research, and published systematic reviews and meta-analyses. It introduces no new experimental sample, pooled estimate, or formal systematic-review claim. Historical models are evaluated alongside contemporary psychometric evidence, with special attention to what each design actually measures and which inferences it can support. The organization follows the scientific problem from perception and temporal control through theory, operational measurement, empirical performance, high-risk applications, design, and human–artificial coordination. The resulting framework treats awareness as a measurable, task-specific explanatory construct whose mechanisms and limits remain open to comparative investigation Moens et al. (2026) Stanton et al. (2017).

Dynamic Systems, Control, and the Temporal Structure of Awareness

A dynamic control environment contains variables that continue to evolve while operators observe, interpret, and act. Actions alter subsequent observations, and delays, disturbances, feedback loops, and hidden states complicate inference. In a power network, a measurement may reflect a downstream disturbance rather than its cause; in aviation, the meaning of an altitude indication depends on flight phase, trajectory, and control mode. Situation awareness accordingly has a temporal structure: evidence arrives intermittently, an internal account of the system is updated, and decisions are made before the future is directly observable. The scientific unit is a coupled process linking environment, observation, knowledge, forecast, and control. Classical cognitive explanations emphasize human representations, whereas ecological work asks which constraints are directly disclosed by the environment and its instruments Endsley (1995) Vicente & Rasmussen (1992).

Three timescales must be considered. Millisecond-to-second changes govern display sampling and detection; seconds-to-minutes changes govern diagnosis, prediction, and corrective action; extended changes in operating regime govern learning, strategic priorities, and shifts in expected system behavior. The same operator can be well informed about present pressure but unaware of cumulative thermal drift; another can anticipate a dangerous trajectory while missing an abrupt alarm. Experimental tasks should manipulate rates of change, signal latency, decision deadlines, and availability of prospective cues independently where possible. Otherwise a supposed deficit in situation awareness may be an artifact of inadequate response time, motor restrictions, or the informational content of the display. Human-system design is evaluated by whether relevant causal and temporal relationships are available when action remains useful Smith & Hancock (1995) Stanton et al. (2001).

Dynamic conditions also make the observer part of the process under investigation. Supervisory automation can preserve nominal output while hiding intermediate changes from an operator; the appearance of stability can therefore coincide with deteriorating knowledge of contingencies. Bainbridge's analysis of the ironies of automation Bainbridge (1983) provides a foundational account of how taking routine control away from a human may leave demanding exception-handling tasks. The operative problem is the relation between mode of control, access to feedback, maintenance of procedural competence, and the time available to rebuild a working model. Awareness should be measured across the complete action-observation cycle, including recovery after a disturbance, rather than at an arbitrary instant detached from operational history Onnasch et al. (2014).

Historical Formation of Situation Awareness Research

The contemporary situation-awareness literature emerged from earlier traditions of aviation psychology, vigilance, human-machine interaction, and skilled control. Long before the expression became a specialized research term, investigators studied how operators select information from instrument arrays, coordinate perception with action, and maintain an understanding of systems under uncertainty. The rise of complex automated cockpits and command environments sharpened a recurrent question: what does the operator know about a system when immediate response is required? In 1991 Sarter and Woods described situation awareness as a critical yet insufficiently specified phenomenon, identifying a need for conceptual and empirical precision rather than relying on its intuitive appeal Sarter & Woods (1991). This challenge shaped subsequent debates over whether awareness is a cognitive product, a dynamic perceptual activity, a distributed achievement, or a label assigned after successful performance.

Endsley's 1995 theoretical treatment Endsley (1995) supplied a particularly influential model of information perception, comprehension, and projection, relating these levels to goals, expectations, attention, working memory, experience, and decision processes. A companion methodological paper formalized the logic of pausing realistic simulations and probing operators' knowledge of task-relevant facts Endsley (1995). In the same research period, Smith and Hancock Smith & Hancock (1995) argued for an externally directed, adaptive conception that places ongoing interaction and environmental coupling at the center of explanation. These publications established a productive disagreement over the locus of awareness while enabling investigators to design more explicit tests. Their historical significance must be distinguished from the proposition that any single framework has settled all questions about the construct.

Research subsequently expanded from individual aviation operators to supervisory control, distributed teams, industrial systems, and complex organizational settings. Mode-confusion studies highlighted the consequences of unexpected automation states Sarter & Woods (1995), and ecological interface design identified the value of making functional constraints perceivable Vicente & Rasmussen (1992). Work on team and distributed situation awareness broadened the unit of analysis beyond an isolated person's memory Salmon et al. (2008) Stanton et al. (2017). Contemporary systematic reviews now examine whether diverse measures converge, whether they predict performance, and which contexts moderate observed associations Bakdash et al. (2022) Moens et al. (2026). The trajectory of the field is therefore a movement from operational concern through competing theory and instrumentation toward evidence-sensitive comparison across tasks, measurement systems, and organizational forms.

Perception: Detecting Task-Relevant System States

Perceptual awareness begins with the registration and selection of information that can affect control. Relevant elements may include location, velocity, fuel, vital signs, machine configuration, status of adjacent actors, and environmental hazards; their relevance is conditional on the operator's task. Perception requires discriminating signal from noise, noticing a state change, and relating an observation to the correct object and time. An instrument may be clearly visible yet cognitively neglected because attention has been captured by a competing task. Conversely, a warning can be detected without being integrated into any coherent account of what is happening. The first level of the classical model describes access to task-relevant elements, not successful reasoning about those elements Endsley (1995).

Attention is finite and goal directed. Operators sample displays selectively according to current tasks, expected event probabilities, and learned priorities. These strategies can be adaptive when the environment is stable, yet dangerous when rare events violate expectations. Visual fixation and information use must also be distinguished: recording where a person looked reveals sampling behavior, while knowledge probes assess what information was retained or correctly interpreted. A high number of fixations does not establish a high-quality operating picture, and a low count may reflect efficient selective sampling by an expert. Investigations should therefore combine eye movement or interaction traces with content-specific questions and measures of task demands, rather than equating exposure with awareness Hart & Staveland (1988) Moens et al. (2026).

Perception also depends on how machines encode their states. A delayed telemetry update, ambiguous color code, inconsistent spatial mapping, or alert detached from its causal context can make accurate perception improbable even for an attentive operator. Ecological interface design proposes that perceptual displays should expose constraints and meaningful relations rather than isolated measurements Vicente & Rasmussen (1992). This is especially important when the required inference exceeds what a single numeral or icon reveals. A controlled experiment can compare matched informational content in different formats, test anomaly detection latency and missed changes, and evaluate whether improvements persist under workload. Such designs identify display-mediated mechanisms more convincingly than self-reports that merely ask whether an interface felt informative.

Comprehension: Constructing an Explanatory Operating Picture

Comprehension integrates perceived facts into an interpretation of the present situation. A controller who knows that pressure has fallen, a valve has changed position, and demand is rising must establish whether these observations constitute a single causal episode, several independent events, or harmless variation. The process draws on mental models of system structure, learned operating regimes, goals, and the consequences of component relationships. It can also involve identifying conflicts between indicators, distinguishing a sensor error from a physical disturbance, and recognizing that a standard procedure is now inappropriate. Endsley's second level of situation awareness captures this integration of features into task meaning Endsley (1995).

The correctness of a situation model is inherently task dependent. Two individuals can share observations and develop different interpretations because they control different variables or have different responsibilities. A flight crew member may prioritize aircraft configuration while an air traffic controller prioritizes separation among multiple aircraft. This difference is not automatically an error; the empirical question is whether each interpretation covers information required for the actor's own decisions and whether shared dependencies are recognized. Systematic studies should map comprehension queries onto actual work goals and domain constraints, rather than assuming that a single exhaustive description of the environment is possible Salmon et al. (2008) Stanton et al. (2017).

