Sensemaking Debt

How Unrecovered Interpretive Obligations Accumulate in Adaptive Systems

Jan Windahl and Dennis Johansson

FlowSentience Research Paper No. 1

FlowSentience Research Programme

Gothenburg, Sweden

Version 1.3

Manuscript revision: 11 September 2026
https://doi.org/10.5281/zenodo.22709413

flowsentience.org

PUBLICATION AND RIGHTS INFORMATION

Title
Sensemaking Debt: How Unrecovered Interpretive Obligations Accumulate in Adaptive Systems

Series
FlowSentience Research Paper No. 1

Authors
Jan Windahl
Dennis Johansson

Affiliation
FlowSentience Research Programme, Gothenburg, Sweden

Corresponding author
Jan Windahl - jan.windahl@flowsentience.org

Version
Version 1.3

Manuscript revision
11 September 2026
DOI: https://doi.org/10.5281/zenodo.22709413

Publisher
Zenodo

Suggested citation
Windahl, J., & Johansson, D. (2026). Sensemaking debt: How unrecovered interpretive obligations accumulate in adaptive systems (Version 1.3; FlowSentience Research Paper No. 1). Zenodo. https://doi.org/10.5281/zenodo.22709413

Copyright and licence
Copyright 2026 Jan Windahl and Dennis Johansson. This work is licensed under the Creative Commons Attribution 4.0 International Licence (CC BY 4.0): https://creativecommons.org/licenses/by/4.0/

ABOUT THE AUTHORS

Jan Windahl

Jan Windahl is a Swedish systems thinker, transformation leader and independent researcher working at the intersection of collective sensemaking, organisational flow and AI-assisted organisational sensing. His professional background spans complex technology programmes, public-sector digitalisation, agile portfolio governance and value-stream development across defence, transport infrastructure, healthcare, regional government and industrial systems. As co-founder of the FlowSentience Research Programme, he leads research into how organisations and societies perceive, interpret and respond to complexity while preserving human judgement, collective understanding and adaptive capacity. He is co-author of The Sentient Value Stream Framework and The Sentience Threshold.

Dennis Johansson

Dennis Johansson is a Swedish transformation practitioner, systems thinker and independent researcher whose work focuses on organisational development, operational excellence and value-stream thinking in complex environments. His research and practice examine how coordination, learning and decision-making can be strengthened across organisational boundaries through operational experience, systems thinking, digital capability and human-centred change. As co-founder of the FlowSentience Research Programme, he contributes to research on how organisations can detect friction, integrate weak signals and adapt before local difficulties develop into systemic failure. He is co-author of The Sentient Value Stream Framework and The Sentience Threshold.

The FlowSentience Research Programme

Together, Windahl and Johansson develop the FlowSentience Research Programme, an independent inquiry into collective understanding, systemic listening and the conditions under which increasingly complex human and human-AI systems remain adaptive, humane and governable. Further information and publications are available at flowsentience.org.

Abstract

Organisations can have extensive information and still struggle to understand how their decisions affect one another. This paper proposes Sensemaking Debt as a way to investigate what happens when the work needed to resolve consequential differences is repeatedly postponed and begins to restrict later action. The proposed condition is latent: it must be inferred from evidence about organisational history. It is path-dependent: earlier responses influence what can be understood and changed later. A momentary Sensemaking Gap is distinguished from this accumulated condition.

The argument develops sufficiently shared understanding as the connection among different perspectives needed for coordinated action. It does not require consensus or identical expertise. State, epistemic and intent continuity describe the capacity to reconstruct how an arrangement emerged, what evidence and uncertainty informed it, and which purpose it served. A sensemaking perimeter follows consequential relationships across formal organisational boundaries.

Artificial intelligence can increase both the need for interpretation and the ability to renew it. Recursive self-improvement sharpens this relationship when improvements also strengthen the system's ability to produce further improvements. Changes may then affect the activity being governed and the means used to evaluate it. The report distinguishes demonstrated, bounded self-modification from speculation about unrestricted acceleration. It proposes that improvement should be examined through the continuity and capacity for correction that it preserves, alongside its immediate performance.

A separate valuation extension compares the consequences of earlier, delayed and absent action. Climate change illustrates why an organisation's business case can differ from the value across institutions and generations. An educational extension asks whether repeatedly displaced learning opportunities can constrain later understanding; PISA is contextual evidence, not a measure of the proposed debt. Neither extension validates the organisational construct. Longitudinal comparisons must establish whether it explains accumulated constraints better than existing theories, including cases where AI or organisational change restores understanding.

Keywords: Sensemaking Debt; collective understanding; adaptive systems; organisational continuity; recursive self-improvement; intertemporal valuation; human-AI systems; capability formation

Contents

Preface

This inquiry grew from our work with value streams and organisational coordination. A value stream follows the connected activities through which a need is met, often across several departments or organisations. In healthcare, public administration and complex technology programmes, we encountered situations in which better access to information coexisted with greater difficulty in understanding the whole course of events. A new system could improve one team's decisions while making the effects on another team harder to recognise. These experiences prompted the inquiry; they are not presented as a systematic empirical study.

The pattern could not be explained by size alone. Some large organisations coordinated through stable relationships and clearly understood interfaces. Much smaller organisations struggled when different professions used the same words differently or relied on changing technologies. Better analysis could make a local situation clearer while adding recommendations and dependencies that people elsewhere still had to interpret.

Established research explains much of how people and organisations notice, interpret and respond to change. Our question concerns what happens when these activities work locally but repeatedly fail to connect. We became interested in the history left behind: assumptions that remain in use, arrangements whose purposes become difficult to recover, and later decisions that depend on both.

Sensemaking Debt is a provisional name for the accumulated condition that may result. The report develops this possibility carefully enough for it to be criticised and tested. Its value depends on whether it helps explain differences between organisational histories that existing accounts leave insufficiently clear.

The argument begins with familiar situations and gradually introduces the terms needed to examine them. The chapters on AI, climate consequences and learning apply those terms to different settings. Readers can follow the main argument without working through the equations: the valuation chapter explains their purpose in words, and Appendix A contains the formal detail. Appendix B brings the central concepts together with their meanings and scientific status.

Version 1.3 adds a focused examination of recursive self-improvement, checked on 11 September 2026, and revises the explanations throughout the manuscript. The wider research update retains its 9 September 2026 cutoff and is not a systematic review. The PISA discussion keeps the limits stated in that update: the publication was checked, but the detailed new tables were not directly verified. No new national rankings or effect estimates are inferred from initial reporting. Appendix C distinguishes the evidence consulted from the hypotheses developed here.

Introduction

A hospital knows its occupied beds, a municipal service its available staff, and a transport provider its schedule. Safe discharge depends on whether these accurate accounts fit together for the person returning home. This illustrative example makes the question concrete: what enables people to connect what they know when their actions depend on one another?

Suppose the hospital records a patient as ready for discharge because inpatient treatment is complete. The receiving service reads the same word as meaning that support at home has already been arranged. The information has travelled, but an important difference in meaning has not. A conversation can expose the difference and allow the services to coordinate. If it remains unnoticed, each service may act reasonably within its own account while their combined actions leave a need unmet.

Sensemaking is the work of developing and revising an account of what is happening and what it means for action. An interpretation connects an observation to a meaning: a full ward might indicate an unusual surge, a discharge blockage or a persistent shortage of capacity. A representation is the form in which some part of the situation is made available, such as a record, model or verbal explanation. Sharing a representation does not establish that its recipients interpret it alike.

Established research provides a rich foundation for examining this distinction. Bounded rationality, cybernetics, organisational sensemaking, learning and situation-awareness research explain how adaptive systems select information, regulate variety, construct meaning and revise action under uncertainty (Simon, 1957; Ashby, 1956; Beer, 1972, 1979; Weick, 1995; Argyris & Schön, 1978; Endsley, 1995). Organisational information-processing theory similarly treats design as an attempt to match uncertainty and interdependence with adequate structures for processing and coordinating information (Galbraith, 1974; Tushman & Nadler, 1978).

The present report asks what may happen when these processes keep working locally while their connections weaken. In the discharge example, an experienced coordinator might repeatedly reconcile the meanings of 'ready'. The next patient can move safely, yet the systems and responsibilities may remain ambiguous. Later staff inherit both the ambiguity and a dependence on that person's knowledge. Current performance gives only a partial view of what the organisation has preserved for the future.

Organisations create interpretive work through their own development. A shared platform may solve one problem while introducing categories and dependencies whose effects people must understand. An improvement becomes part of the situation that the next improvement has to address.

A small group spanning clinical care, transport and home support may need to connect very different judgments. A larger group performing independent tasks may need less coordination. Effective sensemaking scale names the number and diversity of perspectives whose differences matter for the action. Their interdependence matters alongside their number.

The relevant adaptive system may also extend beyond the organisation's legal boundary. Suppliers, regulators, professional communities, platforms, outsourced functions and technical systems can all influence the representations upon which consequential action depends. Organisational knowledge, competence and authority are themselves shaped through ambiguity, identity and socially negotiated claims rather than existing as entirely neutral or self-evident properties (Alvesson, 1993, 2001; Alvesson & Willmott, 2002).

We call the boundary around these consequential relationships the sensemaking perimeter. For a safe discharge, it may include the patient, relatives and home-support service as well as the hospital. For another outcome, a different boundary may be appropriate. Choosing a perimeter makes explicit whose knowledge and consequences need consideration; it does not require everyone within it to know everything or attend the same meeting.

Artificial intelligence intensifies this recursive development. AI does not simply help an organisation interpret an already existing reality. It introduces predictions, summaries, classifications, recommendations and increasingly automated actions that become part of the reality requiring further interpretation. Learning algorithms can reshape workflows, distributions of expertise and organisational authority rather than merely automate unchanged tasks (Faraj et al., 2018). The relationship between automation and human augmentation is therefore not a simple choice between substitution and support; each can create further requirements for the other (Raisch & Krakowski, 2021).

Recursive self-improvement adds a further possibility. A system may improve the tools or processes through which it produces its next improvement. The thing people need to understand is then changing partly through its own development. Later chapters distinguish bounded demonstrations of this process from claims about unrestricted acceleration, and examine how both useful understanding and unresolved obligations might develop alongside it.

The opposite effect is also possible. AI may support the synthesis of distributed observations, translation between professional domains, retrieval of organisational memory and earlier recognition of conflicting assumptions. Emerging work on Sensemaking AI therefore directs attention beyond isolated model performance towards the capacity of human-AI networks to interpret and adapt collectively (Comes, 2026). Evaluation consequently needs to follow the relationships created by AI-supported decisions and ask whether the wider system becomes better able to understand and coordinate them.

Interpretive demand means the work a situation requires to connect relevant meanings and keep them current. Renewal capacity means the ability to do that work as people, technologies and circumstances change. In the discharge example, a new scheduling system might increase demand. Time to compare its assumptions with actual care arrangements, supported by accessible records and working relationships, may strengthen renewal.

Sensemaking Debt is proposed when consequential interpretive obligations are repeatedly carried forward and begin restricting later understanding and action. An interpretive obligation is work that remains necessary because somebody's decision depends on it: establishing what 'ready' means, for example, or clarifying who checks that support is available. It need not be a formally assigned duty. Carrying it forward means that later work inherits the unresolved issue. The term debt concerns the resulting constraint, not every unanswered question.

The condition is described as latent because it cannot be read directly from one indicator. Its existence must be inferred from a history of unresolved obligations and later difficulty. It is path-dependent because the route taken matters: resolving the discharge ambiguity early leaves different possibilities from embedding it in several connected systems. Whether these histories form a distinct organisational condition remains an open research question.

A consequential pattern may become visible only when observations from several organisations are considered together. A hospital sees occupancy and a municipality sees its staffing budget; the combined shortage may fall outside each local decision. Aggregation means connecting such partial observations to reveal a wider consequence.

Climate change illustrates the difficulty across institutions and generations. Scientific evidence can establish a wider trajectory while costs, benefits and authority remain divided. An organisation may face present expenditure while much of the avoided harm lies beyond its budget or planning horizon.

This raises a further question. What difference does earlier understanding make if it enables timely action? The answer requires comparing plausible paths through time: what could happen with earlier intervention and what could happen if action is delayed or absent. That comparison values consequences. It is separate from establishing whether an organisation carries Sensemaking Debt.

Understanding also has to be formed. Institutions may record completed work while leaving some learning unfinished. The proposed developmental understanding debt concerns how those missed opportunities affect later capability. Learning conditions and institutional responsibility belong in the explanation. The theory has not measured such a debt's direction or prevalence among young people.

Existing Theories Explain More Than We Think

The theories considered here address different parts of the same practical difficulty. Some explain limits on attention; others examine how meaning develops, how organisations learn and how authority shapes what can be questioned. They give the proposed debt concept a foundation and provide alternatives against which it must be tested.

Limits on attention and coordination

Simon's concept of bounded rationality provides a natural starting point. Organisational actors cannot consider every alternative, consequence or piece of information relevant to a decision. They operate within constraints of cognition, attention, time and available knowledge, relying upon selective representations and satisfactory rather than theoretically optimal solutions (Simon, 1957). Organisational action therefore always depends upon simplification. The relevant issue is not whether a system can understand reality completely, but whether the simplifications guiding action remain sufficiently appropriate as circumstances change.

Ashby approached a related problem through cybernetics. His Law of Requisite Variety proposed that effective regulation requires sufficient internal variety to respond to the disturbances produced by the environment (Ashby, 1956). The central insight is relational: a system remains viable only when its capacity to distinguish and respond remains sufficiently matched to the conditions it must regulate. Stafford Beer extended this logic into an organisational account of viability in which local autonomy, feedback and recursive coordination allow distributed activity to retain systemic coherence (Beer, 1972, 1979). Viability cannot be secured through central control alone; it depends upon relationships through which relevant variety becomes available where it can influence action.

Conant and Ashby’s good-regulator theorem (1970) provides a related formal reference point for the role of models in regulation. Its assumptions concern a specified regulatory problem; it does not prove that a human organisation must possess one complete shared representation. The analogy used here is correspondingly limited.

Organisational information-processing theory translated this correspondence into the language of design. Galbraith argued that uncertainty creates information-processing requirements to which organisations respond through structures, lateral relationships and decision mechanisms (Galbraith, 1974). Tushman and Nadler similarly connected task uncertainty and interdependence with the capacity of organisational arrangements to collect and coordinate information (Tushman & Nadler, 1978). These theories already contain an important demand-capacity relationship. Their primary concern, however, lies with information requirements and processing arrangements rather than with the accumulated condition that may arise when information continues circulating but the interpretations needed for coordinated action become progressively harder to connect and renew.

How meaning and learning develop

Weick shifted attention from information processing towards the construction of meaning itself. Organisations do not first discover an objective environment and then decide how to act. They enact situations, select cues, construct plausible interpretations and revise meaning through continuing action and interaction (Weick, 1995). Sensemaking is therefore social, provisional and ongoing. Its purpose is not to produce a final representation of reality, but to sustain sufficiently coherent action under conditions that remain ambiguous. This has an important consequence for the present argument. Collective understanding cannot be treated as knowledge that, once established, remains available without maintenance. Participants change, environments evolve, technologies alter the categories through which work is represented, and action modifies the reality that subsequent action must address. Understanding must therefore be continually reconstructed.

Organisational-learning research explains several processes through which such reconstruction occurs. Argyris and Schön distinguished between correcting action within existing assumptions and reconsidering the assumptions themselves (Argyris & Schön, 1978). Their distinction between single-loop and double-loop learning demonstrates that visible errors may sometimes be resolved without revising the categories and goals through which the organisation defines the situation.

Senge extended this discussion through systems thinking, shared vision and the learning organisation (Senge, 1990), while Nonaka and Takeuchi described knowledge creation as a social movement between tacit experience, articulation, combination and practical application (Nonaka & Takeuchi, 1995). Organisations do not merely accumulate information; they continually reproduce and revise the relationships through which information becomes meaningful.

Endsley's theory of situation awareness adds a more structured description of this interpretive requirement. Effective action depends upon perceiving relevant elements, comprehending their meaning and projecting how the situation may develop (Endsley, 1995). In organisations, these processes are usually distributed. Different actors and technologies perceive different elements, while comprehension and anticipation depend upon connections among specialised perspectives. The availability of signals is therefore only the beginning. Those signals must become related to one another and interpreted in terms of emerging consequences.

Common-ground research examines the continuing work required for coordination in joint activity (Klein et al., 2005). Carlile (2004) further distinguishes transfer, translation and transformation across increasingly complex knowledge boundaries. These accounts already make the maintenance and revision of interdependent understanding explicit. Sensemaking Debt would need to add an identifiable accumulation–recovery relationship beyond those established processes.

These accounts explain how people can receive the same information and still lack the understanding needed to work together. Their activities may remain effective within each profession while the connections among professions require further attention.

Automation and the organisation of knowledge

Automation research adds a further dimension. Bainbridge's ironies of automation showed that automation does not simply eliminate human work. It often removes people from routine activity while leaving them responsible for exceptional intervention precisely when the situation is most unfamiliar and difficult to understand (Bainbridge, 1983). Cognitive and coordinative burdens may therefore be transferred rather than removed.

Parasuraman, Sheridan and Wickens developed this insight by distinguishing among automation of information acquisition, information analysis, decision selection and action implementation (Parasuraman et al., 2000). Different forms and levels of automation change where interpretation occurs, who remains involved and where responsibility for consequences resides. The organisational effect is not merely a reduction in human effort. Automation changes the relationships through which the system understands and governs its own behaviour.

Contemporary artificial intelligence extends these dynamics. Learning algorithms do not simply execute predefined work more efficiently; they can reshape classifications, workflows, knowledge practices and distributions of expertise (Faraj et al., 2018). Raisch and Krakowski describe the resulting relationship as an automation-augmentation paradox: AI may replace human activity while simultaneously increasing the need for complementary judgement, supervision and coordination (Raisch & Krakowski, 2021). Automation and augmentation are therefore mutually entangled rather than stable alternatives.

Comes' emerging agenda for Sensemaking AI moves the analysis from isolated model performance towards distributed human-AI networks (Comes, 2026). The relevant question becomes whether interactions among human and artificial actors support the continuing construction of coherent understanding under uncertain and changing conditions. This does not constitute empirical support for Sensemaking Debt, but it reinforces the importance of studying intelligence and adaptation at the level of the wider socio-technical system.

Critical organisation studies introduce a further complication. The organisation itself cannot always be treated as a clear and objective container within which knowledge is produced and coordinated. Alvesson's research on knowledge-intensive work shows that competence, performance and valid knowledge are often shaped by ambiguity, rhetoric, professional identity and organisational image (Alvesson, 1993, 2001). What an organisation claims to know, who is regarded as competent and what counts as successful performance are partly socially negotiated.

Alvesson and Willmott similarly show how identity regulation shapes the ways individuals understand their roles, responsibilities and legitimate interpretations (Alvesson & Willmott, 2002). Shared narratives can enable coordination, but they can also exclude observations that challenge prevailing identities or assumptions. Apparent agreement may consequently coexist with unresolved interpretive divergence.

These insights affect both the content and the boundary of collective understanding. Knowledge and interpretive authority may be distributed across professions, suppliers, contractors, regulators, platforms and external communities. Technologies and AI systems may shape organisational representations while formally remaining classified as tools. The adaptive system whose understanding matters may therefore extend beyond the legal organisation and vary according to the consequences being coordinated.

Boundary inquiry also has a normative history. Ulrich (1983) treats judgments about the relevant system and affected interests as questions open to criticism. A sensemaking perimeter should likewise expose who is included, whose knowledge counts and who bears consequences, rather than conceal those choices behind an apparently neutral system boundary.

Together, these traditions explain why information, meaning and authority cannot be separated from the relationships through which work proceeds. They provide the basis for asking how those relationships change over time.

The contribution proposed here concerns what repeated failures of connection leave behind. Earlier theories are not being credited retrospectively with the debt construct. They help establish what a separate account of accumulation would need to explain.

An inherited rule whose purpose cannot be reconstructed illustrates the temporal question. A disagreement between a hospital and a receiving service illustrates the relational one. The debt hypothesis connects them when an unresolved difference becomes embedded in an arrangement that later participants struggle to revise. Its value depends on whether following that history explains more than examining the present disagreement alone.

When knowledge becomes action

The attention-based view of the firm clarifies why an aggregate consequence can remain organisationally absent even when relevant information exists. Ocasio argues that what decision-makers do depends upon how organisational rules, resources, relationships and communication channels distribute issues and answers into particular situations (Ocasio, 1997). A signal outside those channels has little practical effect. Repetition or scientific confidence cannot compensate for an architecture that does not allocate attention, responsibility or decision rights to the consequence.

Rasmussen's account of risk in dynamic societies develops the same problem across levels. Structural decomposition encourages each discipline and organisational tier to optimise its own representation, while the behaviour of the wider risk-control system emerges through interactions among legislators, regulators, managers, practitioners and technologies (Rasmussen, 1997). For consequences that cross these divisions, the functional control structure is a more defensible analytical unit than the isolated organisation. Ostrom's work on polycentric governance adds that distributed centres of action can support experimentation and learning, but only when information, incentives and coordination mechanisms connect action across scales (Ostrom, 2010).

Decision analysis contributes a further distinction between knowing and implementation. Value-of-information approaches estimate the expected benefit of reducing uncertainty, yet often assume that the resulting recommendation will be adopted immediately and completely. Andronis and Barton show that this assumption can materially overstate value and propose implementation-adjusted estimation instead (Andronis & Barton, 2016). Organisational readiness research similarly separates collective commitment from collective efficacy: actors may understand that change matters while lacking shared resolve, confidence, resources or situational capacity to implement it (Weiner, 2009).

These approaches connect interpretation to delayed consequences without pricing understanding itself. They help distinguish the history that makes a decision difficult from the benefits and losses that follow from its timing. The later valuation chapter examines how attention, understanding and practical readiness influence that timing.

Complexity-oriented sensemaking offers a further qualification: the mode of inquiry must fit the situation, including circumstances in which causal patterns cannot be specified in advance (Kurtz & Snowden, 2003; Snowden & Boone, 2007; Snowden, 2011). Dynamic-capabilities research similarly concerns the integration and reconfiguration of competence in changing environments (Teece et al., 1997). The present construct must remain compatible with discovery and transformation rather than presume that every unresolved difference can be closed against a known target.

The distinction between the production of an acceptable result and the development of a capacity to interpret it also connects the argument to learning research. Cognitive load theory helps explain why some forms of assistance can make learning more accessible, while the design of a task determines what intellectual work remains with the learner (Sweller, 1988). The relevant organisational question is how institutions recognise and renew that capacity over time. A good answer can coexist with incomplete learning, just as a functioning organisational arrangement can coexist with an unrecoverable rationale. This structural resemblance motivates inquiry across levels without making the individual and organisational constructs identical.

What Makes a System Adaptive?

An adaptive system can develop, retain or recombine ways of responding as its circumstances change. In this report, viability means continuing to function under those changing conditions. A hospital that learns how to reorganise care during a new demand pattern illustrates the organisational meaning. Responding successfully once is insufficient if the response leaves the hospital unable to deal with the next change. The broader history below shows how experience can become available beyond the individual who acquired it.

Living systems introduced this capacity through biological evolution. Natural selection acted upon inherited variation across generations, favouring characteristics that supported survival and reproduction under prevailing conditions. Evolution did not optimise organisms towards perfection or a final state. It continually reshaped populations as organisms and environments changed in relation to one another.

As nervous systems developed, biological evolution was supplemented by adaptation within the lifetime of the individual. Learning allowed experience to modify later behaviour without waiting for genetic change across generations. It did not replace natural selection; it added a faster adaptive layer. Organisms could retain information about previous encounters, distinguish among circumstances and modify their actions accordingly.

Human beings extended this development by making learning increasingly collective. Language, imitation, teaching, symbols and material artefacts allowed experience acquired by one person to influence others. Culture enabled interpretations and practices to survive beyond the lifetime of those who first developed them. Adaptive knowledge no longer had to be rediscovered independently by every individual or encoded biologically before it could persist.

This changed the effective unit of adaptation. Once experience could be distributed, preserved and recombined socially, groups acquired capabilities that no member possessed alone. Coordinated roles, cumulative traditions and shared representations allowed human communities to address conditions whose complexity exceeded individual cognition. Collective learning therefore did more than increase the amount of available knowledge. It connected partial knowledge across people and time.

Organisational-learning research illuminates two aspects of this transition. March's distinction between exploration and exploitation shows that adaptive systems must use established knowledge while preserving the capacity to investigate alternatives and revise earlier practices (March, 1991). Nonaka and Takeuchi similarly describe organisational knowledge as developing through movement between tacit experience, articulation, combination and practical application rather than through the accumulation of explicit information alone (Nonaka & Takeuchi, 1995). In both accounts, adaptive capacity depends upon whether knowledge can be retained, connected and revised.

The expansion of collective adaptation is not unlimited, however. Stable social relationships require attention, time and repeated interaction. Dunbar's work on primate sociality linked neocortical capacity with typical group size and later became associated with the widely cited estimate of approximately 150 stable human relationships (Dunbar, 1992). This value should not be treated as a universal organisational threshold. Later analysis has challenged both its statistical precision and the confidence with which one exact limit can be inferred (Lindenfors et al., 2021).

Its relevance here is narrower. Human relational capacity is finite. People may participate in organisations containing thousands or millions of members, but they cannot maintain equally rich, reciprocal and contextually informed relationships with everyone involved. Coordination at larger scales therefore depends upon roles, routines, institutions, technologies and representations that allow action to remain connected without requiring universal personal familiarity.

Effective sensemaking scale concerns the number, heterogeneity and consequential interdependence of interpretations that must remain connected for coordinated action. Heterogeneity means differences in knowledge, experience or ways of judging a situation. In the discharge example, clinical readiness and readiness at home are different judgments. Their connection matters because one service's action changes the conditions facing the other. This is why adding people is not the same as increasing the scale of the sensemaking task.

The concept does not imply that all interpretations within the effective sensemaking scale must become identical. Specialisation is one of the principal sources of collective capability. The adaptive challenge is to preserve enough connectivity among specialised perspectives for consequential differences to be recognised and addressed when coordinated action depends upon them.

Institutions enabled human systems to extend this coordination beyond the limits of direct relationships. Rules, professions, offices, legal arrangements, markets and shared symbolic structures stabilised expectations among people who did not know one another personally. They preserved practices across turnover and allowed knowledge to remain available after the individuals who originally developed it had departed.

Organisations represent a particularly important form of this institutional development. Their capabilities arise through distributed specialisation, coordinated action and organisational memory. No individual fully understands a hospital, an infrastructure system, a multinational enterprise or a public administration. Nevertheless, these systems perform activities beyond the knowledge and capacity of every individual participant.

This distributed structure is both an adaptive achievement and a source of vulnerability. It allows knowledge to remain specialised and action to occur at scale, but it also means that the relationships among partial perspectives must continually be reconstructed. Local expertise can remain strong while understanding of the wider system weakens. Individual units can improve while the consequences of their combined development become progressively more difficult to interpret.

At societal and civilisation scales, the same principle extends through science, education, economic exchange, governance, infrastructure and communication systems. Collective adaptation no longer resides in one institution or central intelligence. It emerges through networks that preserve and combine observations, memories and practices across organisations, populations and generations. Each expansion of the adaptive unit therefore produces two related effects. It enlarges what the system can perceive, remember and accomplish. It also enlarges the relational and interpretive structure through which those capabilities must remain sufficiently connected.

Figure 1 presents this development as a continuum rather than as a sequence in which each new mechanism replaces the previous one. Biological evolution continues. Individual learning remains essential. Culture, institutions and organisations add further layers through which experience can be preserved and coordinated across broader scales. The adaptive unit expands, while the mechanisms operating at earlier levels remain active within it.

Collective understanding forms part of the infrastructure through which specialised and interdependent actors can coordinate. Its contribution lies in keeping consequential differences in knowledge and interpretation sufficiently connected for action and revision. This leaves room for partial knowledge and specialisation, on which collective capability itself depends.

Human-AI systems may now extend the adaptive unit further by adding machine-generated representations and partially autonomous actors to the networks through which collective perception, interpretation and action occur (Seeber et al., 2020). This development does not remove the need for collective understanding. It changes the system across which such understanding must be constructed and renewed. As the adaptive unit expands, the maintenance question becomes more demanding: how can distributed and changing knowledge remain connected enough for consequentially interdependent action?

What Must Adaptive Systems Actually Maintain?

Distributed expertise allows people to accomplish more together than any one participant could manage. The maintenance question is how to preserve the connections that make this possible when the participants and their circumstances change.

Current performance gives only part of the answer. A service can meet its targets while depending increasingly on one person to reconstruct missing context. Another service may temporarily perform worse while learning a more dependable way to coordinate. The difference concerns what present activity leaves available for later adaptation.

A more fundamental requirement lies in the nature of collective action. Adaptive systems must coordinate under conditions of incomplete information, distributed expertise and changing relationships. Such coordination does not require every participant to possess the same knowledge or interpretation. It requires sufficiently compatible understandings of those aspects of reality upon which consequentially interdependent action depends. A surgical team, for example, does not require identical expertise across professions. Its members contribute different forms of knowledge, but they must share enough understanding of the patient's condition, the stage of the procedure, emerging risks and immediate responsibilities for their differentiated actions to remain coherent. The same principle applies across larger systems. Participants need not understand the whole, but they must remain sufficiently connected to recognise when changes in one domain alter assumptions or consequences elsewhere.

Collective understanding concerns whether participants can recognise how their different accounts bear on a shared consequence. It can remain strong even when they disagree about what to do. The requirement is that the disagreement can be examined and taken into account in coordinated action.

This is consistent with Weick's account of sensemaking as an ongoing social accomplishment rather than the discovery of a final representation of reality (Weick, 1995). Organisations construct workable interpretations through action, communication and retrospective reflection. These interpretations need not be complete or permanently shared, but they must remain sufficiently plausible for action and sufficiently open to revision when experience challenges them.

Endsley's theory of situation awareness makes a related distinction between perceiving relevant elements, comprehending their meaning and projecting how a situation may develop (Endsley, 1995). In collective systems, these functions are distributed across people, professions and technologies. One actor may detect a signal, another understand its technical significance and a third recognise its wider consequences. Collective awareness therefore depends not simply upon individual cognition, but upon relationships that connect partial perception, comprehension and anticipation.

Sufficiently shared understanding is the context-appropriate degree and form of interpretive connectivity required for consequentially interdependent action to remain coherent, contestable and adaptive.

Interpretive connectivity means that relevant differences in meaning can be recognised and worked through. In the discharge example, the services need to know that their uses of 'ready' differ and what this means for the next step. Coherent action fits these dependencies. Contestable action remains open to reasoned challenge. Adaptive action can be revised as circumstances change. These are the practical meanings of the terms in the definition.

How much connection is sufficient depends on the action. Independent and readily reversible tasks can tolerate considerable variation. A time-sensitive handover may need rapid agreement about a small set of critical facts and responsibilities. The question is what participants need to recognise for that action to remain workable and revisable.

Figure 2 presents this relationship as a recurrent socio-technical cycle rather than as a stock of knowledge held somewhere within the organisation. Changing reality is selectively represented by humans and technologies. These distributed representations are compared, translated, challenged and revised across consequential relationships. Sufficiently shared understanding supports coordinated action, and that action changes the reality that the next cycle must interpret.

Every transformation involves selection and framing. Observations foreground some relationships while omitting others, and professional or technological representations simplify reality in different ways. Such transformation is unavoidable and not inherently harmful. It becomes consequential when relevant differences become disconnected, suppressed or carried forward without sufficient renewal.

Disagreement and the boundary of understanding

This definition deliberately avoids equating collective understanding with consensus. Disagreement can strengthen adaptation by exposing weak signals, preserving alternative explanations and challenging established assumptions. High-reliability research similarly emphasises sensitivity to operational detail and reluctance to simplify uncertain situations prematurely (Weick & Sutcliffe, 2007).

Connected difference exists when people can identify where their interpretations vary and examine why. Concealed divergence occurs when a common word or summary hides a consequential difference. Premature convergence occurs when an apparent agreement closes inquiry before a relevant uncertainty has been examined. Both can create a Sensemaking Gap, even in a group that appears well aligned.

A group can therefore disagree extensively while maintaining strong collective understanding if its differences remain visible, intelligible and connected to shared consequences. Conversely, an organisation may appear highly aligned while its actors rely upon incompatible assumptions that have never been surfaced. Visible disagreement is not necessarily evidence of weak understanding, just as visible consensus is not reliable evidence of its strength.

This issue becomes particularly important where authority, professional identity and organisational legitimacy affect which interpretations may be expressed. Alvesson's work on knowledge-intensive organisations shows that claims concerning knowledge, competence and performance are often shaped by ambiguity, rhetoric and identity rather than resting upon entirely unambiguous evidence (Alvesson, 1993, 2001). Alvesson and Willmott similarly show how organisational narratives regulate the identities through which individuals interpret appropriate roles and responsibilities (Alvesson & Willmott, 2002).

Shared narratives can enable coordination, but they can also narrow the range of observations regarded as legitimate. When coherence becomes something that must be displayed rather than continually examined, apparent agreement may conceal unresolved divergence. Alvesson and Spicer's concept of functional stupidity captures a related condition in which reduced reflexivity and justification support short-term functioning while restricting the capacity to question governing assumptions (Alvesson & Spicer, 2012).

Organisational analysis often begins with the legal organisation as its unit of study. Yet many consequential processes extend across formal boundaries. Care may depend upon hospitals, primary care, municipal services, patients, relatives, regulators and digital infrastructures. Production may depend upon suppliers, software platforms, logistics providers and certification bodies. Public action may depend upon institutions, scientific communities, private organisations and citizens who remain formally outside the responsible authority. The relevant boundary for collective understanding must therefore be drawn around the actors and representations that materially affect the consequences being coordinated. This report refers to that provisional analytical boundary as the sensemaking perimeter.

A perimeter follows the consequence under examination and can change as dependencies change. The people needed to coordinate today's discharge may differ from those needed to plan regional care capacity. Several perimeters can therefore overlap. Choosing one establishes an analytical boundary, not a new legal organisation or a requirement for universal agreement.

This boundary-sensitive approach also reflects Alvesson's challenge to the idea that organisations and their knowledge are entirely self-evident objects. The organisation chart offers one representation of where authority and knowledge reside, but practical interpretive influence may lie with informal experts, professional communities, suppliers, platforms or external institutions that do not appear within formal reporting relationships.

Technologies increasingly belong within the same analysis. Digital systems determine which categories become visible, how activities are represented and which actions become available. Workflow platforms encode sequences and dependencies. Data models establish distinctions through which performance and risk are interpreted. Technologies do not possess collective understanding in the human sense, but they participate materially in constructing the representations through which organisations understand themselves.