Comprehension failures can arise from overconfident causal attribution, misleading interface groupings, delayed or inconsistent feedback, and automation modes that alter the meaning of a control input. A factually accurate data stream may still be informationally inadequate if it conceals the operating logic that binds observations together. Mode confusion exemplifies this problem: the user may correctly perceive a displayed value but misunderstand which controller currently governs the system, producing an inaccurate interpretation of why the value changed Sarter & Woods (1995). The strongest empirical design tests not only whether an operator names the current mode but whether that operator can identify its consequences for authority, transitions, limits, and anticipated system response.

Projection: Anticipating Future States and Decision Windows

Projection extends current comprehension into estimates of what the system is likely to do next. Its object is a time-indexed expectation: a conflict may develop, a parameter may exceed a threshold, a patient may deteriorate, or an automated mode may hand authority back to a human. Accurate projection requires both a sufficiently correct present model and knowledge about rates, constraints, and likely interventions. It therefore draws on working models of dynamics, anticipated external events, and alternative action paths. The third level of Endsley's model Endsley (1995) specifies this prospective dimension and explains why possession of present-state facts alone may be insufficient for timely control.

Projection is measurable through explicit near-future questions and through the calibration of expectations against later observed states. A probe might ask which aircraft pair will approach a separation limit within a specified interval, whether a control process will cross a boundary if left unchanged, or which display state should follow a commanded mode transition. The evaluation must specify both horizon and acceptable error because forecasting five seconds ahead differs materially from forecasting several minutes ahead under changing inputs. Experimental designs can vary uncertainty, horizon, intervention options, and feedback availability to distinguish limitations in prediction from limitations in detection and comprehension Endsley (1995) Bakdash et al. (2022).

Effective projection supports anticipatory control by revealing when action is still feasible. The operator may intervene early to preserve safety margins, allocate attention to an impending critical phase, or request assistance before workload peaks. Yet forecasts are not infallible internal simulations: uncertainty grows under incomplete observation, unmodeled disturbances, and novel machine behavior. Evaluating projection should therefore include confidence calibration and performance under unexpected transitions, rather than rewarding confident point predictions alone. Research on adaptive awareness Smith & Hancock (1995) and supervisory control Bainbridge (1983) jointly suggests that system design must make both likely trajectories and uncertainty available to the person accountable for subsequent decisions.

Goal-Directed Processing and the Three-Level Model

The three-level model organizes situation awareness into perception, comprehension, and projection while locating each operation within an operator's goals and knowledge structures. Goals determine which elements warrant attention; expectations guide search; experienced operators use stored patterns to interpret changing inputs. The model does not require every judgment to proceed through three conscious serial steps. Perception, comprehension, and projection can be recursively coupled: noticing an unexpected state may alter the current goal, a new hypothesis may redirect visual search, and an anticipated event may determine which sensor is checked next. This architecture helps explain how an observer maintains a continuously revised understanding of a dynamic environment Endsley (1995).

A useful empirical interpretation distinguishes levels as functional accomplishments rather than assuming they represent isolated psychological modules. A person might correctly identify a warning and still be unable to infer its implications. Another might predict a conflict using learned patterns while struggling to verbalize every intermediate step. These dissociations matter for experimental probes and intervention design. A training program emphasizing display scanning may improve acquisition of individual cues without improving prediction; a causal schematic may improve interpretation while leaving rare-event detection unchanged. Researchers should identify the cognitive target of each intervention and measure each level with task-specific criteria rather than one undifferentiated global score Endsley (1995) Moens et al. (2026).

The model also requires a distinction between a product and the processes producing it. Situation awareness as a momentary informational state is not the same as vigilance, working memory capacity, expertise, or decision accuracy, even though all may influence it. Construct overlap emerges when an investigator asks only whether a participant selected the correct action: the response may reflect motor speed, incentives, or procedural habit in addition to knowledge. Theoretical clarity requires specifying what information the person possessed at a relevant moment and whether that representation supported the demands of the control task. A causal study should then separately model resource constraints, interface information, the resulting state of awareness, and downstream behavior Sarter & Woods (1991) Bakdash et al. (2022).

Ecological Accounts and Perception–Action Coupling

Ecological approaches redirect attention from internally reconstructed descriptions to information available in the operator's active relation with the environment. A cockpit instrument, process diagram, or physical control surface can make important constraints perceptually accessible. Smith and Hancock Smith & Hancock (1995) characterized situation awareness in terms of adaptive, externally directed consciousness, highlighting responsiveness to conditions as they change. In this view, awareness is sustained through seeking information, testing expectations, and adjusting action. The research implication is that a static question about remembered facts captures only part of an adaptive process, especially when an operator normally relies on consulting the world rather than retaining every measurement in memory.

Ecological interface design provides a concrete engineering program closely related to this perspective. Vicente and Rasmussen Vicente & Rasmussen (1992) developed a rationale for displays that reveal functional constraints and means–ends relationships in complex work domains. For example, a display that preserves the relation between material balance, pressure, and system limits may help an operator diagnose unfamiliar combinations of disturbances more effectively than several isolated meters. The intervention is not simply an aesthetically simplified screen; it changes the informational structure available for perception and inference. A rigorous comparison keeps underlying process data and operator goals constant while varying how relationships are exposed, then assesses detection, diagnosis, prediction, and control over novel disturbances.

The ecological and internal-state approaches can be compared through controlled predictions. If operators have access to a persistent external display, the importance of explicitly retained facts may decrease for tasks where rapid reacquisition is possible; if access is interrupted, internal representations and forecasting may become more decisive. The distinction can be tested by manipulating display persistence, information latency, alarm fidelity, and opportunities to interrogate system history. The question is how situational knowledge is accomplished with available environmental resources, not which vocabulary deserves exclusive ownership of the phenomenon. Complementary metrics can assess external information pickup, inferred constraints, and answers to factual state probes Smith & Hancock (1995) Vicente & Rasmussen (1992) Moens et al. (2026).

Distributed Cognition and Team Situation Awareness

Complex control is frequently organized across multiple individuals and artifacts. A surgical team, air traffic sector, or grid operations center does not require every participant to maintain an identical representation of all variables; relevant information is divided by role, expertise, and access. Distributed cognition provides a way to study information as it moves across people, displays, procedures, and symbolic artifacts. Hutchins Hutchins (1995) analyzed cognition in real work as a system involving coordinated representations, making it possible to ask how information can be preserved or transformed across a socio-technical arrangement. The operative unit becomes an organized information-processing system whose performance depends on coordination among its components.

Research on distributed situation awareness formalizes the idea that knowledge can be compatible and coordinated without being fully shared. In an energy-distribution case study, Salmon and colleagues Salmon et al. (2008) represented different actors' awareness in a collaborative operating environment. Their approach emphasizes which elements each participant requires and how relevant relations circulate through the network. A field study should therefore identify information ownership, handoff channels, message timing, and consequences of incorrect or delayed exchange. Counting how many team members know the same facts can be misleading when role-appropriate specialization, rather than uniform knowledge, is the design goal Stanton et al. (2017).