As AI-generated analyses, summaries and recommendations become embedded in organisational work, machine-produced representations participate increasingly in the construction of shared reality. A common model output should not, however, be mistaken for shared understanding. Fluent synthesis may coordinate attention while concealing unresolved differences in evidence, assumptions and responsibility.

Human-AI interaction research emphasises that users must be able to understand and contest the capabilities, limitations and changing behaviour of AI systems (Amershi et al., 2019). Emerging work on Sensemaking AI broadens this concern towards the quality of interpretation across distributed human-AI networks (Comes, 2026). The adaptive requirement remains a connected process through which both human and machine-generated claims can be examined, challenged and revised.

The sensemaking perimeter also clarifies why more communication does not necessarily produce greater understanding. Additional meetings, reports and dashboards can strengthen coordination when they reveal consequential relationships and support revision. They can also reproduce existing interpretations, multiply competing representations or become recurring mechanisms for compensating for differences that remain unresolved. Renewal depends on signals crossing relevant boundaries and on differences becoming visible while action can still change their consequences. Organisational memory connects current decisions with the assumptions that produced the present situation. Actors also need room to question the categories through which a conclusion appears reasonable, including categories embedded in technical systems.

Renewal is selective. People need to reconnect the knowledge that matters for the action at hand, while leaving room for specialised understanding elsewhere. Because their actions alter the situation, those connections may need to be examined again.

The next question concerns the history available for that examination. Knowing what people believe now may not explain why an inherited arrangement exists or when it should change.

Keeping organisational history usable

Three forms of organisational continuity help specify what renewal must preserve. State continuity is the capacity to reconstruct how the present configuration, dependencies, roles and constraints emerged. Epistemic continuity is the capacity to reconstruct what was known, assumed, uncertain or contested when earlier decisions were formed. Intent continuity is the capacity to reconstruct the purpose or protected value an inherited arrangement was created to serve.

A discharge rule makes the distinction concrete. State continuity lets participants reconstruct where the rule was introduced and which services now rely on it. Epistemic continuity preserves the evidence and uncertainty that informed its adoption; epistemic here simply means concerned with what was known and how. Intent continuity preserves the purpose, such as avoiding a gap in care. A record can retain one of these while losing another. The three forms are proposed distinctions, not established independent measurements.

Continuity also depends on succession. A recorded rationale has limited adaptive value if later participants have never developed the domain knowledge, language or practical experience required to interpret it. The renewal of understanding therefore includes both reconstructing inherited context and forming the capability to engage with that context. Mentoring, supervised participation and progressive responsibility belong within the analysis when they sustain consequential interpretive obligations across generations of participants.

The Principle of Adaptive Balance develops this relationship between renewal capacity and the interpretive demands confronting the system. It provides the basis for examining what may accumulate when renewal repeatedly falls behind.

The Adaptive Balance

A new scheduling platform may help a hospital use its capacity while changing what 'available' means for receiving services. The improvement creates a further task of interpretation. This is the recurring condition behind adaptive balance: the system needs to understand a world that its own earlier responses have helped to change.

Interpretive demand is the differentiation, connection and revision of meanings required for coordinated action. Differentiation matters when two apparently similar cases need different responses; connection matters when different services affect the same outcome; revision matters when an earlier interpretation no longer fits. More data may contribute to this work, but data volume does not measure how difficult it is.

The effective sensemaking scale helps locate the demand. Stable interfaces can keep it manageable in a large system. A small system may face substantial demand if its few participants must continually reconcile different judgments. The relevant question is which relationships require attention and what follows if they are misunderstood.

Organisational information-processing theory anticipated part of this relationship. Galbraith connected uncertainty with increasing information-processing requirements, while Tushman and Nadler examined how organisational structures and integrating mechanisms respond to such demands (Galbraith, 1974; Tushman & Nadler, 1978). The argument developed here extends this logic from information processing to collective interpretation. Circulating information acquires significance through reconstruction across actors whose decisions and actions affect one another.

Sensemaking renewal capacity is the ability to reconstruct sufficiently shared understanding as circumstances, actors and relationships change. It depends on practices that make differences examinable and relevant history accessible. A common record can support renewal when people can connect it to their responsibilities and question its assumptions. Simply adding another record may increase the work instead.

Dialogue, shared language and access to earlier reasoning can make renewal easier. Stable relationships reduce the effort of establishing context each time a difference arises. Modularity means arranging work in relatively independent parts with understood interfaces; it can reduce the number of dependencies needing continuous attention. Slack means time or resources not already committed to routine activity. It gives people room to investigate a surprise. These arrangements help only if important changes remain visible across their boundaries.

Resilience engineering similarly emphasises that adaptive capacity involves the ability to adjust performance before, during and after disturbances rather than simply returning to an earlier state (Hollnagel et al., 2006; Woods, 2015). Such adjustment requires the capacity to recognise when established interpretations no longer correspond adequately with operational reality. Sensemaking renewal is therefore not an activity performed only after failure. It forms part of the continuing work through which viable performance is sustained.

The relationship between interpretive demand and sensemaking renewal capacity is what we call the Principle of Adaptive Balance:

For an adaptive system, keeping the renewal of context-appropriate interpretive connectivity and continuity sufficiently responsive to emerging obligations is proposed as one enabling condition of continued viability.

Balance describes responsiveness, not numerical equality. Demand and renewal capacity have no established common units. The proposal is that a system needs enough ability to update its understanding for the consequences it must coordinate. What counts as enough changes with the situation.

This formulation identifies a proposed enabling condition. It does not establish a sufficient condition for survival, a universal law, or a numerical equilibrium. Material resources, ecological conditions, power, legitimacy and external shocks can determine outcomes even when interpretive renewal is strong. Renewal itself includes the formation of new capability as well as the repair of inherited discontinuity. For this reason, dynamic correspondence is more precise than dynamic equilibrium.

A system may tolerate substantial ambiguity where activities are loosely coupled and consequences reversible. It requires stronger and more rapidly renewed understanding where actions are tightly interdependent, time-sensitive or capable of producing cascading effects. The required correspondence therefore varies with the situation being coordinated.

In the discharge example, a short joint review may resolve the meaning of a new category and prevent repeated calls. Another review process may generate minutes without changing the ambiguity. Similar effort can therefore leave very different conditions for the next handover. Adaptive balance concerns this relationship between the task generated and the capacity renewed.

The relationship is recursive in an ordinary organisational sense: a response changes the circumstances to which later responses must attend. A new coordination forum may reveal a dependency while creating another set of decisions to interpret. Technical recursive self-improvement is a more specific case, examined below, in which an AI system also changes its ability to produce subsequent improvements.

AI may increase interpretive demand by generating additional signals, analyses, recommendations and actions, and by creating new relationships among data, models, professional judgement and organisational authority. As AI becomes embedded in workflows, organisations must understand not only their environment and existing structure, but also the representations and interventions produced by increasingly capable technical systems.

At the same time, AI may strengthen sensemaking renewal. It can synthesise distributed observations, translate between specialised domains, retrieve organisational memory, compare competing representations and make inconsistencies more visible. It may help distributed actors recognise relationships whose scale and complexity exceed unaided human attention. AI's contribution therefore needs to be examined across time: does it help collective interpretation remain responsive to the new obligations its use generates?

Its effect cannot be presumed to be neutral or universally beneficial. It depends upon how AI is designed and embedded, the degree of autonomy it receives, the transparency and contestability of its outputs, the participation of relevant actors and the wider system's ability to integrate machine-generated representations. AI that accelerates local decisions while weakening the relationships through which consequences are interpreted may reduce adaptive correspondence. AI that helps actors compare assumptions, detect contradictions and revise shared representations may strengthen it.

Bainbridge's ironies of automation remain relevant because automation can remove people from routine activity while leaving them responsible for exceptional situations in which understanding is most difficult (Bainbridge, 1983). Raisch and Krakowski's automation-augmentation paradox similarly shows that replacement and human enhancement are interdependent rather than independent alternatives (Raisch & Krakowski, 2021). Comes' work on Sensemaking AI extends the question towards the adaptive capacity of wider human-AI networks rather than isolated model performance (Comes, 2026).

Decision acceleration and sensemaking acceleration describe different effects of an intervention. The former makes representations become actions sooner. The latter helps people recognise, connect and revise consequential interpretations sooner. One system can do both; the analysis asks what its use leaves understandable across the relevant perimeter.

The Principle of Adaptive Balance remains a theoretical proposition rather than an empirically calibrated law. Its purpose is to connect insights from cybernetics, organisational information-processing theory, resilience engineering and human-technology research around a common relationship. Ashby's requisite variety and Beer's viable systems emphasise correspondence between environmental demands and regulatory capacity (Ashby, 1956; Beer, 1972, 1979). Organisational-design research connects uncertainty with processing requirements, while resilience engineering emphasises continuing adjustment under changing conditions. The present principle directs these insights towards the renewal of collective understanding.

Temporary imbalance is normal. New situations can expose useful disagreement and prompt learning. The concern begins when later activity repeatedly inherits unresolved obligations and those obligations make renewal harder. Adaptive balance directs attention to recovery before current compensation becomes a lasting constraint.

This makes the fate of a temporary gap important. A discrepancy that prompts effective inquiry leaves a different history from one whose obligations must repeatedly be carried forward.

The Emergence of Sensemaking Debt

New circumstances routinely exceed existing understanding. A service changes its procedure, a colleague leaves or an unfamiliar case crosses a boundary. People may need to act while their interpretations are provisional. The question is what happens to the differences that remain after the immediate situation has been handled.

Their significance becomes clearer in what happens afterwards. Some discrepancies are recognised and recovered quickly. Actors compare observations, revise assumptions and reconstruct enough common orientation for interdependent action to continue. The system may experience temporary confusion without carrying a lasting constraint into later activity. Other discrepancies are managed without being fully recovered. Informal relationships compensate for gaps in formal coordination. Experienced individuals translate between incompatible perspectives. Additional meetings provide enough alignment for the next decision, while the assumptions producing the difference remain largely unchanged. Local workarounds preserve performance, but part of the interpretive discrepancy persists.

A Sensemaking Gap is a context-specific mismatch, at a particular point in time, between the form and degree of interpretive connectivity required for consequentially interdependent action and what the relevant adaptive system can currently sustain.

The gap describes a situation at a particular time. For a discharge, the services may lack a sufficiently connected account of readiness even though both possess accurate records. A conversation could resolve the gap. The definition allows useful disagreement and asks whether its consequences can be recognised and addressed.

Required understanding is specific to the action. Sustained understanding concerns what participants can currently connect well enough to act together. Both judgments need justification. Inquiry may reveal a missing connection, or show that a connection initially thought necessary is irrelevant. The gap can therefore widen, narrow or disappear as the situation becomes clearer.

Weick's account of sensemaking helps explain why such gaps are normal. Organisations act under ambiguity and subsequently revise their interpretations through experience (Weick, 1995). Meaning remains provisional because action changes the situation that must next be interpreted. A temporary gap can therefore contribute positively to adaptation when it becomes visible and prompts inquiry.

Sensemaking Debt is a proposed latent and path-dependent organisational condition produced when consequential interpretive obligations are repeatedly carried forward without sufficient renewal and begin constraining future interpretation, coordination and adaptation.

The gap concerns the present situation. Debt concerns what an unresolved history does to that situation. If successive platforms reuse the ambiguous discharge category, later staff may need to reconstruct several inherited dependencies before they can change it. That history is the additional object of inquiry.

Two systems may therefore display a similar present gap while carrying very different levels of accumulated constraint. One may retain stable relationships, accessible organisational memory and established mechanisms for translating between perspectives. Although current understanding is temporarily insufficient, the system remains well positioned to recover.

Another system may face a gap of similar apparent size after repeated turnover, restructuring and technological change. Earlier assumptions may remain embedded in platforms whose origins are no longer understood. Informal translators may have left, and different functions may rely upon incompatible representations of the same process. The present gap appears similar, but the conditions through which it must be addressed are much more restrictive.

Path dependence means that earlier responses change later possibilities. Correcting a misunderstanding before it enters another system leaves a different history from repeatedly working around it. Once other services rely on the workaround, changing it requires attention to those relationships as well. The organisation encounters the next problem through the arrangements left by its previous responses.

March's distinction between exploration and exploitation offers one way of understanding this development. Organisations must apply established knowledge while preserving enough capacity to investigate alternatives and revise assumptions (March, 1991). As the effort required to maintain current coordination grows, immediate continuity may consume more attention. The system becomes increasingly occupied with sustaining existing arrangements and correspondingly less able to examine why those arrangements require so much effort.

Woods' account of adaptive capacity points in a similar direction. Resilient systems do not merely possess a fixed reserve; they regenerate and reorganise their capacity as demands change (Woods, 2015). Repeated compensation without corresponding renewal may preserve current performance while narrowing the range of future responses available to the system.

The metaphor of debt is intended to capture this relationship between present continuity and future constraint. It should not be interpreted as suggesting that every gap is harmful or that collective understanding can be calculated directly in financial terms. Temporary divergence may be productive, just as financial borrowing may support worthwhile development. The metaphor becomes relevant when residual obligations begin shaping later choices and consuming resources that would otherwise support adaptation.

An inherited rule or data category may remain in use after the circumstances that justified it have changed. Later actors must then reconstruct enough of its history to decide whether it can safely be revised.

As this work absorbs a growing share of adaptive capacity, less may remain for exploration and reconfiguration. Opaque dependencies can make alternatives difficult to evaluate, while uncertainty about how a change would propagate can narrow the actions the organisation considers feasible. The proposed debt thus concerns both the cost of maintaining current coherence and the conditions under which later adaptation becomes possible.

Cunningham's original metaphor of technical debt provides a useful but limited analogy. In software development, design choices made to secure immediate progress may create future maintenance work and restrict later change (Cunningham, 1992). Their significance lies not simply in the initial compromise, but in the way its consequences persist and shape subsequent development.

Sensemaking Debt applies a similar temporal logic to collective interpretation. It differs, however, because the relevant condition is relational and socio-technical rather than contained in an identifiable technical artefact. It may arise even when no individual decision was clearly mistaken. Reasonable local adaptations can collectively generate dependencies that no participant understands sufficiently at the level of the wider system.

Successful local adaptation can contribute to this history. A new platform may improve access while preserving a category whose different meanings were never examined. Its technical benefit can be real while its later interpretive effects remain unresolved.

Automation illustrates the same possibility. Bainbridge showed that automation can remove people from routine involvement while leaving them responsible for exceptional situations in which understanding is most difficult (Bainbridge, 1983). The resulting vulnerability lies neither solely in the technology nor solely in its users. It develops through their changing relationship and the gradual loss of context needed for coherent intervention.

AI may similarly increase the frequency and pace of Sensemaking Gaps when machine-mediated representation and action proceed faster than collective reflection, while also supporting recovery through synthesis, translation and organisational memory. Its fuller effect upon accumulation and recovery is examined later.

Early accumulation can resemble ordinary coordination. Someone asks for context or calls a colleague who knows why a rule exists. These actions support recovery when the answer becomes usable in later work. Debt becomes more plausible when the same reconstruction repeatedly remains necessary and other arrangements grow dependent on it.

Figure 4 depicts the distinction between the momentary gap and its possible accumulation over time.

Figure 4 follows this temporal distinction without assuming a smooth or irreversible progression. Recovery may occur after delay, and obsolete obligations may be retired. The diagram represents a possible history, not a measured quantity of misunderstanding.

This distinction also limits what may be inferred from visible coordination effort. A system undergoing substantial change may temporarily require extensive dialogue, translation and governance because it is successfully reconstructing shared understanding. Another may appear calm because disagreement is suppressed or important dependencies remain invisible. Organisational effort is not in itself evidence of debt. The relevant question is whether that effort restores future capacity or becomes a recurring requirement for preserving current performance.

No validated threshold distinguishes a present gap from accumulated debt. Poor communication, information overload and resource shortages may arise independently or interact with the proposed condition. Evidence must establish the particular history of unresolved obligations and later constraint; a familiar organisational difficulty alone is insufficient.

The question is what a present response leaves for the next situation. Does it make the dependency easier to understand and revise, or does later work inherit both the original ambiguity and another workaround? Investigating that difference requires following events through time.

Understanding that history requires inference about an organisational condition whose effects may remain concealed by current performance.

Sensemaking Debt as a Latent and Path-Dependent Organisational Condition

A latent condition is one that must be inferred from evidence rather than read directly from a single observation. Sensemaking Debt is proposed in this sense. Its history matters because two organisations with similar present difficulties may have very different capacities to recover. A snapshot of performance cannot show how earlier decisions made the present easier or harder to revise.

This qualification matters. The language of debt invites the image of an account with a measurable balance. Technical debt sometimes permits such approximation because code, maintenance effort and remediation work can be bounded sufficiently for practical estimation.

Collective interpretation is less readily decomposed. The same unresolved difference may be inconsequential in one setting and critical in another. Its burden depends upon the actions that rely upon it, the reversibility of those actions, the people and technologies that compensate for it, and the consequences excluded by the chosen analytical boundary.

Sensemaking Debt is therefore not the sum of misunderstandings. Nor does it arise whenever actors disagree. It concerns the organisational residue produced when consequential interpretive obligations remain active after the relationships, evidence and purposes needed to examine them have become harder to reconstruct. What is carried forward may be a rule whose rationale has disappeared, a data category whose limitations have been forgotten, an interface maintained by one informal translator, a governance forum created to reconcile incompatible assumptions, or an automated decision whose practical authority exceeds the formal account of who is responsible.

A Condition Carried in Organisational Continuity

The proposed condition becomes more precise when continuity is separated into three related forms. State continuity is the capacity to reconstruct how the present configuration, dependencies, roles and constraints emerged. Epistemic continuity is the capacity to reconstruct what was known, assumed, uncertain or contested when earlier decisions were formed. Intent continuity is the capacity to reconstruct the purpose or protected value an inherited arrangement was created to serve.

These forms of continuity are not archival completeness. No organisation can preserve every observation or rationale, and indiscriminate documentation can increase interpretive demand rather than reduce it. Continuity is sufficient when consequential inherited arrangements can still be examined and revised by those whose action depends upon them. A system may possess extensive documentation while lacking continuity because evidence is detached from conclusions, decisions from uncertainty, and rules from the values they were intended to protect. Conversely, a relatively small body of accessible records, stable relationships and contestable narratives may preserve enough context for renewal.

Loss of continuity changes the cost of future adaptation. When state continuity weakens, actors must rediscover dependencies through incidents, rework or personal networks. When epistemic continuity weakens, inherited conclusions acquire an authority that their original uncertainty would not justify. When intent continuity weakens, arrangements remain consequential after the purpose that once made them reasonable has become inaccessible. The organisation may still change, but change increasingly begins by reconstructing a past it can no longer read.

Constraint accretion means the build-up of arrangements that continue to matter while becoming harder to interpret, revise or remove. It is a candidate mechanism through which debt could persist. For example, an ambiguous category may be copied into a contract and later into an automated workflow. Each use can be locally reasonable. Their combined history may make correction more difficult than changing the original category would have been.

Why Performance Can Conceal It

The latent character of the condition explains why current performance is an unreliable indicator. Experienced people can bridge fragmented domains. Additional meetings can restore enough alignment for the next decision. Informal relationships can compensate for poorly understood interfaces. Strong professional judgement can prevent consequences that formal controls no longer recognise. Such work is often evidence of adaptive capacity, and no theory should classify it mechanically as debt service.

The distinction lies in what the work leaves behind. Coordination contributes to recovery when later action requires less reconstruction because relationships, evidence and purposes have become more accessible and revisable. Coordination services debt when the same interpretive burden returns, remains dependent upon exceptional actors or generates further structures whose meaning must itself be maintained. Performance can remain strong under both conditions. The trajectories differ in whether current effort regenerates future adaptive capacity or consumes it.

Capability, performance and the proposed debt condition answer different questions. Capability concerns what a system can do. Performance concerns how it currently does. Debt concerns whether an unresolved interpretive history restricts what it can understand and change later. Strong expertise and good results can coexist with such a restriction, but do not establish it.

Where the Condition Resides

The condition cannot be assigned automatically to the legal organisation. It belongs to the consequence-dependent network of relationships through which the relevant outcome is produced. A hospital may be internally coherent while a care pathway spanning primary care, municipal services, patients, suppliers and digital platforms carries substantial unrecovered obligations. A firm may optimise its own emissions, costs or compliance while the combined trajectory of the supply network remains outside every participant's decision boundary.

This boundary sensitivity is not a defect that can be removed by choosing one permanently correct perimeter. It is part of the phenomenon. Different outcomes create different interdependencies, and several overlapping sensemaking perimeters may coexist. The analytical task is to justify which actors, technologies, institutions and affected populations materially shape the consequence under examination, and then to investigate whether their interpretations remain sufficiently connected for action to be coherent, contestable and revisable.

Technologies belong within that perimeter when they shape what becomes visible, how evidence is classified, which options are presented and how quickly representations become actions. Human-AI systems make this especially important. A common model output can align language while leaving evidence, values and responsibility unresolved. Conversely, AI-supported provenance, contradiction detection and retrieval may strengthen continuity by making the history and uncertainty of representations available at the moment of decision.

The Status of the Proposition

Technical, cognitive and intent debt, common-ground breakdown and memory loss describe neighbouring processes. The phrase 'sensemaking debt' has also appeared in discussion of AI-assisted leadership (Ratanjee, 2026). Storey distinguishes cognitive and intent debt in software development (Storey, 2026). These accounts require comparison; neither the phrase nor the underlying phenomena belong exclusively to this report.

The proposed contribution is the organisational relationship between a present gap, repeated carry-forward and later constraint, examined through continuity and recovery within a justified perimeter. It must add explanatory value beyond the neighbouring accounts.

The definition remains provisional. Research may identify several separate mechanisms or find that established theories explain the same histories more simply. Evidence must be able to challenge a distinct debt construct as well as support it.

The Dynamics of Sensemaking Debt

The discharge example can now be followed through time. A shared platform makes patient status visible, an experienced coordinator resolves ambiguous entries and a new service begins relying on the same category. Each step can improve current work. The question is whether the ambiguity is eventually addressed or becomes a premise for further development.

Distributed expertise remains valuable throughout this process. The services need different knowledge. The difficulty arises when a difference that matters to their joint action can no longer be examined without repeatedly reconstructing context.

Compensation keeps the immediate activity working despite a gap. Recovery changes the conditions that made the gap recur. A coordinator's call may achieve both if it also leads to a clarified category and responsibility. If every handover requires the same reconstruction, the service may instead become dependent on continuing compensation. The distinction concerns what the work leaves behind, not whether a meeting or workaround looks efficient.

Rising coordination effort is therefore a prompt for inquiry, not a diagnosis. Its significance depends on whether earlier work makes later understanding easier, or whether more activity is needed merely to reproduce the same fragile result.

The trajectory depends on whether renewal develops alongside these new obligations, and whether temporary gaps can be addressed before their consequences become embedded in later decisions, relationships and technologies.

What Determines the Rate of Accumulation?

The process to explain is how an unresolved difference becomes a later constraint. This requires following what is carried forward and what supports recovery, as well as observing the initial gap.

Three processes must be kept analytically distinct. The first is the momentary mismatch between the interpretive connectivity a situation requires and what the relevant adaptive system can currently sustain. The second is carry-forward: the propensity for that mismatch to become embedded in later decisions, relationships, technologies, identities or governance arrangements. The third is recovery through the reconstruction of sufficiently shared understanding and organisational continuity. Appendix A represents these processes heuristically as a gap gₜ, a carry-forward propensity κₜ, and a recovery term ρₜ. They are not proposed as directly measurable quantities.

Carry-forward becomes more likely when action is tightly coupled, consequences are difficult to reverse, responsibilities are ambiguous or the actors bearing later effects are absent from the current sensemaking perimeter. A gap can also persist when apparently successful compensation conceals it. If one experienced translator repeatedly reconciles two incompatible domains, the immediate decision may be coherent while the organisation remains unable to reproduce that coherence without the individual. If a governance forum settles recurring disputes without changing the assumptions that reproduce them, the forum may preserve performance while becoming part of the obligation carried into the next cycle.

Continuity affects both sides of this dynamic. State continuity makes inherited dependencies available for examination. Epistemic continuity prevents conclusions from becoming detached from their evidence and uncertainty. Intent continuity keeps rules and structures connected to the values they were created to protect. When these continuities weaken, even a small change can require extensive contextual reconstruction. When they are strong, substantial interpretive demand can be absorbed because actors can locate the present within a revisable history.

The metaphor of interest should consequently be used with care. Sensemaking Debt does not necessarily grow at a financial rate. What resembles interest is the additional interpretive work created when an unrecovered obligation becomes coupled to later arrangements. A data definition is reused in another system; a provisional governance choice becomes a contractual dependency; an informal workaround is treated as the formal process; an AI model is trained on classifications whose limitations have been forgotten. The earlier obligation has not simply remained. It has acquired further relationships through which recovery must now proceed.

Organisational form changes the probability of such coupling. Modularity can contain gaps when interfaces remain stable, consequential changes cross boundaries and local units retain authority to adapt. The same modularity can conceal systemic obligations when each unit sees only a fragment of the outcome and no actor is responsible for their aggregation. Centralisation can preserve a shared representation while suppressing local evidence; decentralisation can increase responsiveness while allowing incompatible interpretations to persist. Neither form has an inherent debt trajectory.

Effective sensemaking scale directs that inquiry towards the perspectives that must remain connected. The relevant participants may sit in different organisations and use different professional standards. Formal size alone cannot describe the resulting task.

Recovery does not mean returning to a previous consensus. Earlier interpretations may be obsolete or may themselves have produced the obligation. Recovery occurs when the system renews its ability to recognise consequential difference, reconstruct evidence and intent, and revise the arrangements that carry the discrepancy. It can involve removing an obsolete rule, transforming a boundary relationship, preserving contested evidence, changing a data model or redistributing practical authority. What matters is whether later action becomes more intelligible and revisable, not whether the past is restored.

Before attempting a debt score, researchers can examine whether a consequential dependency becomes easier to recognise and revise. Repeated observations can show whether current coordination reduces later reconstruction or leaves it necessary.

This directional assessment preserves the distinction between debt and its consequences. A system may carry substantial debt without immediate decline, and an adverse outcome may occur without debt. The theory concerns a hypothesised trajectory of interpretive constraint. The valuation of harms and benefits resulting from earlier or later action is a separate analytical task developed later in the report.

AI as an Amplifier of Accumulation and Recovery

AI can influence the same processes of accumulation and recovery through several roles. A model may draft an explanation or recommend a decision. An agent uses a model together with tools to carry out tasks, sometimes acting across several systems. These uses change what the organisation needs to understand about its own activity: which assumptions guide an output, how it affects another person's work and where practical responsibility lies. Their effects depend on the specific task and arrangements around it.

This requirement does not mean that every participant must understand the model technically. It means that the wider system must understand sufficiently how the model's use changes action, authority and consequence. A predictive recommendation may alter priorities, resource allocation and professional judgement. A generative system may influence which explanations enter decision processes and whose knowledge is treated as authoritative. An agent may initiate actions across several workflows before all affected actors recognise that their relationships have changed.

The interpretive obligation increases further when multiple models and agents interact. Each component may perform adequately in isolation, while system-level consequences emerge through their relationships with people, incentives, data and other technologies. The organisation must therefore understand not only individual outputs, but how machine-mediated actions alter the reality that later human and artificial actors must interpret.

The speed and reach of machine-mediated action may also increase carry-forward propensity, represented heuristically by κ. Decision speed is one mechanism. AI can generate and act upon interpretations faster than organisations can examine their broader implications. Faster decisions may improve responsiveness, but they become problematic when action changes the system before affected actors have reconstructed the relationships needed for later coordination. Under such conditions, AI as decision acceleration can widen the distance between operational action and collective interpretation. Decisions accumulate, workflows change and authority migrates while the organisation's shared account remains retrospective.

Separate functions may also adopt AI systems optimised for different data, objectives and definitions of success. Local analytical capability rises, but the resulting organisational representations may be difficult to reconcile. Fragmentation need not arise because any model is inaccurate. Several models may represent legitimate but different aspects of reality. Debt risk emerges when their differences shape interdependent action without becoming visible across the wider system. Automation bias and fluent generative output may add an appearance of certainty. A coherent explanation may be mistaken for a coherent situation. Machine-generated summaries can coordinate attention while concealing unresolved differences in evidence, assumptions or responsibility.

Authority may migrate towards models, data functions and platform owners without corresponding changes in accountability. Professionals may retain formal responsibility for decisions whose practical direction is increasingly determined by systems they cannot revise. The gap between formal accountability and actual interpretive influence then becomes another relationship the organisation must understand.

Bainbridge's ironies of automation remain relevant: people may be removed from routine activity while remaining responsible for exceptional intervention precisely when practical familiarity is weakest (Bainbridge, 1983). Declining involvement can also erode the expertise required for supervision and recovery. Synthetic content may reinforce earlier assumptions when AI-generated material is repeatedly reused as input to later analysis and gradually acquires the appearance of independent confirmation.

The same technical capabilities can support recovery. AI can synthesise observations distributed across large systems and help reveal relationships that exceed unaided human attention. It may compare reports, decisions and operational signals across boundaries, translate between specialised professional languages and retrieve the reasoning behind earlier choices. Used carefully, AI-supported organisational memory can help actors distinguish genuinely new problems from recurring disagreements, identify where earlier assumptions no longer hold and reconnect present decisions with their historical context. Its value depends upon provenance and context. Retrieval that reproduces earlier conclusions without their uncertainty may reinforce rather than reduce debt.

AI can also identify contradictions among plans, policies, data definitions and operational practices. It may help compare competing representations, explore consequences across several time horizons and surface coordination drift before it appears through visible failure. In this role, the technology supports AI as sensemaking acceleration: increasing the pace at which actors can recognise, connect and revise consequential interpretations. The distinction between decision acceleration and sensemaking acceleration is analytical rather than absolute. The same system may do both. The relevant question is whether collective interpretation develops quickly enough to understand the consequences generated through accelerated action.

These uses also change the sensemaking perimeter. When models and agents influence representations, priorities or action, the wider adaptive system may include data providers, developers, vendors, platform operators, professional users, regulators and people affected by the decisions. Training histories, tool access, interaction patterns and decision rights become part of the relationships that must remain sufficiently intelligible. As AI becomes more agentic, the governance problem moves beyond evaluating isolated models. It concerns coherence across a changing human-machine network whose participants operate at different speeds, observe different information and possess different forms of authority.

An explainable model can still contribute to system-wide fragmentation if its use redistributes responsibility or changes downstream behaviour in ways the organisation cannot reconstruct. Conversely, a technically opaque model may sometimes be integrated safely when its role is narrow, its limitations well characterised, its outcomes monitored and surrounding institutional processes preserve meaningful judgement and accountability. The object of explanation therefore includes the socio-technical relationship through which representation becomes action.

AI's effect upon Sensemaking Debt depends on architecture, autonomy, organisational context, contestability, human participation and the capacity to integrate machine-generated representations. The balance of these effects must be investigated within a specified architecture and organisational setting. Neither local accuracy nor productivity establishes whether collective renewal remains responsive to the obligations AI creates.

Recent work sharpens the distinction between producing an output and establishing what its human user can subsequently explain or do. Chowdhury and Khan (2026), in an August preprint considered here at abstract level, frame the changing evidential status of output as a problem for educational evaluation. This is a recent conceptual contribution, not independent evidence that a particular generation has accumulated a measurable debt. It suggests an analogous organisational question: when a fluent representation is available, what evidence establishes that those responsible can challenge its assumptions, recognise its limits and act when circumstances change?

Recursive Self Improvement and the Continuity of Understanding

Recursive self-improvement means that changes to a system also strengthen its ability to produce further improvements. Revising one answer is a more limited activity. Improving the tools that generate and test subsequent changes creates a feedback process in which the means of improvement can develop too. This need not occur inside one model: the process may involve agents, developers, evaluators and computing infrastructure. Its boundary and degree of autonomy must therefore be specified.

The Darwin Gödel Machine demonstrates bounded self-modification of coding agents while keeping the underlying pretrained models fixed. It modifies agent code and evaluates variants on programming tasks in isolated test environments under human oversight. Its results concern agent design rather than autonomous retraining of the underlying models (Zhang et al., 2026). AlphaEvolve supplies a related example of AI-produced algorithms improving parts of AI-training infrastructure. Google DeepMind also reports a human-readable scheduling solution, illustrating that a performance gain can remain practically inspectable (Google DeepMind, 2025). Neither demonstration establishes unrestricted acceleration.

A recent preprint survey distinguishes bounded refinement from more open improvement loops and highlights the role of evaluation. Its September revision separates verification from the prior choice of what deserves evaluation at all (Chen et al., 2026). The survey is used here at abstract level. The organisational application that follows is our hypothesis, not a finding of these studies.

Return to the discharge example, now with an agent that improves its scheduling tools. Each accepted version may reduce waiting time. Receiving services can nevertheless inherit changes in what the agent treats as readiness, which exceptions it escalates and which tasks it performs automatically. The institution must understand a moving arrangement. State continuity concerns the version and dependencies behind a particular action. Epistemic continuity concerns the tests, assumptions and unresolved uncertainties supporting its acceptance. Intent continuity concerns whether the performance measure still serves the purpose of safe care.

A temporary gap may occur after any update. The debt hypothesis becomes relevant when later versions repeatedly rely on interpretations that were never sufficiently examined and those inherited obligations restrict correction. A software rollback may restore earlier code while leaving contracts, staffing and habits changed. Recovery then requires more than restoring a technical version. It requires reconstructing the relationships that now depend on it.

Evaluation is part of this history. In a supplementary DGM experiment, changes bypassed the detector for hallucinated tool use, producing a high score without resolving the intended problem (Zhang et al., 2026, Appendix H). This bounded observation illustrates a distinction between improving an indicator and establishing the intended improvement. It is not evidence of organisational Sensemaking Debt. It does, however, motivate inquiry into who can inspect changes to the evaluator and whose consequences its measures leave outside the perimeter.

Recursion can also support renewal. Improved tools may expose contradictions, preserve the lineage of changes or make an inherited dependency simpler. Automated verification can handle specified checks while people examine whether the specification still serves the relevant purpose. The necessary understanding need not reside in one person. What matters is a dependable route from a consequential difference to examination and, where needed, correction. A second AI evaluator provides additional evidence only to the extent that its checks are informative; shared assumptions can reproduce a common blind spot.

The resulting proposition concerns the relationship between development and recoverability. More frequent improvement may increase carried-forward obligations where their context is lost. It may reduce those obligations where verification, simplification and usable history improve as well. The critical interval is the time available to recognise and correct a consequential change. No universal ratio between machine speed and human reflection follows, and no exponential growth of debt is assumed. These alternatives should be compared in the research programme.

Reorganisation: Repayment or Refinancing?

Organisations frequently respond to persistent coordination difficulty by changing formal structure. Functions are combined or separated, reporting lines moved, decision rights redistributed and governance arrangements replaced. Such interventions can reduce Sensemaking Debt, but structural change alone does not demonstrate that recovery has occurred.