Distributed accounts create measurement challenges. An individual freeze-probe score may be high while critical dependencies between roles remain poorly coordinated. Conversely, one participant's inability to answer another participant's specialized question may be compatible with excellent collective functioning. Team analysis can use role-specific queries, communication networks, interaction traces, and system-level outcome measures to detect gaps or bottlenecks. Researchers should distinguish shared mental models, information accessibility, communication quality, and synchronized action; each can contribute differently to team situation awareness. The theoretically significant variable is whether the system contains sufficiently accurate and available representations at the locations and times where decisions are made Hutchins (1995) Stanton et al. (2017) Moens et al. (2026).

Mental Models, Expertise, and Recognition

Expert operators rely on knowledge of system structure, normal operating ranges, failure signatures, and the effects of candidate interventions. Such knowledge shapes what is noticed and how rapidly an unexpected pattern becomes meaningful. An expert may interpret a small conjunction of indicators as an emerging fault while a novice sees unrelated values. A mental model is therefore both a resource for comprehension and a source of selective attention. Situation awareness increases when the model accurately represents current control constraints, but experienced expectations can also bias perception when systems behave unusually. The cognitive account must explain both efficient pattern recognition and vulnerability to unfamiliar exceptions Endsley (1995) Sarter & Woods (1995).

Expertise is task specific. Knowledge that supports aircraft approach monitoring may transfer imperfectly to a novel cockpit interface whose automation states or alert priorities differ. Similarly, experience with routine industrial disturbances may not equip an operator to diagnose rare coupled failures. Training evaluations should separate procedural recall from causal understanding and distinguish performance on familiar cases from adaptive responses to unseen events. The most informative tasks include disturbances that invalidate stereotyped interpretations without introducing arbitrary tricks. Correct situation awareness can then be assessed by whether the participant revises a working explanation when the environment provides contradictory evidence Vicente & Rasmussen (1992) Stanton et al. (2001).

The development of expertise may involve improved attentional allocation, richer causal models, better retrieval of operating constraints, and enhanced prediction of downstream states. Experimental methods can trace these possibilities through scan paths, verbal protocol analysis where appropriate, time-stamped probes, and longitudinal simulation performance. Each measurement has limitations: think-aloud procedures can change task demands, eye tracking cannot disclose full reasoning, and questionnaires can overestimate metacognitive certainty. Converging evidence is stronger when experts and novices are matched on access to system information and comparisons account for differences in routine operating knowledge Moens et al. (2026) Hart & Staveland (1988).

Attention, Workload, and Prospective Control

Situation awareness interacts closely with mental workload but cannot be inferred directly from workload alone. Excessive task demands can reduce the time available to inspect displays and integrate information; very low demand during prolonged automated monitoring can also permit disengagement. A person may report high workload while maintaining an accurate model through disciplined prioritization, or low workload while overlooking a gradually developing anomaly. The NASA Task Load Index introduced by Hart and Staveland Hart & Staveland (1988) measures multiple subjective aspects of workload, not the accuracy of an operating picture. Treating workload as a proxy for awareness confounds resource demand with task-relevant knowledge.

Tasks can compete for attention across monitoring, communication, diagnosis, and motor response. A radio call arriving during a mode transition may divert attention at precisely the moment when a new controller state must be understood. Conversely, well-designed anticipatory displays can reduce the search required to identify a developing conflict. Researchers should manipulate concurrent demands and evaluate whether perceptual acquisition, causal comprehension, or projection changes. When performance is preserved through extra effort, workload measures and situation probes may diverge; this divergence is scientifically valuable because it reveals compensation and the remaining resilience of the operator Endsley (1995) Onnasch et al. (2014).

A complete process model includes information demand, current attention allocation, task-relevant representations, forecast confidence, control selection, and system feedback. Longitudinal measurements make it possible to study whether awareness drops before a failure or is reconstructed rapidly after corrective feedback. Such analyses should account for endogenous task difficulty: difficult events can simultaneously increase workload, lower situation-awareness scores, and worsen outcomes, producing associations that do not demonstrate a direct causal pathway. Prospective designs and randomized interface changes can help identify mediating mechanisms, but direct causal evidence requires temporal ordering and well-defined control conditions Bakdash et al. (2022) Yuan et al. (2026).

Mode Awareness, Automation Surprise, and Supervisory Control

An automated system changes its behavior according to internal modes, triggering conditions, available authority, and programmed transitions. Mode awareness is the operator's knowledge of which mode is active, what the mode does, how it can change, and what consequences will follow from intervention. Sarter and Woods Sarter & Woods (1995) analyzed automation-related mode errors in supervisory control, making clear that incorrect expectations about machine behavior can arise even when the interface is functioning as designed. The psychological event is not merely a missed icon: it is a failure to maintain a model linking machine state, human command, and expected system response.

Automation surprises are especially consequential during infrequent transitions. Routine automation may reduce action demands while also depriving the human of continuous feedback needed to maintain an accurate representation. When control must be recovered, the operator may need to recognize a transition, identify what the machine has already done, assess whether a previous plan remains valid, and choose an intervention within a narrow window. Bainbridge Bainbridge (1983) anticipated this paradox, and Onnasch and colleagues Onnasch et al. (2014) synthesized experimental evidence about performance tradeoffs associated with stages and levels of automation. Experimental tasks should measure not only nominal performance but also detection of automation failure, mode identification, recovery latency, and knowledge of constraints.

A design program for mode awareness exposes transition causes, authority boundaries, predicted effects of commands, and recovery options. Testing must evaluate how much of this additional information the operator can use under time pressure; an elaborate explanation shown after the critical decision may have little operational value. Alert design should distinguish a change in automation authority from a change in process state, while keeping their causal connections visible. The aim is calibrated supervision in which human attention and machine control are coordinated through interpretable feedback and recoverable action paths, rather than a simplistic maximum level of automation Parasuraman et al. (2000) Lee & See (2004).

Research Design and Operational Definitions

Empirical investigation of situation awareness begins by defining what the operator must know to perform a specified task at a specified time. Investigators establish a work domain, identify relevant state variables, model causal relations and future contingencies, and construct questions linked to those requirements. A general question such as whether a participant is aware of the situation is scientifically weaker than a targeted question about a changing separation, control authority, or expected threshold crossing. Operationalization should specify the unit of analysis, time window, expected correct answer, available information, and conditions under which an answer counts as adequate. The methodological account developed by Endsley Endsley (1995) connects situation-awareness queries to goal-directed task requirements rather than to indiscriminate memory testing.

A strong study distinguishes at least four elements: exposure to state information, accuracy of the resulting representation, performance of an action, and the subsequent state of the controlled system. These elements may diverge, particularly when actions are constrained by system latency or when machine automation compensates for a human mistake. Research designs include controlled simulations, within-subject interface comparisons, between-group training comparisons, retrospective analyses of operational events, and field observation. Randomized manipulations of displays or task demands can support causal inference about the intervention, while concurrent correlations between awareness scores and task outcomes primarily support association. The quality of the control condition, representativeness of scenarios, and independence of outcome scoring determine which conclusions are justified Bakdash et al. (2022) Yuan et al. (2026).

The research question determines the appropriate measure. If the goal is to determine whether a training program improves knowledge of future system states, future-state probes are central. If the goal is to understand coordination during handoffs, role-specific information flow and shared dependencies may be more informative. If the goal is to estimate detection of rare hazardous events, event-level hit and miss rates are indispensable. Measurement choice also determines the underlying concept of situation awareness: a subjective confidence report, a knowledge probe, and an eye movement trace do not represent interchangeable observations. The 2026 meta-review Moens et al. (2026) emphasizes conceptual alignment and the limited psychometric evidence available for many instruments.