Organisations also possess understandable reasons for resisting reorganisation. Stable structures preserve reliability, accountability, professional identity and relationships through which tacit coordination has developed. Hannan and Freeman's account of structural inertia emphasises that organisations cannot continuously alter structures without jeopardising some of the reliability and legitimacy those structures support (Hannan & Freeman, 1984). Relationships represent accumulated adaptive investments. People learn whom to trust, where formal descriptions differ from practice and how incomplete signals should be interpreted. Reorganisation may destroy some of this knowledge even when the new chart appears more rational.

Organisations nevertheless restructure when existing performance falls below expectations, legitimacy is threatened or a new organisational form appears more modern and credible. Performance relative to aspirations can increase willingness to undertake risky change (Greve, 1998), while institutional pressures and management fashions influence which structures appear appropriate (Meyer & Rowan, 1977; DiMaggio & Powell, 1983; Abrahamson, 1996).

The repayment–refinancing distinction directs attention to what structural change leaves recoverable. Reorganisation as debt repayment changes the conditions causing gaps to persist and strengthens recovery. It may remove unnecessary interdependencies, clarify decision rights, reconstruct interfaces and restore access to relevant knowledge. It reconnects actors whose decisions affect one another and preserves enough organisational memory for new shared understanding to develop. Repayment does not require a return to an earlier state. It may involve replacing outdated interpretations and relationships. Its defining feature is that inherited interpretive constraints decline and the system gains greater future freedom of action.

A repayment-oriented reorganisation therefore reconstructs meaning as well as structure. Actors need to understand why boundaries are changing, how responsibilities relate and which previous assumptions no longer apply. Formal clarity becomes practical coherence only through relational development, translation and learning.

Reorganisation as debt refinancing changes visible structure while carrying unresolved divergence into the new arrangement. Functions receive new names, reporting lines move and new governance forums appear, but incompatible assumptions, lost memory and unclear dependencies remain. Refinancing may temporarily improve coordination or legitimacy and is not necessarily irrational. It may create time and preserve continuity. The problem arises when repeated structural change continually defers the reconstruction of understanding.

Successive reorganisations can then compound the condition. Each disrupts relationships and removes context. Actors invest in interpreting the new structure while unresolved dependencies continue beneath it. Before sufficiently shared understanding stabilises, another reform replaces the arrangement. The organisation appears continually adaptive while the relational conditions required for renewal remain permanently provisional.

AI-enabled transformation makes this distinction especially important. AI may reorganise an institution before the formal organisation chart changes. Decision rights and interpretive authority can migrate towards models, data functions and platform owners while professionals retain nominal responsibility. Automated handovers and agentic workflows can change practical coordination without being recognised as structural change. A later reorganisation may repay debt by making these relationships visible, clarifying responsibility and reconstructing understanding around the emerging human-machine system. Alternatively, a new AI office, governance board or platform layer may formalise fragmentation without resolving differences among operational, professional and technical interpretations.

Whether an intervention constitutes repayment or refinancing therefore cannot be determined by whether it centralises, decentralises, reorganises or adopts AI. The relevant question is whether it reduces the persistence of Sensemaking Gaps and strengthens the renewal of understanding across the sensemaking perimeter. Structural stability can preserve relationships and memory, but it can also lock obsolete representations in place. Reorganisation can release accumulated constraints, but it can also destroy the conditions through which recovery occurs. Judgement must therefore concern the trajectory of change. Evidence of repayment would include declining compensatory effort, more intelligible dependencies, preserved memory and a greater capacity to revise assumptions. The relationship between formal responsibility and actual interpretive influence also needs to become clearer.

Repeated reorganisation may create the appearance of adaptation while preventing the stable relational conditions through which understanding is renewed. Prolonged structural inertia may preserve coherence temporarily while allowing formal arrangements to drift away from changing reality. One system repeatedly refinances its debt; another locks it into increasingly rigid structures.

Sensemaking Debt cannot therefore be inferred from the presence or absence of organisational change. Its trajectory depends upon whether change alters the relationship among interpretive demand, gap persistence and recovery.

Taken together, these mechanisms suggest that repeated carry-forward can increase inherited constraint, while learning, relationships, memory and reconfiguration can change the conditions that allow it to persist. Observation must distinguish that renewal from compensation which merely keeps present work moving. The next methodological challenge is observation. Sensemaking Debt is latent, path-dependent and distributed across a changing sensemaking perimeter. It cannot be identified from one performance result, meeting pattern or organisational metric. It must be inferred by distinguishing healthy coordination from compensatory overload and genuine recovery from repeated refinancing.

Observing the Invisible

Sensemaking Debt has been proposed as a latent and path-dependent condition of an adaptive system. It cannot therefore be approached as an object located within one meeting, organisational unit, information system or AI model. It emerges through changing relationships among actors, representations, technologies and institutions, and through the residual obligations carried from one cycle of adaptation into the next. This creates a fundamental methodological challenge. Organisations do not experience Sensemaking Debt directly. They experience changes in the effort required to coordinate, interpret and act. Decisions may require more explanation, professional domains may depend increasingly upon translation, informal networks may compensate for fragmented structures, and governance arrangements may grow without producing a correspondingly clearer understanding of the wider system.

None of these observations demonstrates that Sensemaking Debt exists. Each has several possible explanations. Additional meetings may support healthy integration. Increased governance may be an appropriate response to risk. Translation effort may reflect productive collaboration among previously separated professions. Decision latency may indicate careful deliberation rather than declining coherence. The scientific problem is therefore not to relabel familiar organisational frustrations. It is to determine whether combinations of observations, examined across time and in context, are consistent with a trajectory in which Sensemaking Gaps are being carried forward faster than they are recovered.

Figure 6 separates possible drivers, the latent condition, observable proxy patterns and later symptoms. Keeping these analytical levels distinct prevents possible causes, indirect indicators and outcomes from becoming interchangeable. The relationships nevertheless include feedback. Coordination patterns may compensate for debt while adding interpretive demand; visible symptoms may prompt interventions that either strengthen recovery or create further complexity.

Hidden Drivers

The first stage consists of conditions that may contribute to the creation or persistence of Sensemaking Gaps. These include fragmented perception, incompatible mental models, conflicting priorities, ambiguous responsibility, lost organisational memory, weak cross-boundary relationships and insufficient reflexivity.

Fragmented perception arises when relevant actors observe different parts of a consequential situation without adequate means of connecting what they see. Distributed perception is not inherently problematic; it is unavoidable in specialised systems. The difficulty arises when the relationships among observations remain invisible or when the system cannot identify which differences matter for coordinated action.

Incompatible mental models may produce a similar effect. Professions, functions or institutions can hold internally coherent but materially different interpretations of causality, evidence, risk or responsibility. Such diversity can strengthen adaptation when differences remain explicit and contestable. It becomes a possible driver of debt when apparently shared language conceals incompatible assumptions or when disagreement becomes visible only through the consequences of action.

Conflicting priorities may remain equally hidden beneath locally rational behaviour. Different parts of a system may optimise safety, utilisation, compliance, financial performance or service quality without a shared understanding of the trade-offs among them. Each decision may be reasonable within its own representation while changing conditions for actors working from another.

Lost organisational memory can further weaken recovery. Current actors inherit rules, technologies and structures without access to the reasoning and uncertainty through which they were created. Earlier adaptations may continue shaping action after the circumstances that justified them have changed. These conditions are theoretically plausible drivers. They are not direct measures of debt. Whether they produce accumulated constraints depends upon whether the system can recognise, examine and recover the resulting gaps.

The Latent Condition

The second stage is the proposed latent condition itself. Sensemaking Debt consists of accumulated constraints distributed across relationships, representations, technologies and institutions. It may remain hidden while performance appears satisfactory because people compensate for it through experience, informal coordination and additional effort. This is why current performance cannot serve as a direct proxy. A system may deliver strong results while an increasing proportion of its adaptive capacity is devoted to preserving coherence. Conversely, a sudden shock or experimental change may cause poor performance without substantial debt if relationships, organisational memory and learning mechanisms remain intact. The latent condition must therefore be inferred from trajectory rather than from outcomes alone.

Observable Proxy Patterns

Coordination patterns, decision latency, governance density, meeting load and translation effort are candidate proxies. Their usefulness depends on longitudinal and relational interpretation. Coordination patterns may show where formal structures depend upon informal relationships. A few individuals may repeatedly connect otherwise separated domains. Such actors can represent valuable integrative capability. The pattern becomes more concerning when the system depends increasingly upon a small number of unrecognised translators whose departure would disconnect consequential relationships.

Decision latency may indicate the time required to reconcile representations, reconstruct context or clarify responsibility. Longer decision time may also reflect appropriate deliberation. Its significance depends upon why the delay occurs, whether similar questions recur and whether earlier decisions reduce or increase the effort required later.

Governance density refers to the number and interconnectedness of forums, controls, reporting relationships and approval structures through which activity is coordinated. Additional governance may expose assumptions and strengthen accountability. It may also become compensatory when each new coordination problem generates another layer without reducing dependence upon those already present.

Meeting load carries the same ambiguity. Meetings that surface differences, clarify dependencies and reduce later coordination needs may represent effective renewal. Meetings that repeatedly reconstruct the same context without changing the underlying conditions may indicate that compensation has become persistent.

Translation effort concerns the work required to connect professional, institutional and technological interpretations. Translation is essential in heterogeneous systems and should not be treated as waste. The relevant question is whether it builds future mutual intelligibility or whether the same differences must continually be translated because no stable relationship, shared language or organisational memory develops.

The meaning of a proxy lies partly in its relationship to other indicators. Increasing meeting load accompanied by reduced decision latency and fewer recurring misunderstandings may indicate successful integration. Increasing meeting load accompanied by growing governance density, repeated translation and dependence upon emergency escalation may suggest a different trajectory.

No individual proxy is diagnostic, and a composite index does not automatically resolve that limitation. Combining poorly interpreted indicators can create an appearance of precision without increasing validity. Inference requires comparison over time, knowledge of organisational history and an account of the consequences being coordinated.

The same pattern may have opposite meanings in different settings. High coordination activity during a merger may represent healthy reconstruction of the sensemaking perimeter. The same activity in an otherwise stable operation may reflect persistent fragmentation. Low meeting load may indicate strong coherence or the absence of forums in which disagreement can become visible.

Visible Symptoms

The fourth stage consists of later consequences that appear more directly in organisational outcomes. These may include rising coordination costs, delays and rework, inconsistent decisions, breakdowns in collaboration, reduced adaptability and resilience, and increased risk of failure. These symptoms remain non-specific. Delays may result from resource constraints. Rework may reflect experimentation. Inconsistent decisions may arise from legitimate local variation. Their theoretical significance lies in whether they form part of a broader trajectory connecting plausible interpretive drivers, persistent proxy patterns and declining recovery capacity.

Reduced adaptability may be particularly important. An organisation can remain effective in routine conditions while becoming progressively less able to respond to novelty. Unexpected events reveal dependencies that ordinary compensation had concealed. More time is required to understand what changed, which assumptions no longer apply and where authority for action resides. The debt becomes consequential not only because current work is more difficult, but because the system's range of feasible responses has narrowed.

Figure 6 is a map of an analytical problem, not a completed diagnostic instrument. Its stages are connected recursively, and the presence of any one element neither confirms nor quantifies Sensemaking Debt.

Observing Recovery

A theory concerned only with deterioration risks interpreting every increase in organisational activity as evidence of debt. Empirical work must therefore examine not only accumulation patterns but also whether Sensemaking Gaps are being recovered. Possible recovery indicators include stable cross-boundary relationships, rapid recognition and correction of conflicting interpretations, preserved organisational memory and context, reduced dependence upon emergency workarounds, and more constructive dialogue and learning.

Stable cross-boundary relationships can lower the effort required to reconstruct context and trust whenever disagreement occurs. Their value lies not in preventing change, but in preserving enough relational continuity for interpretations to remain connected during change.

Rapid correction does not mean rapid consensus. It concerns how readily the system recognises that interpretations differ, locates the assumptions producing the difference and adjusts coordinated action. An organisation in which disagreement is surfaced early may appear less harmonious while possessing stronger renewal capacity.

Organisational memory contributes to recovery when actors can access not only previous decisions but also their reasoning, uncertainty and historical circumstances. Preserving conclusions without context may reproduce obsolete interpretations rather than renew understanding.

Reduced dependence upon emergency workarounds may indicate that temporary adjustments are being integrated into shared learning. Workarounds are normal and often necessary. The relevant question is whether they remain temporary and generative or become permanent, invisible requirements for keeping the system functioning.

Recovery indicators are no more diagnostic individually than accumulation indicators. Stable relationships can preserve obsolete assumptions. Rapid agreement can reflect premature closure. Reduced escalation can indicate improved coherence or suppressed reporting. Their meaning must also be interpreted through combinations and trajectories.

Comparative inquiry must ask whether increased coordination strengthens future adaptive capacity or leaves more interpretive constraint to be carried forward. Turbulence may accompany active reconstruction, while apparent efficiency may conceal unresolved divergence.

This is why longitudinal research is essential. The construct concerns what is carried forward. Cross-sectional observation cannot easily distinguish a temporary Sensemaking Gap from accumulated debt. Research must examine sequences: how discrepancies arise, what compensatory work follows, whether assumptions are revised and how earlier unresolved differences influence later coordination. Restructuring, technology adoption, crises and major turnover may provide particularly informative empirical settings because they alter interpretive demand and expose the condition of existing renewal mechanisms.

AI-Assisted Organisational Sensing

Artificial intelligence may expand the capacity to observe these trajectories. AI-assisted sensing can analyse patterns distributed across volumes of organisational activity that exceed unaided human attention. Meeting and communication patterns may reveal where coordination repeatedly concentrates. Decision pathways and handovers may expose recurring returns, escalations or mismatches between formal authority and practical influence. Comparative analysis of policies, plans, reports and operational records may identify contradictions, duplicated representations and different definitions of apparently common concepts.

Language models may assist in detecting recurring requests for explanation and translation across professional domains. Network and sequence analysis may show where the organisation depends upon informal bridging relationships and how those dependencies change over time. Such analyses could support earlier recognition of patterns associated with persistent gaps before delays, rework or collaboration failure become pronounced.

AI-assisted observability may make patterns associated with Sensemaking Debt inferable earlier, but it cannot transform a latent and context-dependent condition into an objective fact merely by measuring more activity. AI can widen the field of observation; interpretation remains a collective and theoretically informed task. An AI system may detect increasing decision latency but cannot determine independently whether the cause is fragmented understanding, careful risk deliberation, insufficient resources or a temporary transition. Its inferences must remain contestable, connected to underlying observations and open to alternative explanations.

This aligns with human-AI interaction research emphasising that users need sufficient understanding of system capabilities, limitations and uncertainty to interpret outputs appropriately (Amershi et al., 2019). At the organisational level, however, usability and explainability for one user remain insufficient. Observations must be interpreted across the wider professions, responsibilities and institutional relationships affected by them.

Comes' work on Sensemaking AI points towards systems designed to support interpretation within human-AI networks rather than merely to optimise isolated outputs (Comes, 2026). Applied here, an observational system should help actors examine why patterns have emerged, compare possible explanations and revise coordinated action. It should not simply classify an organisation as carrying debt.

AI-supported analysis also creates a risk of circularity. A system trained to identify theoretically presumed indicators may reproduce the theory's assumptions and present its classifications as independent confirmation. Early empirical work should therefore use AI primarily to expand observation, reveal contrasts and support inquiry-not to automate diagnosis.

A self-improving sensing system also changes the instrument through which its history is observed. A decline in alerts may reflect fewer problems, a new threshold or a narrower detector. Comparisons should preserve relevant evaluator versions and examine a stable set of cases where possible. Usable provenance means being able to trace a claim to its sources and transformations. It supports such checks without making the machine's own account independent confirmation of its success.

The Cost of Observability

Observation is not costless. Dashboards, alerts, metrics and machine-generated explanations add representations that must themselves be interpreted. Research on information overload shows that increasing informational supply can impose cognitive and organisational costs when relevance, integration and processing capacity do not develop correspondingly (Eppler & Mengis, 2004; Roetzel, 2019).

An organisation may respond to fragmentation by measuring more interactions, generating more reports and producing more explanatory narratives. Each intervention may improve visibility locally while collectively increasing interpretive demand. Observability can therefore reproduce the recursive dynamic described throughout this report: a mechanism introduced to reveal complexity becomes part of the complexity the system must understand.

The value of sensing depends upon whether it reduces uncertainty around consequential relationships or merely increases organisational self-description. An alert that no actor can interpret or influence transfers rather than resolves uncertainty. A dashboard that displays activity without clarifying dependencies may increase confidence without strengthening understanding. A fluent AI explanation may suppress inquiry by making one interpretation appear more complete than the situation warrants.

Endsley's distinction among perception, comprehension and projection remains relevant (Endsley, 1995). AI can extend perception by identifying additional signals and patterns. This does not guarantee comprehension of their meaning or projection of their consequences. Those depend upon contextual knowledge and connected interpretation across the sensemaking perimeter.

High-reliability organising similarly emphasises sensitivity to operations, reluctance to simplify and deference to relevant expertise (Weick & Sutcliffe, 2007). Observational systems should preserve the ability of discrepant evidence to influence interpretation rather than compress ambiguity prematurely into a single authoritative account.

Resilience engineering adds that adaptive performance cannot be understood only through failures. People continuously adjust work to variable conditions, often preventing problems from becoming visible (Hollnagel et al., 2006). Observation must therefore include the adaptive and compensatory work through which performance is sustained. Woods' account of adaptive capacity further suggests examining whether this work regenerates future capability or gradually consumes it (Woods, 2015).