Freeze-Probe Measurement and the SAGAT Tradition

The Situation Awareness Global Assessment Technique, or SAGAT, is based on interrupting a dynamic simulation at selected moments, withholding current task displays, and asking questions about the relevant state. Responses are assessed against an externally specified situation record. The technique permits a distinction between what the simulated system actually contained and what the operator knew about its content, relationships, and likely developments. A prompt might request the location of a hazard, the identity of a controlling machine mode, or the expected state of a parameter after a specified interval. Endsley's measurement framework Endsley (1995) established the rationale for probing knowledge at operationally meaningful moments rather than treating retrospective impressions as equivalent evidence.

Probe design requires a defensible specification of relevance and an adequate sampling of situations. If an assessment repeatedly asks for easily memorized numerical values, it may overrepresent one kind of factual recall and underestimate richer understanding. If it asks about obscure system details unrelated to decision demands, an expert may score poorly for rationally ignoring them. The researcher must therefore derive probes from task analysis, balance the three functional levels where appropriate, specify scoring rules, and examine item properties across participants and scenarios. Freeze points should be distributed to avoid predictable preparation and to sample relevant operating regimes. The measure is strongest when the probe bank has content validity and independent correctness criteria Endsley (2021) Moens et al. (2026).

An extensive synthesis by Endsley Endsley (2021) examined 243 studies of direct objective measures and reported that SAGAT showed sensitivity in 94% of the assessed studies, compared with 64% for SPAM and 73% for other real-time probes. These percentages describe evidence classifications in the reviewed studies, not the probability that an individual answer or a specific implementation is correct. The same review evaluated concerns about interruption and memory reliance and found support for SAGAT across many experimental settings. Its conclusions are important but remain situated within the reviewed designs; direct application to a particular domain requires an independently validated set of queries, a suitable simulation, and careful interpretation of how interruption may affect the phenomenon being studied.

Online Probes, Subjective Ratings, and Behavioral Indicators

The Situation Present Assessment Method, often abbreviated SPAM, asks operators task-related questions during an ongoing scenario rather than freezing the simulation. This method can preserve the continuity of an operational task, and response correctness may reveal whether relevant information is available when needed. It also introduces methodological complications: the probe itself competes for attention, response time may reflect task workload or willingness to interrupt an ongoing activity, and a participant may choose to delay an answer while prioritizing control. Consequently, probe latency is not a pure measure of situational knowledge. In the cross-study synthesis Endsley (2021), SPAM and SAGAT were comparably predictive of performance, but SPAM showed lower reported sensitivity and more frequent concerns about intrusiveness.

Subjective ratings ask participants or observers to evaluate awareness, understanding, or readiness. Such instruments can be logistically useful in field environments where freezing operations would be unacceptable. They may capture metacognitive confidence, perceived uncertainty, or professional judgments about information management. However, an individual can feel well informed while holding an inaccurate model, and a trained observer can infer awareness from behavior without direct access to hidden knowledge. The systematic comparison by Salmon and colleagues Salmon et al. (2009) found differences between freeze-probe and rating approaches in a military planning context, illustrating the danger of treating instrument names as guarantees of measurement equivalence. Results must be interpreted according to each measure's actual informational target.

Behavioral metrics include scan patterns, communication frequencies, control interventions, incident recognition, and the latency of corrective actions. These indicators reveal activity that can support the construction of awareness, but they are often multiply determined. A timely action could arise from correct understanding or from habitual response; an apparently delayed action could be deliberate risk management. Physiological and eye-tracking metrics can describe attentional and arousal dynamics without directly demonstrating comprehension or projection. The comprehensive 2026 meta-review Moens et al. (2026) classified 38 identified instruments across self-ratings, observer ratings, probing techniques, and physiological metrics, underlining that their complementary strengths depend on the question rather than on a universal ranking of technologies.

Psychometric Validity and Method Disagreement

The validity of a situation-awareness score depends on the inference made from it. Content validity asks whether probes cover information required by actual control goals. Construct validity asks whether the observations behave as a theory predicts while remaining distinguishable from adjacent constructs such as workload, confidence, or general experience. Criterion validity concerns associations with appropriate external outcomes. Reliability asks whether scores are sufficiently stable across equivalent items or conditions for their intended purpose, while responsiveness asks whether they detect consequential changes. A tool can be sensitive to a display manipulation and still measure only a narrow part of awareness. Instrument quality must therefore be reported at the level of the intended task and inference Moens et al. (2026).

Multiple operationalizations often yield different results because they observe different aspects of information processing. Freeze probes privilege explicit factual and predictive answers; ratings emphasize judgment; eye tracking observes sampling; communication analysis reveals information distribution; and task outcomes incorporate many determinants beyond knowledge. Disagreement among these measures can expose a flawed instrument, but it can also show that the underlying phenomenon is multidimensional or process dependent. The existence of different scores should motivate an explicit theory of which operations connect them. Cross-method convergence is informative only when the measures are intended to reflect the same latent property, with comparable time windows and reference objects Salmon et al. (2009) Stanton et al. (2017).

Moens and colleagues Moens et al. (2026) synthesized fourteen systematic reviews encompassing more than 477 distinct primary studies and identified 38 measurement tools. Their conclusion was that psychometric evidence remains limited for most measures, although probing techniques, especially SAGAT, have the most extensive supporting validity evidence in the review corpus. This finding supports a qualified preference for validated objective probes when knowledge accuracy is the primary target. It does not make probe scores universal standards for every team, longitudinal field process, or human–AI collaborative system. Researchers should publish their task analyses, item sets where possible, scoring rules, missing-data practices, and validity evidence so that measurements can be scrutinized and independently adapted.

Simulation, Field Observation, and Causal Identification

Simulations give researchers control over disturbances, information timing, machine modes, and outcome scoring. They can create comparable events for different operators and ethically test rare hazards that would be unacceptable in real systems. Their major limitation is representativeness: a simulation may reproduce physical state transitions while omitting organizational pressures, communication norms, or the costs of actual failure. The degree of fidelity required depends on the hypothesis. A study of visual grouping may require highly accurate displays and visual task demands, while a study of interprofessional handoffs may require realistic authority boundaries and communication structures. Experimental validity depends on reproducing the mechanisms under investigation, not simply maximizing cosmetic similarity Endsley (1995) Stanton et al. (2017).

Field observation captures real workflows, institutional conditions, and nonroutine disturbances, but it presents limits on randomization, interruption, and access to ground-truth states. Researchers may use logs, interviews tied to time-stamped events, communication recordings where ethically permitted, and observational coding of decisions and transitions. Retrospective accounts are especially vulnerable to hindsight, memory reconstruction, and the tendency to reinterpret early warnings after outcomes are known. A valid historical reconstruction should separate contemporaneous information available to the operator from information obtained in an investigation. In both field and simulation settings, incident narratives are best used to develop testable mechanisms rather than as proof that a particular mental state caused an outcome Sarter & Woods (1991) Bainbridge (1983).

Causal inference requires attention to confounding by task difficulty, operator expertise, environmental uncertainty, and automation quality. A positive correlation between awareness and performance might arise because easy situations permit accurate knowledge and effective action simultaneously. A manipulated display may improve both through parallel routes, without awareness mediating the full effect. Sequential experiments can help: first establish that an intervention changes a specified awareness component, then test whether the change predicts an independently measured control outcome while addressing alternative pathways. Multi-condition designs can vary workload and system complexity to evaluate moderation. Meta-analytic associations provide useful evidence but do not by themselves resolve the direction of causation Bakdash et al. (2022) Yuan et al. (2026).