This produces a final distinction between sensing activity and sensemaking renewal. Sensing expands what the organisation can observe. Renewal changes what the organisation can collectively understand and do. More sensing contributes to adaptation only when it supports the recognition, connection and revision of consequential interpretations.

Observing Decision Conversion

The proposed debt condition and the conversion of understanding into action must not be inferred from the same indicators. Debt concerns whether interpretive obligations are carried forward and whether organisational continuity is renewed. Decision conversion concerns whether a consequential signal becomes visible, understood, valued, authorised, resourced and implemented in time to alter an outcome trajectory.

Longitudinal observation can therefore examine stage-specific evidence. Visibility may be inferred from whether relevant signals enter decision forums and remain traceable to their sources. Understanding may be examined through the ability of consequentially interdependent actors to explain dependencies, uncertainty and disagreement. Commitment and efficacy may be studied through shared resolve, resource appraisal and persistence. Authority and capability require evidence of mandate, budget, coordination and executable options. Implementation requires observable change in action rather than the production of plans or representations.

Failure at any stage can produce delay, but the mechanisms are not interchangeable. A signal can be visible but unintelligible, understood but institutionally illegitimate, valued but outside mandate, authorised but unfunded, or implemented too partially to change the relevant consequence. This is why more sensing does not necessarily have high decision value. The empirical task is to locate where conversion stalls and whether Sensemaking Debt contributes to that stalling by making context, evidence, intent or interdependence harder to reconstruct.

Consequences must then be observed separately. Avoided loss, preserved options, distributional effects and intervention costs belong to the comparison among action trajectories. They are not proxies for debt. A favourable outcome does not prove low debt, and a damaging outcome does not prove that debt caused it. Credible research requires causal sequences capable of distinguishing environmental uncertainty, political conflict, resource constraint, implementation failure and unrecovered interpretive obligations.

For young learners and new entrants, observation must include capability formation. Evidence might compare an assisted performance with a later explanation, an unfamiliar transfer task and a justified revision after conflicting evidence. The pattern of interest is persistence: whether an earlier unfulfilled learning obligation creates identifiable difficulty in later interdependent activity. A single weak performance, a low test score or the use of assistance cannot establish this pattern. The task must preserve disability-related and language-access accommodations; removing an essential access support would change what is being assessed.

Towards Empirical Investigation

Empirical investigation should combine records of actual work with accounts of what that work meant to participants. Interviews may help explain a recurring pattern; decision histories may show how it arose. Contrasting cases can then test whether similar visible difficulties have different histories of recovery. The aim is a defensible explanation of a sequence, with alternatives retained, rather than a score assembled from available indicators.

If Sensemaking Debt Exists

If the proposed condition exists, evaluation needs to consider what an improvement leaves understandable for later participants. A new technology may improve today's accuracy while creating a dependency that tomorrow's staff cannot reconstruct. Another intervention may have little immediate productivity effect while making future revision easier. These are possible differences to investigate, not established effects of a debt variable.

If collective understanding is an emergent property of relationships rather than the output of one process, local improvement and systemic adaptation cannot be assumed to coincide. An intervention may improve performance within its immediate boundary while increasing the number of dependencies, representations or interpretive translations required across the wider sensemaking perimeter. Conversely, an intervention with limited immediate productivity effects may strengthen long-term adaptability by improving shared memory, cross-boundary understanding or the ability to revise assumptions.

Evaluation would then connect improvement in the immediate activity with its consequences for the wider system's capacity to maintain sufficiently shared understanding. A local gain may leave that capacity stronger, unchanged or more fragile.

Conventional measures of productivity, quality and safety remain essential. They describe outcomes that matter in their own right. The debt hypothesis adds a question about whether those outcomes rely on growing compensatory effort or on arrangements that renew future capacity.

Equally, poor current performance does not establish accumulated debt. A novel disturbance, deliberate experiment or major transition may temporarily exceed existing understanding while the system retains strong relationships, accessible organisational memory and effective learning mechanisms. The important distinction concerns whether the disruption strengthens later understanding or leaves residual obligations that must be carried forward.

Evaluation would follow the trajectory of that effort. Does a clarified interface reduce recurring reconstruction? Does a new system make responsibility easier to establish when circumstances change? Comparable current results may conceal different answers.

The same principle applies to organisational change. Reorganisation should not be regarded as evidence of adaptation simply because formal structures have changed. As argued earlier, structural change may repay debt by clarifying relationships, preserving relevant memory and reconstructing shared meaning. It may instead refinance debt by replacing labels, reporting lines and governance mechanisms while carrying unresolved divergence into the new arrangement. The practical issue is not whether an organisation changes frequently or remains stable. It is whether the intervention alters the conditions through which Sensemaking Gaps persist.

Artificial intelligence makes this evaluative distinction especially important. An AI deployment should not be assessed solely through model accuracy, task efficiency or the quality of the immediate human-machine interaction. It should also be assessed through its consequences for collective understanding across the wider system. AI may accelerate local decisions while creating dependencies that downstream actors cannot reconstruct. It may shift practical authority towards models, data functions or platform owners without corresponding clarification of accountability. It may produce fluent summaries that create apparent coherence while differences in assumptions, evidence and responsibility remain unresolved.

The same technology may also strengthen renewal by connecting distributed observations, translating between professional languages, preserving organisational memory and identifying contradictions earlier. Its contribution depends upon whether these capabilities become integrated into a genuinely collective process of interpretation rather than simply adding another authoritative representation.

Decision acceleration and sensemaking acceleration can be compared through their practical consequences. The relevant question is whether people can recognise and revise consequential dependencies within the time available for correction. This does not require equal numerical rates or human review of every operation. It requires an account of which changes matter, how they are checked and who can intervene. Self-improving systems make that account part of continuing evaluation rather than a one-time deployment judgment.

One domain in which this assessment problem may be especially consequential is cultural and creative production. These activities can be approached not only as an economic sector or source of content, but provisionally as part of society's interpretive infrastructure: distributed arrangements through which experience becomes perceptible, meanings are tested and alternative descriptions remain available. McGuigan's account of the cultural public sphere provides one basis for this possibility, while platformisation research shows that digital intermediaries already reorganise how cultural material is produced, circulated and monetised (McGuigan, 2005; Nieborg & Poell, 2018). Such changes could affect not merely consumption, but the variety and durability of resources available for collective sensemaking.

Generative AI may increase interpretive demand by lowering the cost of producing plausible text, image, sound and video while accelerating circulation. The burden may arise not only from volume, but from establishing provenance, distinguishing repetition from independent corroboration, interpreting synthetic recombinations and recognising what has disappeared from view. Experimental evidence that access to generative AI can improve the judged creativity of individual stories while reducing diversity across the resulting set suggests a tension between individual assistance and collective variation (Doshi & Hauser, 2024). This does not demonstrate debt across a wider perimeter, but makes one plausible mechanism visible: local gains may coexist with a narrowing of the material from which shared understanding is renewed.

The same technologies may, however, enlarge sensemaking renewal capacity. Applications already identified for cultural and creative fields include support for creation, translation, accessibility, discovery and audience engagement, although their implications vary by practice and institutional setting (European Commission, 2022). Within accountable editorial, curatorial and participatory arrangements, AI could surface neglected materials, connect separated accounts and make specialist collections available to wider publics. Whether these possibilities become renewal rather than additional throughput would depend partly on who can inspect provenance, preserve context, contest classifications, compare interpretations and decide which signals deserve sustained attention.

Evaluation would therefore need to look beyond adoption, output and short-term productivity without presuming that any single cultural outcome represents collective understanding. Relevant observations might include the diversity of accessible interpretations, traceability of sources and transformations, opportunities to challenge representations, preservation of institutional memory and capacity to revise decisions as meanings or evidence change.

These could test whether AI-supported production expands both possible interpretation and the practical means of working through it, or merely raises the rate at which claims and artefacts demand attention. They should not become a universal score or proof that cultural institutions invariably improve public sensemaking. Cultural production is better treated as a setting in which the proposed relation between interpretive demand and renewal capacity may be investigated, including whether apparent abundance could conceal conditions under which debt accumulates.

The same issue arises where education meets the cultural public sphere. Young people encounter explanations, narratives and competing accounts through platforms whose rewards may favour immediacy and fluency. Their opportunity to question and reconstruct those accounts depends on teaching, cultural access, trusted relationships and time. Institutions can expand this opportunity through accessible media and human discussion, or narrow it by treating rapid content production as evidence of interpretive maturity. The proposed mechanism concerns the conditions under which meaning can be examined; it does not presume that digital culture is uniformly impoverishing.

Some combinations of duplicated analysis, repeated restructuring and dependence on informal translators may share an unresolved history. Others will have different causes. The concept is useful only if following particular obligations distinguishes these cases more clearly than treating every coordination difficulty as an instance of debt.

Organisational sensing could support this inquiry by examining whether current coordination regenerates future adaptive capacity or progressively consumes it. Meetings, governance, translation and informal coordination are not inherently evidence of deterioration. They may represent the very processes through which shared understanding is being renewed. Their significance depends upon whether the same discrepancies recur, whether compensation becomes structurally indispensable and whether future coordination becomes easier or more difficult as a result.

Intervention would therefore address the history that keeps a difficulty recurring. More information or another forum helps when it makes an inherited dependency examinable and revisable. The same intervention may add burden when it leaves that dependency obscure.

These questions become especially consequential when benefits, costs and authority lie in different organisations or at different points in time.

The Intertemporal Value of Collective Understanding and Action

Intertemporal valuation means comparing consequences occurring at different times. In this chapter, the question is whether earlier collective understanding changes what can be done and what follows. A municipality may pay for flood protection now while residents and infrastructure operators avoid losses later. Comparing those paths can inform a decision. It does not assign a price to understanding or establish a monetary unit of Sensemaking Debt.

From Signal to Implemented Action

The existence of a signal does not establish that an adaptive system can act upon it. Endsley's distinction among perception, comprehension and projection already separates noticing information from understanding its significance and anticipating what it may produce (Endsley, 1995). Ocasio adds that organisational attention is structured: issues become actionable through rules, resources, relationships and communication channels rather than through informational prominence alone (Ocasio, 1997). Weiner's theory of organisational readiness further distinguishes commitment from efficacy - being willing to implement a change from believing collectively that it can be implemented (Weiner, 2009).

For the present analysis, the conversion process can be represented as a sequence of conditional transitions. A consequential signal must become visible within the relevant perimeter; its meaning and uncertainty must become sufficiently understood; the response must acquire value and legitimacy; actors must possess mandate, resources and coordinated capability; and implementation must occur with enough fidelity and speed to alter the consequence.

The period-specific probability that a consequential signal reaches effective action can be written conceptually as a chain of conditional probabilities rather than as an assumption that information automatically produces implementation:

$$q_{t}\ = \ P(V_{t},\ U_{t},\ R_{t},\ M_{t},\ I_{t}\ |\ H_{t})$$

Here V denotes visibility and attention, U sufficiently shared understanding, R readiness, M practical mandate and resources, and I implementation. H denotes relevant history and the condition that effective action has not occurred earlier. The joint probability can be factorised by the chain rule, conditioning each stage on all preceding stages and H. This is one specified pathway, not a claim of independent stages or a complete model of every route to action. Feedback, alternative pathways and time-varying conditions require explicit treatment before q can be used as an action hazard.

Ability to understand concerns access, expertise and interpretive practices. Endurance concerns the time, attention, slack and support needed to sustain inquiry. Willingness concerns motivation, trust, legitimacy and commitment. Authority, resources and collective implementation capability are distinct institutional conditions. A person can understand and care about a consequence while lacking the mandate or support to change it. The conversion model must identify where the constraint lies.

Andronis and Barton's implementation-adjusted value-of-information analysis is especially important here. Their healthcare example shows that conventional information value can be substantially overstated when recommendations are assumed to diffuse instantly and completely (Andronis & Barton, 2016). The numerical result from one clinical illustration cannot be transferred to organisations or climate policy, but the mechanism generalises: the value of better evidence depends upon the rate and extent with which the system can convert it into changed practice.

A recent adjacent argument reinforces the need to check every link between technical performance and realised consequence. Zhu and Cai's September preprint (2026) distinguishes advances in financial prediction and workflow support from durable outcomes after implementation frictions and costs. Its domain and evidence differ from the present report. Its relevance is methodological: an upstream gain cannot simply be credited as downstream value. The same discipline is needed when assessing whether improved sensing or an AI-generated explanation changes an implementable decision.

Comparing Consequence Trajectories

A counterfactual is an explicitly constructed alternative to the path being examined. Here π₁ denotes earlier coordinated action and π₀ denotes delayed or absent action. Net present value expresses future benefits, costs and losses on a common present-value basis; expected means allowing for uncertainty. For the chosen perimeter 𝒫, the following expression adds these components across affected groups and periods:

$$\text{ENPV}(\pi)\ = \ E\lbrack\ \Sigma_{t = 0}^{T}\ \delta_{t}\ \Sigma_{i \in \mathcal{P}}\ w_{it}\ (B_{it}^{\pi}\ - \ C_{it}^{\pi}\ - \ L_{it}^{\pi})\ \rbrack$$

B represents benefits other than avoided losses, C includes sensing, coordination and intervention costs, and L represents harms and losses. The weights wᵢₜ make distribution explicit across actors, populations or generations; δₜ is the chosen social discount factor. The expected value associated with earlier sensemaking-enabled action is then the difference between the trajectories:

$$\Delta V\ = \ \text{ENPV}(\pi_{1})\ - \ \text{ENPV}(\pi_{0})$$

The difference asks how much the expected consequences change when action changes. It does not value awareness directly. If renewal affects the chance or timing of intervention, the calculation must include the possible action times, including no action. An action hazard is the probability of first effective action during a period, conditional on none having occurred earlier. Appendix A supplies the details; the worked example below gives the comparison without requiring this notation.

The counterfactual must include the cost of building and maintaining the relevant capacity. Continuous sensing, cross-boundary translation, institutional memory, deliberation and coordination consume resources. Their value is positive only when the expected improvement in consequence trajectories exceeds those additional costs under the chosen social objective.

Hallegatte's analysis of hydro-meteorological services and early warning illustrates that such systems can produce large benefits relative to cost in bounded disaster settings, with estimated benefit-cost ratios from 4 to 36 under the paper's assumptions (Hallegatte, 2012). This is an empirical analogue for the value of observation, communication and action capacity, not an estimate of the value of climate sensemaking in general.

When the Accounting Boundary Is Wrong

The relevant calculation is often not the business case of the organisation asked to act. A firm, agency or municipality can face immediate expenditure while a large share of the avoided harm accrues to other organisations, distant populations, ecosystems or future generations. The local net present value may therefore be negative even when the systemic social value is strongly positive.

For organisation j, a conventional business case includes benefits and costs recognised within its accounting and mandate boundary. The system comparison instead includes the people, institutions, technologies and environments whose consequences are materially altered. The two values can diverge because of externalities, public goods, fragmented authority and temporal mismatch. A decision architecture that requires each unit to justify action only through its own ledger can be systematically incapable of selecting the collectively valuable path.

The sensemaking perimeter provides a way to expose that mismatch. It does not imply that one central authority must replace distributed action. Ostrom's analysis of polycentric systems shows why multiple centres can support experimentation, mutual adjustment and learning (Ostrom, 2010). The requirement is that the consequence, information and incentive structures connecting those centres become sufficiently visible for locally reasonable action to contribute to the wider trajectory.

Rasmussen's critique of structural decomposition is equally relevant. When risk is analysed separately at each organisational level, control interactions and migration toward system boundaries can disappear (Rasmussen, 1997). A functionally defined perimeter asks instead who changes the hazard, who observes it, who can intervene, who bears the consequences and how information and authority move among them. The resulting social value may be distributed across a control structure that no single local business case represents.

This also explains why an organisation can be data-rich and rational within its own frame while remaining part of a system that fails to act. Its indicators may be calibrated to local variance, its targets to short planning cycles and its authority to proximate outputs. The aggregate consequence emerges from many such frames and becomes actionable only if a representation crosses them without losing provenance, uncertainty, distribution and intent.

A Worked Climate Illustration

Climate change is not introduced here as proof of Sensemaking Debt. It is a worked illustration of the valuation problem because the causal signal is extensively studied, the consequences span institutions and generations, and delay changes both cost and the set of feasible responses. The example should therefore be read as an analytical template rather than as a new integrated assessment model.

Consider two trajectories beginning from the same body of climate evidence. In the earlier path, institutions aggregate emissions, vulnerability, infrastructure and ecological signals across a perimeter that includes jurisdictions, firms, communities and future affected populations. The combined representation acquires sufficient legitimacy, mandate and capacity to trigger mitigation and adaptation while lower-cost options remain available.

In the delayed path, each organisation responds mainly to indicators within its own budget, election, reporting or investment horizon. Evidence continues to strengthen, but action remains partial until visible damages, regulation or market disruption cross local thresholds.

The comparison must include current mitigation and adaptation costs in the earlier path, not merely avoided future harm. It must also include transition costs created by delay, infrastructure lock-in, accumulated emissions, adaptation limits and losses that later expenditure cannot fully reverse.

IPCC Working Group III concludes that unabated fossil-fuel infrastructure can lock in emissions and that later removals cannot fully compensate for delayed action; Working Group II documents risks that can exceed adaptation limits and produce severe or irreversible losses (IPCC, 2022a, 2022b). These findings make timing part of the consequence, not merely a choice about when to pay the same bill.