Process Tracing and Multimodal Measurement

A dynamic state of awareness develops through a sequence of information requests, perceptual selections, hypotheses, communications, and interventions. Process tracing attempts to observe this sequence without collapsing it into a single end-of-task score. Researchers may align event logs, screen transitions, eye movements, response probes, control commands, and team communications along a common timestamp. A change in display fixation followed by a revised prediction and preventive action can support a temporal explanation, provided the intervening cognitive content is assessed rather than assumed. Such designs can identify when an information gap emerged, whether it was detected, and how rapidly the operating picture was repaired Endsley (1995) Stanton et al. (2017).

Multimodal measurement requires careful synchronization and error modeling. Sensor latency, inconsistent clocks, selective recording, and missing observations may make one modality appear to precede another when it did not. Researchers should define event windows before analysis, record the precision of each instrument, and guard against circular scoring when an intervention is evaluated using behavior already included in the awareness metric. Experience-sampling and observational coding are useful for some questions but can perturb the underlying process. Methods should be combined to triangulate explicit knowledge, information-gathering behavior, social coordination, and performance rather than to create an unvalidated composite score that obscures how each variable contributes Moens et al. (2026) Salmon et al. (2009).

For repeated real-world observations, hierarchical models may distinguish within-operator fluctuations from stable differences among operators, teams, and sites. A high average score at one institution may reflect less demanding cases rather than superior awareness. Equally, performance effects of a display may vary by event type, task phase, and expertise. Transparent reporting includes scenario sampling, instrument reliability, the number of independent operators and teams, clustering of observations, and intervals around effect estimates. Studies that publish reproducible task descriptions and deidentified scoring protocols enable stronger independent comparison across the diverse domains reviewed in recent syntheses Bakdash et al. (2022) Moens et al. (2026).

The Association Between Awareness and Performance

The assumption that better situation awareness improves control has motivated decades of research, yet the empirical relation is more modest and more variable than intuitive accounts often imply. Bakdash and colleagues Bakdash et al. (2022) meta-analyzed 678 effect sizes reported across 77 papers and found an overall association between situation-awareness measures and task performance of approximately r = .26. This positive average is compatible with practical significance in selected operations, but it also indicates that awareness measures explain only part of performance variation. The included methods differ in measurement, task demands, and outcome definition. The pooled association should not be translated into a universal causal coefficient or used to predict how much performance will improve when a particular interface is changed.

An additional meta-analysis published in 2026 synthesized 87 studies and 265 effect sizes, finding a moderate positive association between situation awareness and task performance and substantial moderation of antecedent relationships by study characteristics Yuan et al. (2026). Its findings reinforce the importance of task context in interpreting awareness scores. Measurement differences can matter because an operator's knowledge of state, a colleague's assessment of that knowledge, and the operator's confidence are not equivalent predictors. A study may also inflate an apparent association when an awareness instrument includes variables too similar to its performance criterion. The scientific task is to examine whether task-specific awareness predicts independently assessed outcomes after accounting for common influences such as expertise, scenario complexity, and workload.

The strongest performance argument is therefore conditional. Correct awareness can make an effective decision possible; it does not ensure that the chosen action will be executed, accepted by the machine, or beneficial under unpredictable disturbances. Some environments allow automation or procedures to deliver correct outcomes despite inadequate human understanding. Others depend heavily on early detection and forecast. Research should report the relevant decision opportunities, rate of irreversible system change, available alternatives, and scoring independence. Comparing knowledge accuracy with recovery behavior across matched contingencies offers a more precise account than assuming that all successful performance is mediated by a single broad psychological construct Bakdash et al. (2022) Onnasch et al. (2014) Kaber & Endsley (2004).

Automation Allocation and the Routine–Failure Tradeoff

Automated assistance may support information acquisition, analysis, selection of actions, and execution. Parasuraman, Sheridan, and Wickens Parasuraman et al. (2000) distinguished the types and degrees of automation across these processing stages, creating a conceptual tool for examining why different allocations produce different human consequences. Information displays that summarize sensor relations may improve awareness, whereas systems that independently select and execute actions can reduce operator involvement and opportunities to update a working model. The appropriate level of automation depends on task variability, reliability of machine advice, recovery demands, and the distribution of authority between human and machine. An automated controller's nominal accuracy is therefore an incomplete measure of human-system quality.

Kaber and Endsley Kaber & Endsley (2004) examined level of automation and adaptive automation in a dynamic control task. Their reported findings distinguished benefits for primary performance from benefits for situation awareness and workload: lower-level automation produced stronger primary performance in the studied configuration, intermediate levels facilitated higher situation awareness without a corresponding improvement in primary performance, and adaptive-cycle characteristics influenced other workload and monitoring outcomes. This is a particularly informative experiment because the findings caution against treating awareness, workload, and output as a single success dimension. The conclusions apply to the tested paradigm, automation implementations, and participant conditions; replication across other work domains is required before deriving universal design rules.

The meta-analysis of eighteen experiments by Onnasch and colleagues Onnasch et al. (2014) found a broad tradeoff. Greater degrees of automation generally favored routine performance while showing disadvantages for situation awareness and performance when automation failed. The synthesis identified especially consequential shifts when automation progressed from supporting information analysis toward selecting actions. These findings create a plausible mechanism by which efficient routine automation can leave operators with less capacity to diagnose sudden exceptions. For design and evaluation, they motivate paired tests: performance while automation works as expected and resilience when its assumptions, sensors, or control policies fail. A system that excels only under nominal conditions leaves the full human-system control problem unexamined.

Aviation, Flight Decks, and Air Traffic Operations

Aviation supplied many influential examples because pilots and controllers must integrate distributed data under strict temporal and spatial constraints. Position, trajectory, aircraft configuration, weather, clearance, and traffic separation become meaningful only through their relation to a flight plan and the changing environment. A pilot can recognize the current altitude yet misunderstand the active guidance mode, while an air traffic controller can track a target without anticipating a convergence conflict. These examples illustrate why perception, comprehension, and projection should be scored separately. Theoretical treatments of dynamic aviation awareness Endsley (1995) and investigations of supervisory mode errors Sarter & Woods (1995) made this domain central to the study of human-machine coordination.

Modern displays change which facts are immediately available and which must be inferred. Highly capable avionics can reduce manual control activity while creating new obligations to recognize transitions between managed and selected modes, interpret automation limits, and intervene during unanticipated behavior. An experimental aviation study should therefore manipulate mode transparency and alert timing while independently measuring mode identification, future-trajectory questions, control response, and recovery after a failure. The same principle applies to air traffic interfaces where a conflict predictor can focus attention but also introduce dependence on algorithmic forecasts. If an operator stops checking the underlying traffic geometry, a predictor failure may have consequences that routine accuracy metrics conceal Bainbridge (1983) Parasuraman et al. (2000).

Aviation evidence also illustrates the difficulty of generalizing from laboratory tasks to operational performance. Pilots are subject to training standards, crew procedures, regulatory constraints, fatigue, and institutional communication routines that a simplified simulation may omit. Historical incidents can identify plausible pathways of mode confusion, but precise psychological attribution requires contemporaneous records and careful separation of known states from later reconstructions. Robust evaluation should use scenario families varying novelty, workload, and information availability, with expert review of what an operator could reasonably have inferred at each decision point. The research result should specify the operating conditions under which an intervention improved knowledge or control, rather than assert a universal aviation effect Sarter & Woods (1995) Stanton et al. (2017).

Industrial Process Control and Networked Infrastructure

Industrial control presents a distinctive problem: important relationships are often hidden inside distributed processes whose state is represented by remote sensors and symbolic displays. Operators monitor pressure, flows, demand, component states, or network topology, sometimes across geographical distance. A correct observation of one variable may mislead if the configuration of the wider system has changed. Displays that expose functional dependencies can support diagnosis by making relationships perceptible; ecological interface design Vicente & Rasmussen (1992) is relevant precisely because local readings do not fully describe a complex work domain. Evaluation should include rare but physically plausible disturbances and assess whether the operator recognizes constraints before a threshold is violated.

A naturalistic study of energy distribution by Salmon and colleagues Salmon et al. (2008) documented how relevant information was distributed among interacting agents in a real operational domain. The investigators represented what information elements were used and how they were shared, supporting a system-level account of situation awareness. The study is valuable because it examines actual coordination arrangements instead of assuming that the operating picture resides in one controller. Its case-study design is suited to describing structures and generating hypotheses, while broader causal estimates would require comparisons across situations or organizations. The findings motivate mapping which role has access to which state information, when information is transmitted, and what happens when a communication link fails.

An industrial research program can combine simulated process faults with field studies of organizational response. In simulation, investigators control the location and timing of disturbances, the transparency of display information, and automation policies. In field studies, logs and structured observations reveal actual handoffs, workarounds, and coordination pathways. Both designs should distinguish a failure of human knowledge from an inaccessible or inaccurate system state. A control room with no reliable telemetry cannot be judged solely by operator awareness scores; observability is a property of the work system. This distinction joins cognitive explanation to engineering requirements for sensor integrity, information provenance, and resilient communication Salmon et al. (2008) Stanton et al. (2017).

Clinical Monitoring, Emergency Coordination, and High-Risk Teams

Clinical and emergency work often requires several professionals to monitor evolving conditions while sharing responsibility for decisions. Patient deterioration, equipment status, resource availability, and treatment timing may be visible to different people and only partially represented in a common record. Situation awareness in this setting concerns timely access to relevant indicators, interpretation of changing risk, anticipation of likely consequences, and coordinated handoff of the information necessary for action. The distributed perspective Stanton et al. (2017) is useful because a team may be effective when specialized members hold complementary knowledge and communicate dependencies accurately. An isolated awareness score may miss whether a critical update reached the person empowered to intervene.

Clinical awareness can be studied through representative simulations, case-based projections, observational coding, and time-stamped communication analysis. A freeze-probe technique may be suitable in simulation but inappropriate during patient care; observer ratings and behavioral measures can be more feasible in natural settings while carrying different validity limitations. The contemporary meta-review Moens et al. (2026) stresses that suitability depends on the intended construct, context, and operational constraints. Investigators should also assess whether clinicians understand the limitations of monitors and clinical decision-support outputs, because a display can accurately reproduce a device reading without making the physiological cause or urgency of the situation clear.

Research design must respect the difference between awareness and clinical expertise or outcome. A physician may accurately anticipate deterioration yet face a resource shortage; another team may obtain a good outcome through an established protocol despite imperfect recognition of a causal process. Studies should therefore define the relevant informational state, clinical action opportunity, independent outcome metric, and confounding clinical severity. Educational interventions may focus on cross-checks, forecast-oriented briefing, and timely escalation, with effectiveness evaluated using task-specific evidence rather than a general assertion that good teamwork always produces better awareness. Clinical recommendations require qualified professional review when they move beyond research synthesis into practice directives Moens et al. (2026) Stanton et al. (2017).

Road Transport, Remote Systems, and Variable Autonomy

Road vehicles and remotely controlled systems illustrate how changing machine authority affects an operator's knowledge of the environment. A driver supervising an automated assistance feature must understand what the system currently controls, which obstacles it can detect, when its performance may deteriorate, and how quickly manual intervention could become necessary. A remote operator faces additional constraints from camera coverage, sensor fusion, network latency, and the reduced availability of peripheral cues. The same situation-awareness construct spans these cases, but its operational questions differ: immediate object detection, interpretation of automation mode, and anticipation of a safe takeover window may have distinct temporal requirements Parasuraman et al. (2000) Sarter & Woods (1995).

The human-system research problem is especially acute when routine automatic operation creates long periods of low engagement followed by abrupt requests for action. A nominally effective controller may reduce manual work while weakening the operator's opportunities to maintain an accurate forecast of hazards. The automation literature Bainbridge (1983) Onnasch et al. (2014) establishes a general basis for examining such tradeoffs, but domain-specific experiments must measure realistic takeover requirements instead of directly importing results from aviation or industrial simulations. Key factors include the operator's monitoring task, available warning time, road or workspace complexity, reliability of machine alerts, and feedback that conveys current control authority.

Evaluation should distinguish readiness to intervene from demonstrated control capability. A participant may correctly explain the upcoming hazard but be unable to execute a safe maneuver because of an inadequate transition interval. Conversely, a safety system may complete a maneuver without the human knowing why it worked. Studies can manipulate alert timing, mode displays, and changes in the environment, then collect pre-intervention state probes, forecast accuracy, response time, and objectively scored system outcomes. This design treats awareness as one experimentally analyzable contributor to safe control and gives designers evidence about when additional information, better authority handoff, or more automation is needed Kaber & Endsley (2004) Bakdash et al. (2022).

Interface Design for Understanding and Anticipatory Control

An interface supports situation awareness when it makes information necessary for correct interpretation and timely action accessible at the right moment. The design question is not how much information can be displayed, but which state variables, functional relationships, uncertainty estimates, and action consequences an operator must establish. Ecological interface design Vicente & Rasmussen (1992) supplies a principled approach to exposing constraints of the work domain, while the three-level account Endsley (1995) differentiates whether a display supports perception, comprehension, or projection. For a dynamic process, a useful visualization may show current values, their rates of change, operating boundaries, and projected consequences of plausible interventions. These representations should permit operators to test and revise interpretations without forcing them to infer an entire causal structure from scattered indicators.

Design also involves temporal visibility. A warning that arrives after a critical threshold is crossed supplies retrospective information but may not preserve a preventive action window. A machine mode indicator can tell the operator what is active without explaining why a transition occurred or what it means for command authority. Better feedback can associate the current mode with its controlling assumptions, transition triggers, and implications for expected system behavior. Such transparency needs selective presentation because indiscriminate explanatory detail can compete with the operator's immediate task. Studies should compare interpretable warnings, state histories, and future-state visualizations under matched decision requirements, measuring both information acquisition and controlled outcomes Sarter & Woods (1995) Parasuraman et al. (2000).

User-centered evaluation should include unfamiliar disturbances and circumstances in which machine recommendations are unreliable. Ordinary task completion can fail to reveal whether a display leaves its users dependent on familiar scripts. Appropriate outcome measures include anomaly recognition, causally correct diagnosis, calibrated uncertainty, projection accuracy, recovery latency, and resilience of control after an unexpected transition. A successful intervention may improve one component without changing another. The Kaber and Endsley automation experiment Kaber & Endsley (2004) and the broader tradeoff synthesis Onnasch et al. (2014) illustrate why the combination of nominal productivity and self-reported ease is insufficient evidence of better awareness.

Training, Briefings, and Coordinated Recovery

Situation-awareness training should target the information skills that a specific operational domain requires. These may include disciplined scan patterns, identifying system constraints, recognizing early anomalies, checking mode transitions, articulating competing explanations, and projecting near-term hazards. Teaching a generic slogan about being alert offers little basis for measurement. Training objectives become testable when defined through a task analysis: participants should be able to state which variables are critical, explain why a change matters, forecast what would happen without intervention, and identify evidence that could falsify their initial hypothesis. Simulation exercises can vary scenario familiarity and decision timing to distinguish rehearsed performance from transferable understanding Endsley (1995) Vicente & Rasmussen (1992).