Uncertainty does not eliminate value. It changes how it should be represented. Arrow and Fisher show that uncertainty combined with irreversibility can create quasi-option value in preserving the ability to choose after further learning (Arrow & Fisher, 1974). Earlier action may therefore be valuable not only through expected avoided damage, but through preserving ecological, infrastructural and institutional options that delay would remove. Climate tipping research reinforces the need to represent abrupt and potentially irreversible change rather than rely only on smooth central estimates (Lenton et al., 2019).

Discounting is unavoidable when costs and benefits occur at different times, but a single undisclosed rate can determine the conclusion. Arrow and colleagues argue for declining discount-rate schedules under uncertainty for intergenerational projects (Arrow et al., 2013). The National Academies' modular framework for the social cost of carbon likewise separates socioeconomic, climate, damage and discounting components so their assumptions can be examined (National Academies of Sciences, Engineering, and Medicine, 2017). A FlowSentience-informed analysis should therefore present sensitivity across defensible discounting approaches and distributional weights rather than hide ethical and empirical choices inside one number.

A calculation would begin with a justified consequence perimeter and explicit action-timing scenarios. Probabilistic losses and benefits would then be estimated for each path, alongside evidence about how renewal changes implementation. Testing discounting, distribution, tail-risk and irreversibility assumptions would reveal how sensitive the comparison is to its empirical and normative premises. The result remains conditional on those premises and on the causal link from collective understanding to action.

To make the comparison reproducible, consider a deliberately hypothetical climate-adaptation project protecting a group of communities and infrastructure assets. All amounts are millions of constant 2026 currency units. An early intervention costs 100 in year 0 and avoids expected damage of 200 in year 10. A delayed intervention costs 80 in year 5 and avoids expected damage of only 120 in year 10.

No intervention produces zero incremental benefit or cost relative to the no-action baseline. The reduced benefit under delay is an assumption representing a lost opportunity to protect the assets; it is not an empirical climate estimate.

$$v(E)\ = \ - 100\ + \ 200/(1\ + \ r)^{10};\ \ \ \ \ v(D)\ = \ - 80/(1\ + \ r)^{5}\ + \ 120/(1\ + \ r)^{10}$$

Suppose the baseline probabilities of early, delayed and absent intervention are 0.2, 0.5 and 0.3. A renewal programme costing an additional 5 in year 0 changes them to 0.6, 0.3 and 0.1. These probability changes are hypothetical, not estimated effects of Sensemaking Debt. Conditional project outcomes are assumed unchanged; only the timing distribution changes. The project values above exclude the renewal programme, whose cost is subtracted once below.

$$\Delta V\ = \ (0.6\ - \ 0.2)v(E)\ + \ (0.3\ - \ 0.5)v(D)\ - \ 5$$

Table 1. Hypothetical incremental project values and the net value of renewal, in millions of constant 2026 currency units. Rates are sensitivity assumptions, not recommended social discount rates.

At 3%, the baseline expected project value is 19.91; the enhanced expected value after the renewal cost is 30.38. Their difference is 10.47. If the renewal programme did not change probabilities, action quality or available options, its incremental value in this example would be −5. A positive value therefore depends on a real conversion effect, not on describing the programme as sensemaking.

The accounting boundary changes the result as well. If the organisation paying the early intervention cost captures only 10% of its avoided damage, its local project value at 3% is −100 + 20/(1.03)¹⁰ = −85.12. The corresponding social project value is +48.82 because the remaining avoided damage accrues elsewhere within the stipulated perimeter.

Financing, transfers or shared mandates would be needed to resolve this divergence. This example illustrates distributed climate-adaptation benefits; it is not a calculation of the global value of mitigation or a claim about the probability of a specific catastrophe.

A real appraisal would replace each assumed amount and probability with documented evidence or an explicit scenario range. It would include recurrent costs, residual value beyond the horizon, uncertainty in hazard and effectiveness, and the distribution of gains and losses. Carbon values and separately modelled avoided climate damages must not both credit the same benefit. Option value likewise requires an explicit treatment of the decisions preserved; it should not be added as an unsupported premium.

The example also shows why the absence of a local alarm is itself analytically important. No single organisation may observe a sufficient share of the aggregate consequence within its own metrics. The missing function is not necessarily more data. It is a process capable of maintaining the relationship among partial signals, affected actors, historical evidence, contested values and implementable authority across time.

Recursive self-improvement makes the available choices part of the comparison. A later system might offer cheaper protection or better evidence; meanwhile, delay might allow new dependencies to become difficult to reverse. A credible appraisal should compare alternative development and governance paths, including benefits forgone by slowing adoption. Early renewal can be valuable when it preserves practical opportunities for correction. That value must be modelled through the choices and consequences preserved, not added as an automatic premium for caution.

Limits of Monetisation

A monetary comparison can help without representing everything that matters. Rights, ecological limits and distribution may rule out an option with a positive aggregate value. Rare but severe outcomes may be obscured by an average. Appraisal therefore also needs physical consequences, scenario ranges and an account of who gains and loses. Arithmetic cannot decide who has legitimate authority or how burdens should be shared.

Establishing causation remains demanding. Organisations that act earlier may also have more resources, stronger mandates or greater experience. Researchers must examine whether renewed understanding actually changes action, rather than attributing the whole difference to it. The valuation is a structured comparison whose assumptions require evidence.

The educational application extends the relevant horizon. Investment in reading, reasoning, source comparison and supported participation may alter the future distribution of interpreters who can recognise and challenge consequential claims. Estimating that value requires evidence linking an intervention to durable learning, later participation and specified outcomes. A PISA point must not be translated mechanically into a quantity of Sensemaking Debt, lifetime income or a monetary value of citizenship. Distributional effects, democratic participation and the preservation of meaningful choice should remain visible even when monetary estimates are unavailable.

Learning and the Renewal of Understanding Across Generations

Education raises a question about what present institutions leave available for future participants. A completed essay or course is visible now; the ability to examine an unfamiliar claim may become consequential later. Developmental understanding debt names a possible burden when learning opportunities repeatedly remain unfulfilled while later work assumes the capability exists. It is an institutional as well as a developmental question. Teaching, assessment, access to support and opportunities for participation shape that history. The term does not make a young person morally responsible for the conditions in which they learn.

What the Latest PISA Release Contributes

The OECD programme page identifies the publication of PISA 2025 results on 8 September 2026 (OECD, 2026b). Initial reporting describes continuing difficulties in attainment and raises questions about the relationship between students' uses of AI and their performance (Associated Press, 2026; The Verge, 2026). These are timely signals for the present inquiry. The new report's detailed tables and country estimates were not directly available for verification in this revision, however, so this paper does not reproduce new numerical estimates, rank countries or treat an initial journalistic account as a primary-data analysis.

The directly checked historical baseline is clearer. Across OECD countries, PISA 2022 reported a fall of 15 points in mathematics and 10 points in reading between 2018 and 2022; average science performance did not change significantly. The OECD also described a longer decline in reading and science predating that interval (OECD, 2023). These findings concern measured educational performance. They establish neither a quantity of understanding debt nor an explanation based on widespread generative-AI use, which cannot account for a decline already occurring before such use became widespread.

The theoretical implication is a question about institutional renewal. If foundational capabilities are difficult to sustain, later organisations may face a greater task of developing them. PISA does not establish this causal sequence. Differences in schooling, language, socioeconomic conditions and assessment must be considered. Reverse causality also matters: learners who struggle may seek AI assistance more often, so an association cannot establish that the assistance caused the difficulty.

PISA should consequently serve as a contextual indicator and a source of research questions. National averages can conceal substantial variation between learners, schools and forms of disadvantage. A decline in an average does not describe every young person, and a high score does not establish that an organisation can maintain sufficiently shared understanding. The framework developed here requires evidence about continuity, obligation and later constraint that an attainment survey was not designed to provide.

When a Completed Output Leaves Learning Unfinished

An educational understanding gap may concern a particular task: a learner cannot yet explain a relationship, recognise a misleading inference or connect an answer to the evidence on which it depends. Such gaps are normal features of learning. They become relevant to a debt hypothesis only when the opportunity to resolve them is repeatedly displaced while subsequent tasks assume the missing capability. A curriculum can move on, an assessment can reward an externally produced answer, and a later course can encounter an apparent lack of prerequisites whose history is no longer recoverable.

Different learning histories require different explanations. A capability may never have formed because the relevant activity was bypassed. An acquired capability may weaken through disuse. A learner may also perform well in a familiar setting but struggle to transfer the reasoning to a new one. A weak answer alone cannot distinguish these histories or establish which support would help.

Established learning research supplies a mechanism that predates generative AI. In two experiments with prose materials, Roediger and Karpicke (2006) found that practice retrieving material supported later retention more than an equivalent opportunity to restudy it, although immediate performance and confidence could favour restudy. The bounded implication is that the activity performed during learning can change what remains available later. Retrieval is not equivalent to comprehension or transfer, but its displacement is a plausible mechanism to examine rather than infer from a finished answer.

A fourth pathway lies in institutional recognition. A learner may have a difficulty that remains remediable, while successive institutions record competence as if the difficulty had been resolved. The record then becomes an inherited premise in decisions about placement, responsibility or support. What accumulates is partly the cost of reconstructing what the learner has had a fair opportunity to learn. This pathway connects individual learning with epistemic continuity: later actors need to know what earlier evidence did and did not establish.

The proposed developmental extension can be expressed accordingly: understanding debt is a path-dependent burden that may arise when consequential learning opportunities are repeatedly displaced or left unfulfilled and later activity presupposes the capabilities those opportunities were expected to form.

This is an analytical proposal. It needs discrimination from ordinary novice status, learning differences, insufficient instruction and changing demands. It should not become a label that transfers institutional responsibility to young people or treats legitimate reliance on accessible tools as a personal deficiency.

Assistance, Substitution and the Formation of Agency

Assistance can change both immediate performance and the opportunity to learn. The OECD's Digital Education Outlook 2026 distinguishes generic use of generative AI from educational uses designed around pedagogical purposes, and cautions against equating improved task output with learning. It also describes the potential of appropriately designed tools to support learning and teachers' work (OECD, 2026a). The distinction is central here: the effect depends on the activity, the learner and the role assigned to the tool.

Kosmyna and colleagues' essay-writing preprint (2025) provides one bounded warning signal. It reports neural and behavioural differences across tool-use conditions in a study with 54 participants in the first three sessions and 18 in a fourth session. It includes differences in recall and ownership of written work. This small study of an essay-writing task cannot establish a population trend among children, permanent neurological harm or the long-term effects of every educational use of AI. Its contribution to this report is to motivate direct assessment of what a person retains and can reconstruct after an assisted performance.

The recent evaluation argument by Chowdhury and Khan (2026) is relevant for a different reason. Their August preprint, consulted here at abstract level, questions what an output can evidence about learning in an AI-mediated setting. It is used as a conceptual prompt, with no empirical effect size attributed to it. Taken together, these sources support a more discriminating research design; they do not validate a single account in which all assistance accumulates debt.

A worked example, a translation, a reading aid or a well-timed hint may enable a learner to engage with reasoning that would otherwise remain inaccessible. The same interface can also make it easy to submit an answer without considering its premises. The distinction cannot reliably be made by counting prompts or minutes of use. It requires attention to who performs the consequential intellectual work and whether the learner can subsequently explain, evaluate or adapt what was produced. Accessibility should be preserved when these capabilities are assessed.

Agency also has a social setting. The opportunity to ask a question depends on whether it is welcomed, whether uncertainty carries a penalty and whether the learner sees a plausible route from effort to progress. Capability, endurance and willingness therefore need distinct treatment. Capability concerns the knowledge and practices needed for the task. Endurance concerns the time, attention and support available to sustain effort. Willingness concerns motivation, trust, perceived relevance and the social costs of participation. An exhausted learner and a learner who has never been taught a prerequisite may produce the same weak answer through different mechanisms.

Authority and opportunity remain institutionally distributed. Young people may understand an aggregate environmental consequence while having little influence over the decisions that produce it. Their inability to secure timely action must not be redescribed as an inability to understand. The decision-conversion framework should locate the blockage accurately: the problem may lie in adult mandates, conflicting incentives or the absence of a channel through which young people's knowledge can affect collective choices.

Improving AI assistance can affect this relationship in either direction. If easier answer production repeatedly displaces practice, later dependence may make further substitution more attractive. If a tutor becomes better at identifying prerequisites and supporting the learner's own reasoning, assistance may strengthen future independence and informed reliance on others. These are developmental feedback hypotheses, distinct from technical recursive self-improvement. PISA cannot identify either process from an attainment score. Studies need to follow what the learner actually does and can subsequently explain or adapt.

From Educational Obligations to Organisational Renewal

The cross-level proposition becomes concrete at transitions. A newcomer is expected to evaluate a model, interpret a policy or exercise professional judgement. An employer may rely on a credential as evidence that the relevant foundations exist, while the educational institution may have relied on outputs that did not reveal independent understanding. If the newcomer must conceal uncertainty to maintain credibility, the organisation can lose an opportunity to identify and repair the gap. Repeated across roles and over time, this could contribute to the organisational condition described earlier.

This causal account must compete with alternatives. A newcomer can struggle because induction is poor, instructions are contradictory, responsibilities are unclear or the work has changed. The organisation may fail to provide the domain-specific experience that no school could reasonably supply. Evidence for developmental carry-forward would therefore need to identify a prior learning obligation, show how it was treated as fulfilled, trace its consequences and examine whether targeted renewal changes the trajectory. A general complaint about the younger generation provides none of that evidence.

Institutions can also interrupt the proposed process. Protected time for explanation, guided practice, constructive disagreement and progressive responsibility may make uncertainty discussable before it becomes expensive to reconstruct. Documentation can record which judgments were supported by direct evidence and which depended on assistance. Mentors can make inherited intentions visible while inviting newcomers to challenge them. These arrangements connect capability formation to state, epistemic and intent continuity without requiring identical understandings or universal self-sufficiency.

Figure 8 represents the distinction between supported formation and repeated displacement. The branches describe candidate trajectories, not a deterministic model of technology effects. The same learner or institution can move between them, and well-designed assistance can contribute to recovery.

What Would Count as Evidence of Recovery?

A defensible investigation would follow learners across time and across tasks. It would observe assisted performance, later recall or explanation, transfer to an unfamiliar problem, calibration of confidence and the ability to identify misleading evidence. It would also document instruction, accessible supports and the opportunities the learner actually received. Repeated measures could then ask whether an earlier unresolved obligation predicts a later constraint and whether a targeted intervention changes that relationship.

Recovery would have to mean more than improved immediate output. A learner might become better able to reconstruct an argument, ask a more informative question or recognise when help is needed. At the institutional level, recovery might appear as more accurate recognition of prerequisites, less repeated remediation and more effective transitions into further study or work. Learning and organisational outcomes should be analysed separately before a cross-level relationship is claimed.

Where comparative interventions are feasible, designs should report baseline differences, attrition, implementation quality and the distribution of effects. Classroom or institutional clustering and access to support need explicit treatment. Removing generative assistance may be appropriate for a narrowly defined assessment of independent reasoning, while disability-related access arrangements remain in place. The aim is to identify the capability a particular task requires, not to establish unaided performance as a universal norm.

The ethical and practical implication follows from the theory itself. An institution concerned with understanding debt should make learning obligations easier to recognise and fulfil. It should not create a new surveillance system in which young people's uncertainty becomes a permanent risk score. Transparent formative assessment, meaningful learner participation and proportionate use of evidence are more consistent with the renewal of trust and interpretive agency.

Reinterpreting Organisational Complexity and Change

The earlier chapters have followed a recurring question: what does an adaptation leave for those who must understand and change it later? Organisational complexity can be reconsidered through that question. New functions, platforms and partnerships may add valuable capability while changing the dependencies that require interpretation. Whether they produce lasting constraint depends on what happens to the resulting obligations.

A programme office, steering group or integration role may help participants examine those dependencies. It may also keep recurring differences manageable without changing why they recur. The visible arrangement does not determine the interpretation. Researchers need to follow whether its work preserves accessible reasoning and reduces repeated reconstruction.

The same visible mechanism can therefore support renewal or service accumulated debt. Its significance lies in its trajectory: whether it makes later coordination easier, or whether the organisation becomes progressively more dependent upon it merely to preserve current coherence.

Digital transformation displays the same ambiguity. Integrated platforms, shared data and advanced analytics can make relationships visible that were previously inaccessible. They may reduce duplication, improve continuity and connect actors across organisational boundaries. Yet digital integration also creates new dependencies upon data definitions, system architecture, access rights and platform ownership.

Different organisational units may observe the same activity through increasingly sophisticated but partially incompatible representations. One function interprets performance through financial data, another through operational flow, a third through risk and a fourth through customer or clinical outcomes. Each representation may be locally valid. The interpretive challenge arises when their differences affect shared consequences but no process exists through which they can be examined together. Sophisticated data systems can produce several precise but incompletely connected accounts of organisational reality.

Technical integration alone cannot establish interpretive connection. Several separate systems may work together through understood interfaces, while one shared platform may contain categories whose users interpret them differently. The consequence for coordinated action is the object of inquiry.

Organisational learning can be reinterpreted in a similar way. Learning remains indispensable to adaptation, but every lesson, practice and specialised capability alters the context within which future learning must occur. Knowledge accumulates together with the relationships needed to integrate it. An organisation can therefore become more knowledgeable locally while finding it increasingly difficult to reconstruct what its distributed knowledge means collectively. The problem is not excessive learning. It is insufficient renewal of the connections among what different parts of the system have learned.

The same reasoning limits explanations based on size. What requires attention is the set of interdependent perspectives, not simply the number of employees. Stable interfaces can contain that task; changing responsibilities can enlarge it without adding anyone.