Team training introduces information-handling and coordination skills. A briefing can identify which actor monitors which variable, who has authority to intervene, and which signals require immediate escalation. A handoff can make explicit the current state, a causal interpretation, the most likely developing threats, uncertainties, and the next decision boundary. Distributed situation-awareness research Salmon et al. (2008) Stanton et al. (2017) suggests that the value of such practices lies in maintaining the right information at relevant nodes of the system. Effectiveness requires testing not merely communication volume or subjective teamwork but whether critical information arrives and is understood before the opportunity for action disappears.

Evidence of training benefit must be separated from familiarity with the assessment itself. Participants who repeatedly answer the same freeze probes may learn the question bank without improving their ability to detect a novel fault. More persuasive designs use blinded scoring, alternate but equivalent scenarios, delayed retention tests, and transfer to disturbances outside the training set. For team interventions, analytic units should account for clustering by crew or work group. When an intervention targets a clinical setting, its implications for patient care require domain-specific clinical expertise and appropriate review. The present monograph analyzes research methods and does not issue an operational or clinical protocol Moens et al. (2026) Bakdash et al. (2022).

Artificial Systems and Human–Artificial Situation Models

In the Artificial Era, dynamic control increasingly includes machine agents that classify conditions, predict trajectories, recommend actions, and sometimes act autonomously. This changes the distribution of information and control authority. A human may know the state of an external process while misunderstanding the machine's confidence, scope, or intent; a model may generate useful forecasts while its human partner cannot identify which inputs or constraints underpin them. Situation awareness in human–artificial systems should therefore examine at least three distinct objects: the world state, the artificial controller's operative state, and the relationship between the human's expectations and the machine's behavior. The automation taxonomy Parasuraman et al. (2000) and the literature on calibrated reliance Lee & See (2004) provide foundations for differentiating these objects.

Artificial systems can hold and update representations at speeds or scales unavailable to a human supervisor, but system-level coordination depends on the accessibility, fidelity, and usability of these representations. A useful interface may expose the machine's current control authority, relevant observations, anticipated events, known limitations, and rationale at a level appropriate for human decision-making. Such design demands empirical validation. A fluent explanation is not evidence that the underlying prediction is correct, and a correct forecast does not guarantee that its recipient will understand its uncertainty. Studies can compare different forms of machine transparency, evaluate the operator's mode and projection accuracy, and introduce failures that require human challenge or takeover Sarter & Woods (1995) Onnasch et al. (2014).

The disciplinary expansion From Homo to Artificial creates further research questions about awareness as an accomplishment of heterogeneous human–artificial arrangements. These questions can be formulated without presupposing that computational representations are identical to human psychological experience. Operational tests might measure whether distributed state estimates remain mutually compatible, whether machine-generated alerts preserve causal provenance, how information loss propagates through a partnership, and whether authority transfers degrade shared prediction. Distributed cognition Hutchins (1995) and distributed situation awareness Stanton et al. (2017) supply methods for evaluating these configurations as systems. The philosophical horizon motivates inquiry; evidence about actual machine capacities, human comprehension, and performance must come from observable task results.

Causal Explanation and Construct-Validity Debates

Situation awareness is often invoked as an explanation for operational success or failure, yet its scientific value depends on specifying causal pathways rather than retrospectively naming a desired state. The classic critique by Sarter and Woods Sarter & Woods (1991) warned against conceptual vagueness, and later meta-analytic work Bakdash et al. (2022) quantified the limits of empirical prediction. A theory should explain why a certain informational deficit occurs and how that deficit changes the set of actions available to the operator. A mere statement that poor awareness caused a poor decision may conceal missing feedback, inadequate automation design, insufficient time, conflicting goals, or constrained organizational authority.

Different schools offer partially overlapping explanations. The three-level approach organizes the accuracy of task-relevant knowledge; ecological accounts emphasize active perception of constraints; distributed accounts analyze how representations and information responsibilities are organized across people and artifacts. These approaches need not be forced into a single composite index. Their comparative predictions differ: an environmental representation might support successful action with limited factual recall, while a disconnected operator may depend more strongly on retained knowledge; a team may coordinate well with specialized and nonidentical information stores. Suitable experiments should manipulate accessibility of external information and communication while measuring internal answers, actions, and network-level state propagation Smith & Hancock (1995) Hutchins (1995) Salmon et al. (2008).

The distinction between antecedent, mediator, moderator, and outcome is central. Expertise and interface quality can influence awareness; workload and task novelty may moderate that influence; awareness may mediate the relation between information design and some decisions; and system performance emerges after additional processes of action selection and control execution. Reverse causation is also possible when successful actions make later states more predictable. Meta-analyses of awareness and performance Bakdash et al. (2022) Yuan et al. (2026) justify continued investigation of the construct but do not establish a simple universal causal chain. Future theory should specify conditions under which awareness is necessary, sufficient only in conjunction with other capacities, or dispensable because the system is designed to act safely without conscious supervision.

Reproducibility, Heterogeneity, and Limits of the Evidence

The research corpus differs substantially in operational domain, expertise level, simulation fidelity, available time, and the instruments used to score awareness. These differences constrain direct comparison. A freeze-probe study of student operators and an observational study of distributed energy controllers may both invoke situation awareness while measuring different processes at different analytical levels. The meta-analysis by Bakdash and colleagues Bakdash et al. (2022) provides a useful quantitative summary, yet its modest average correlation should be understood within this methodological heterogeneity. The 2026 synthesis of antecedents Yuan et al. (2026) further suggests that task characteristics are important moderators rather than background details to be ignored.

Publication and measurement practices also affect the evidence base. Researchers may preferentially report successful interface interventions, omit unsuccessful probe designs, or analyze repeated event observations as though they were independent participants. Even when statistical significance is established, effect magnitudes and uncertainty intervals are necessary to judge whether an improvement is operationally important. A careful report should identify sample characteristics, task sampling, scoring rules, preregistered hypotheses when applicable, data exclusions, and the level of independence in the analysis. Effects should be tested across both normal and disrupted operating conditions. Where field replication is unavailable, the limits of transfer should remain explicit Moens et al. (2026) Onnasch et al. (2014).

A second limitation concerns the measurement of teams and novel human–artificial systems. Many commonly used instruments were developed for an individual who can answer questions about a bounded simulated environment. Networked settings distribute knowledge among actors and tools, and their informational state changes through communication and delegation. A measure of one person's answers may overlook whether the relevant system as a whole can recognize and correct an emerging problem. The meta-review by Moens and colleagues Moens et al. (2026) documents uneven psychometric support across tools and strongly supports matching measurement to the research purpose. Independent validations, shared task scenarios, and transparent instrument development are essential for accumulating results that can be compared across settings.

Research Agenda for Dynamic Human-System Control

A mature research program should begin with a common conceptual map distinguishing world state, available information, operator knowledge, projected outcomes, machine control mode, and executed actions. It should then test competing mechanisms at the level of specific questions and controlled conditions. A first priority is temporal measurement: how early can a developing hazard be detected, how accurately can its course be projected, and how does that knowledge decay after display interruption or unexpected automation transitions? Studies should use event-triggered probes and independent outcome records with sufficient precision to establish the order of information acquisition, interpretation, prediction, and action Endsley (1995) Sarter & Woods (1995).

A second priority is theory-comparative intervention. Research can contrast memory-dependent displays with persistent constraint visualizations, individual probes with distributed information-network analysis, and status-only automation cues with explanations of mode authority and transitions. Each manipulation should have predeclared predictions for awareness components and independently measured control outcomes. Experiments could also vary the reliability and transparency of machine recommendations, testing whether operators adjust reliance appropriately as conditions change. The relevant dependent variables include calibrated uncertainty, anomaly detection, causal comprehension, forecast quality, transition management, and recovery performance Vicente & Rasmussen (1992) Lee & See (2004) Onnasch et al. (2014).

A third priority is cumulative evidence infrastructure. Researchers need reusable task analyses, validated and openly described probe banks, representative scenario families, and reporting conventions that preserve the distinction between individual, team, and system-level constructs. Replication across expertise levels and real-world contexts should test whether relationships found in one domain survive differences in authority, culture, and technological configuration. Meta-analysis should continue to examine sensitivity to instrument choice and overlapping effect sizes within studies. The strongest scientific advances will arise when awareness is treated as an explicitly specified relation between informational states and achievable control rather than as a catchall explanation for desirable outcomes Bakdash et al. (2022) Moens et al. (2026) Yuan et al. (2026).

Conclusion

Situation awareness in dynamic human-system control is the temporally organized acquisition, interpretation, and projection of information needed for effective action in changing technical environments. Its scientific scope encompasses individual cognition, perceptual coupling with system displays, coordination among specialized human roles, and information distribution across automated components. Endsley's three-level formulation Endsley (1995) remains a useful operational architecture for distinguishing what an operator detects, understands, and anticipates. Ecological and distributed perspectives extend analysis to environmental constraints and collaborative information networks, while automation studies show how machine function allocation shapes both ordinary operation and recovery from exceptions Vicente & Rasmussen (1992) Hutchins (1995) Onnasch et al. (2014).

The empirical record supports the construct with important qualifications. Direct knowledge probes permit focused investigation, and a large systematic review found extensive evidence for SAGAT sensitivity across evaluated contexts Endsley (2021). A recent meta-review nevertheless documents uneven psychometric validation across the many tools available for different purposes Moens et al. (2026). Quantitative synthesis identifies a positive but heterogeneous awareness–performance relationship, including an overall mean correlation near .26 in one major meta-analysis Bakdash et al. (2022). These findings justify the study of awareness as one contributor to control while preventing its reduction to a universal surrogate for safety, expertise, or system performance. The decisive scientific requirement is to identify exactly which informational relation was measured and what inference the design can support.

The practical consequence is a research and design agenda centered on interpretable system states, timely anticipation, calibrated machine assistance, role-appropriate information distribution, and recoverability when conditions depart from expectation. In the Artificial Era, these requirements extend to relations between human operators and artificial controllers whose representations, forecasts, and authority boundaries must be made operationally intelligible. Future progress depends on comparative experiments, reliable instruments, representative disturbances, transparent reporting, and studies that connect local cognitive processes to system-level coordination. The fundamental scientific object remains stable across changing technologies: the relation between what an operating system is doing, what its participants can know, and what they can still do about its future Stanton et al. (2017) Yuan et al. (2026).

References

  • Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), 775–779. https://doi.org/10.1016/0005-1098(83)90046-8
  • Bakdash, J. Z., Marusich, L. R., Cox, K. R., Geuss, M. N., Zaroukian, E. G., & Morris, K. M. (2022). The validity of situation awareness for performance: A meta-analysis. Theoretical Issues in Ergonomics Science, 23(2), 221–244. https://doi.org/10.1080/1463922X.2021.1921310
  • Endsley, M. R. (1995). Measurement of situation awareness in dynamic systems. Human Factors, 37(1), 65–84. https://doi.org/10.1518/001872095779049499
  • Endsley, M. R. (1995). Toward a theory of situation awareness in dynamic systems. Human Factors, 37(1), 32–64. https://doi.org/10.1518/001872095779049543
  • Endsley, M. R. (2021). A systematic review and meta-analysis of direct objective measures of situation awareness: A comparison of SAGAT and SPAM. Human Factors, 63(1), 124–150. https://doi.org/10.1177/0018720819875376
  • Hart, S. G., & Staveland, L. E. (1988). Development of NASA-TLX (Task Load Index): Results of empirical and theoretical research. In P. A. Hancock & N. Meshkati (Eds.), Human mental workload (pp. 139–183). Elsevier. https://doi.org/10.1016/S0166-4115(08)62386-9
  • Hutchins, E. (1995). Cognition in the wild. MIT Press. https://doi.org/10.7551/mitpress/1881.001.0001
  • Kaber, D. B., & Endsley, M. R. (2004). The effects of level of automation and adaptive automation on human performance, situation awareness and workload in a dynamic control task. Theoretical Issues in Ergonomics Science, 5(2), 113–153. https://doi.org/10.1080/1463922021000054335
  • Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80. https://doi.org/10.1518/hfes.46.1.50_30392
  • Moens, L. L., Lydon, S., Cucurachi, S., O’Connor, P., Sauter, T. C., Töndury, G.-A., & Manser, T. (2026). Measuring situation awareness: A meta-review across domains. Human Factors, 68(5), 632–672. https://doi.org/10.1177/00187208251412110
  • Onnasch, L., Wickens, C. D., Li, H., & Manzey, D. (2014). Human performance consequences of stages and levels of automation: An integrated meta-analysis. Human Factors, 56(3), 476–488. https://doi.org/10.1177/0018720813501549
  • Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics—Part A: Systems and Humans, 30(3), 286–297. https://doi.org/10.1109/3468.844354
  • Salmon, P. M., Stanton, N. A., Walker, G. H., Jenkins, D., Baber, C., & McMaster, R. (2008). Representing situation awareness in collaborative systems: A case study in the energy distribution domain. Ergonomics, 51(3), 367–384. https://doi.org/10.1080/00140130701636512
  • Salmon, P. M., Stanton, N. A., Walker, G. H., Jenkins, D. P., Ladva, D., Rafferty, L., & Young, M. (2009). Measuring situation awareness in complex systems: Comparison of measures study. International Journal of Industrial Ergonomics, 39(3), 490–500. https://doi.org/10.1016/j.ergon.2008.10.010
  • Sarter, N. B., & Woods, D. D. (1991). Situation awareness: A critical but ill-defined phenomenon. The International Journal of Aviation Psychology, 1(1), 45–57. https://doi.org/10.1207/s15327108ijap0101_4
  • Sarter, N. B., & Woods, D. D. (1995). How in the world did we ever get into that mode? Mode error and awareness in supervisory control. Human Factors, 37(1), 5–19. https://doi.org/10.1518/001872095779049516
  • Smith, K., & Hancock, P. A. (1995). Situation awareness is adaptive, externally directed consciousness. Human Factors, 37(1), 137–148. https://doi.org/10.1518/001872095779049444
  • Stanton, N. A., Chambers, P. R. G., & Piggott, J. (2001). Situational awareness and safety. Safety Science, 39(3), 189–204. https://doi.org/10.1016/S0925-7535(01)00010-8
  • Stanton, N. A., Salmon, P. M., Walker, G. H., Salas, E., & Hancock, P. A. (2017). State-of-science: Situation awareness in individuals, teams and systems. Ergonomics, 60(4), 449–466. https://doi.org/10.1080/00140139.2017.1278796
  • Vicente, K. J., & Rasmussen, J. (1992). Ecological interface design: Theoretical foundations. IEEE Transactions on Systems, Man, and Cybernetics, 22(4), 589–606. https://doi.org/10.1109/21.156574
  • Yuan, X., Zhao, Y., & Jia, Q. (2026). Antecedents and outcome of situation awareness: A meta-analysis of presence and effect size. Ergonomics, advance online publication, 1–67. https://doi.org/10.1080/00140139.2026.2648584