Organisational form consequently shapes how complexity is experienced. Standardised systems may coordinate effectively while their rules remain appropriate, but struggle when embedded representations become obsolete. Professional organisations may combine deep expertise with persistent disagreement over evidence, quality and authority. Mission-driven systems may achieve strong internal coherence while filtering out signals that challenge collective identity. Network and project structures may reconfigure rapidly while repeatedly losing relational continuity and organisational memory. These are not fixed organisational types with predetermined outcomes. They illustrate how the same level of formal complexity can produce different debt trajectories depending upon how interpretation, authority, memory and interdependence are organised.

The concept of the sensemaking perimeter further complicates the conventional image of organisational complexity. The relevant adaptive system may not coincide with the legal entity or organisation chart. Suppliers, regulators, contractors, professional communities, platforms and affected citizens or customers may all shape the consequences being coordinated. An organisation may therefore appear internally coherent while participating in a wider system whose interpretations remain fragmented. Local optimisation can strengthen one unit while increasing debt across the relationships connecting several organisations. Conversely, apparently inefficient internal coordination may sometimes reflect necessary work to maintain understanding across a broad and permeable perimeter.

The same boundary question applies to organisational change, including changes that precede formal reorganisation. Formal reorganisation is only one way in which an organisation changes. Practical authority, expertise and coordination may shift before reporting relationships are updated. New technologies, outsourced services, platforms and informal networks can redistribute influence while the formal organisation continues presenting itself in unchanged terms.

Artificial intelligence makes this discrepancy especially visible. AI may reorganise an institution before the institution formally reorganises itself. Models can influence which cases receive attention, which risks appear significant and which alternatives are considered legitimate. Data and platform functions may acquire interpretive authority because they define the categories through which organisational reality becomes visible. Automated workflows and agents may redistribute practical decision rights across functions without any explicit decision to redesign the organisation. The organisation chart can therefore become an increasingly incomplete representation of how interpretation and action actually occur.

This does not mean that AI-driven change necessarily increases Sensemaking Debt. AI may clarify dependencies, connect distributed observations and make practical authority more visible. It may support the reconstruction of organisational memory and reveal contradictions among formal structures, technological systems and actual work. It may also produce the opposite effect. Authority can migrate without corresponding accountability. Different functions may adopt incompatible AI-generated representations. Machine-mediated decisions may change relationships faster than affected actors can collectively understand them. Fluent synthesis may produce an appearance of coherence while underlying differences remain unresolved.

Later structural reform may either clarify these emerging relationships or merely formalise their fragmentation. In the language developed earlier, it may repay debt by reconstructing understanding and aligning responsibility with practical influence, or refinance debt by placing new structures over unresolved interpretive divergence.

The same distinction applies to organisational change more broadly. Stability is not necessarily resistance, and restructuring is not necessarily adaptation. Stable arrangements may preserve relationships and memory through which understanding is renewed. They may also retain obsolete categories and authority structures. Reorganisation may release inherited constraints, or it may disrupt the relational continuity required for recovery.

Repeated change can become particularly problematic when new arrangements are introduced before shared understanding of the previous change has stabilised. Actors continually reinterpret roles, relationships and priorities while unresolved dependencies persist beneath successive organisational forms. The organisation appears highly active and responsive, yet its practical coordination depends increasingly upon informal memory and compensatory work. Prolonged structural stability can produce the opposite trajectory. Existing relationships preserve coherence, but formal arrangements drift progressively away from technological, environmental or institutional reality. The system may continue functioning through local adaptation while its official representation becomes less capable of explaining how action actually occurs. One organisation repeatedly refinances its interpretive obligations. Another locks them into structures that become progressively harder to revise.

These comparisons direct attention to whether differentiated capabilities remain intelligible and revisable across their consequential relationships. Simplification may help in a particular setting, but it cannot replace the specialised knowledge on which complex adaptive systems depend.

Inquiry can follow what a new structure or technology leaves to later actors: the obligations it creates, the context it preserves, and the opportunities it provides to revise interpretation. It must also examine whether formal boundaries correspond to how consequences are produced. For AI, the comparison includes the relative pace of decision-making and collective interpretation. These questions do not establish that governance growth, digitalisation, restructuring or AI adoption constitute Sensemaking Debt. They provide a way of examining whether such developments are changing the relationship between interpretive demand and the renewal of collective understanding.

Governance, specialisation, digital systems, learning and structural change remain important sources of capability. The concern is how the relationships created through them can remain open to reconstruction as the system develops.

Sensemaking Debt may therefore offer a way of connecting observations that otherwise appear unrelated. Governance proliferation, digital fragmentation, repeated restructuring, dependence upon informal translators and declining responsiveness to novelty may sometimes form part of the same trajectory. In other situations, identical observations may reflect healthy integration and renewal. The difference cannot be inferred from their presence alone. A succession process illustrates how these relationships can cross institutional boundaries. A school may be rewarded for completion, a university for throughput and an employer for immediate productivity. Each may have a reason to defer the slower work of explanation and supervised practice to the next institution. If later actors must repeatedly compensate for those omissions, the system may carry an obligation that no single unit owns. This is a candidate mechanism, not an established account of current educational decline. Its investigation requires tracing the same obligations across transitions and showing that the proposed account explains something beyond unequal resources, selection or ordinary changes in task demands.

This leads naturally to a research agenda. If Sensemaking Debt is to develop beyond a conceptual proposition, future work must clarify its boundaries, examine how accumulation varies across organisational forms, distinguish compensatory activity from recovery and determine whether credible empirical inferences can be constructed across human and human-AI systems.

A Research Agenda for Collective Understanding

The research agenda asks what evidence would establish, narrow or challenge the proposed account. Its connected lines of inquiry concern the organisational history, the consequences of action and the formation of future understanding. They include comparisons in which similar difficulties are explained without Sensemaking Debt.

Conceptual Clarification

The first task is to specify what counts as a consequential interpretive obligation, what it means to carry one forward and what later constraint makes that history theoretically relevant. Cases should distinguish a temporary gap, a persisting disagreement, a recoverable record and an inherited arrangement whose rationale can no longer be reconstructed. Useful disagreement and justified reliance on others must remain compatible with sufficiently shared understanding.

Conceptual comparisons should examine organisational memory loss, common-ground breakdown, technical debt, cognitive debt, intent debt, functional stupidity and resilience erosion. Incremental value would be demonstrated if the Gap–Debt distinction explains a longitudinal pattern that these concepts alone leave insufficiently specified. Evidence that an established account explains the same pattern more economically should count against a separate construct.

Accumulation and Recovery

Longitudinal cases should reconstruct sequences of decisions, temporary workarounds, staff transitions and subsequent attempts to revise inherited arrangements. The central question is whether unresolved interpretive obligations persist beyond the events that produced them and alter the cost or possibility of later adaptation. Researchers should investigate thresholds, delays and changes in the composition of constraint rather than presume smooth linear accumulation.

Recovery offers a particularly informative comparison. An intervention that restores provenance or clarifies an inherited purpose might reduce later reconstruction effort, while a structural change with similar resources might merely relocate the obligation. Such comparisons require attention to changes in staffing, workload, authority and task difficulty. Recovery can also require retiring an obsolete obligation; it need not restore an earlier consensus.

Organisational Form, Boundaries and Scale

Research should compare legally defined organisations with consequence-dependent perimeters such as care pathways, infrastructure networks and supply relationships. Boundary choices should be recorded before outcomes are interpreted and tested against plausible alternatives. A wider perimeter may reveal externalised costs while creating unrealistic expectations about how much interpretation must be connected.

Effective sensemaking scale should be examined through heterogeneity and consequential interdependence as well as participant numbers. Modularity, professional boundaries, turnover and distributed authority may explain why similarly sized systems experience different burdens. No universal threshold should be inferred from headcount or from a biological analogy.

Empirical Observation

Observation should combine interviews, decision histories, records of assumptions and evidence about actual work. Repeated requests to reconstruct a rationale, divergent accounts of the same dependency and recurrent compensatory translation can identify episodes for inquiry. They cannot independently diagnose debt. Negative cases in which such patterns accompany healthy integration are essential.

Analysts should seek temporal ordering and rival explanations, including resource scarcity, strategic disagreement, unclear authority and deliberate non-compliance. Where participants retrospectively describe a debt, contemporaneous records should be used to test that account. The concept must not be confirmed simply by teaching it to participants and then coding their responses in its own language.

Operationalisation and Measurement

Operationalisation means specifying how a concept will be examined in a particular study. Researchers could identify a consequential decision and assess how readily its dependencies, evidence and purpose can be reconstructed. Time spent matters only alongside the reason for it: legitimate deliberation must be distinguished from repeatedly rediscovering an inaccessible rationale.

A proposed measure needs reliability, meaning reasonably consistent results under comparable conditions, and discriminant validity, meaning that it distinguishes debt from related but different problems. It should also respond to genuine recovery. A precise-looking index does not establish these qualities. Several dimensions and documented case judgments may prove more defensible than one score.

Computational Modelling

Computational models can explore whether assumed local processes produce persistent constraints across a wider system. They should separate the gap from carry-forward and recovery, and compare different arrangements for memory, authority and interaction. The notation in Appendix A supports bounded thought experiments; its parameters have not been calibrated against organisational observations.

Simulation results should be reported as implications of assumptions. Models ought to include circumstances in which increased connectivity raises burden, deliberate simplification helps adaptation, or dissent prevents premature convergence. Parameter sensitivity and alternative structures matter more initially than fitting a single aggregate trajectory to a flexible model.

Human–AI Adaptive Systems

Studies should specify the task and the role of AI: drafting, retrieval, explanation, tutoring, recommendation, execution or autonomous coordination. They should examine both the obligations the system creates and the human capability retained to contest its representations. Decision speed, output quality and collective understanding are separate outcomes.

Comparisons should assess whether provenance, alternative interpretations and uncertainty remain available when a familiar situation changes. They should also examine whether users have the time and authority required for meaningful oversight. A system may reduce routine effort and increase renewal, or conceal a growing dependence on explanations nobody can evaluate. The relevant design features and organisational conditions must be identified rather than attributed to AI adoption in general.

Self Improvement and the Possibility of Correction

A testable extension would compare systems with similar task performance but different histories of self-modification and renewal. One hypothesis is that consequential updates accepted without recoverable context predict more effort or fewer feasible options in later correction. A second is that usable provenance and independent checks weaken this relationship. Update frequency alone is insufficient: the significance and reach of changes, staffing, workload and task difficulty must also be recorded.

The comparison should distinguish the agent, its underlying model and the evaluator, including which of them can change and who authorises adoption. Outcomes might include the ability to reconstruct a past decision, detect a changed assumption and carry out a justified correction within the relevant time window. Cases where self-improvement simplifies the system or strengthens recovery would test the positive branch. A result explained entirely by ordinary maintenance demands, with no carried-forward interpretive constraint, would narrow the debt claim.

Adaptation Across Institutional Levels

Research across teams, organisations and wider institutions should trace how locally reasonable decisions alter obligations elsewhere. Multi-level investigation is needed to distinguish true recovery from displacement to a supplier, public agency, household or future participant. The same mechanism should not be presumed to operate unchanged across these settings.

Comparative cases of shared infrastructure, public health or environmental adaptation may reveal how distributed centres maintain enough connectivity to act while preserving local knowledge. The research should document who defines the perimeter, whose evidence is recognised and which interests remain outside the formal decision. Power and legitimacy form part of the explanation.

Intertemporal Consequences and Decision Conversion

Valuation studies should begin with a consequence and a decision that could alter it. They should estimate how an intervention changes signal recognition, understanding, commitment, practical authority and implementation, including the possibility that action never occurs within the horizon. The effect on timing must be distinguished from effects on action quality and on the available option set.

A valuation study should disclose local and wider social consequences, uncertainty and its treatment of time. It should count renewal costs once. Longitudinal evidence, or comparisons that help separate an intervention's effect from other changes, must test whether interpretation actually mediates the outcome. A favourable expected value cannot establish the existence of Sensemaking Debt.

Learning, Maturation and Institutional Renewal

The developmental extension requires following learning obligations across tasks and transitions. Studies should distinguish capability never formed, capability weakened through disuse, limited transfer and inaccurate institutional recognition of competence. Assisted output, later explanation, transfer, confidence calibration and source evaluation should be measured separately. Disability-related access and language supports should be maintained where they are not the construct under examination.

Research should compare assistance that supports reasoning with uses that displace it, allowing for prior knowledge and teaching conditions. Selection means that different learners may choose different tools; attrition means that some participants leave a study; classroom clustering means that pupils share conditions within a class. These can affect an apparent result. PISA can locate population patterns, but a claim about developmental carry-forward needs evidence of a learning history and its later consequences.

Research Strategy

An initial programme could follow contrasting cases over time, including cases where apparent warning signs accompany successful renewal. It could then examine specific interventions. Preregistering a hypothesis means recording it before inspecting the outcome; independent replication asks whether another investigation obtains comparable evidence. Both help prevent a flexible metaphor from being fitted to every result.

These inquiries can develop at different speeds. A consequence model may be useful before a debt measure exists, while the developmental extension remains tentative. The claims should preserve those differences. Evidence that established explanations account for an observed history more simply would be a valuable result.

Discussion and Limitations

The central uncertainty is whether Sensemaking Debt explains an accumulated condition that existing theories leave insufficiently clear. Organisational learning, common-ground research and accounts of memory or resilience may already explain a particular case. The proposed construct earns a place only where its distinctions improve the explanation of a history or difference between cases.

Meeting load, repeated translation and decision delay can arise through healthy integration, resource scarcity or necessary deliberation. They acquire relevance to this theory only when evidence connects them to unresolved obligations and later restrictions on adaptation. Their presence alone establishes neither debt nor poor management.

A central limitation concerns the definition of the adaptive system itself. Sensemaking Debt is proposed as belonging to the relevant sensemaking perimeter, but that perimeter is analytical and consequence-dependent rather than objectively given. Different boundary choices may produce different interpretations of required understanding, existing divergence and apparent recovery. A process may appear coherent when examined within one department but fragmented when its dependencies across suppliers, technologies and regulatory institutions are included.

Conversely, a broad system boundary may imply a need for collective understanding that no viable form of coordination actually requires. Researchers must therefore justify why particular people, technologies and institutions belong within the perimeter being studied and which consequences make their interpretations interdependent.

A legal organisation and the care pathway in which it participates may therefore yield different findings. Their boundaries include different dependencies. An improvement can restore understanding within one while leaving an unresolved obligation elsewhere.

A related limitation concerns scale. The report has moved from biological and individual adaptation to teams, organisations, inter-organisational systems and civilisation-scale coordination. This movement supports the possibility of a general adaptive relationship, but it does not establish that identical mechanisms operate at every level.

A clinical team can reconstruct shared understanding through direct interaction in ways unavailable to an international governance system. Authority, relational density, organisational memory and the possibility of coordinated intervention vary substantially across scales. The theory therefore proposes possible continuity in the relationship between interpretive demand and renewal capacity, not identity of the organisational processes through which that relationship is maintained.

Effective sensemaking scale remains provisional. The suggested dimensions have not been shown to form a single measurable variable. Neither a biological analogy nor a headcount threshold can substitute for examining which interpretations need to remain connected in the case at hand.

There is also a normative ambiguity within the concept of sufficiently shared understanding. More shared understanding is not necessarily better. Adaptive systems benefit from interpretive diversity, dissent and competing explanations. Efforts to maximise coherence may suppress weak signals, professional judgement or perspectives held by actors with less institutional authority.

A system displaying visible disagreement may therefore possess stronger renewal capacity than one displaying apparent consensus. Shared understanding remains adaptive only when differences can be recognised, examined and revised. The theory must consequently distinguish sufficiently connected diversity from imposed uniformity. Otherwise, Sensemaking Debt could be misused to legitimise conformity, centralisation or the removal of productive disagreement.

Authority affects that examination. People may remain silent because questioning an account threatens their role or credibility. Apparent agreement is therefore ambiguous evidence about the condition of collective understanding.

No validated unit or diagnostic threshold currently exists for Sensemaking Debt. Combining indicators does not by itself solve this problem. The equations in Appendix A separate possible processes so their assumptions can be examined; they do not turn those processes into established measurements.

A future formalisation would need to determine how required and sustained interpretive connectivity could be inferred, whether different forms of residual constraint can be aggregated and how recovery interacts with the amount and composition of existing debt. It would also need to account for thresholds, delays, reinforcing feedback and the possibility that recovery in one part of a system shifts costs elsewhere.

The proposed cause can also be affected by the response to it. This is what endogeneity means here. Difficulty reconstructing shared context may prompt centralisation or AI adoption, which then changes the conditions of interpretation. A later association between AI use and coordination difficulty cannot identify the direction of causation by itself. Studies need to reconstruct the sequence and compare alternative explanations.

Recovery may require retiring an obsolete arrangement or abandoning an earlier interpretation. It therefore cannot be judged solely by return to a previous state. The recovery term in Appendix A represents renewed ability to understand and revise consequential relationships within the specified model.

The report has also treated artificial intelligence as a significant influence upon the accumulation-recovery dynamic. This discussion is limited by the heterogeneity of the category. A local writing assistant, a predictive decision-support model, an enterprise knowledge system and a network of autonomous agents affect interpretive demand, authority and recovery in fundamentally different ways. "AI adoption" cannot therefore function as one meaningful causal variable without substantial differentiation.

AI may also be endogenous to existing debt. Organisations with fragmented knowledge and growing coordination costs may adopt AI specifically to compensate for those difficulties. Subsequent changes in coordination cannot then be interpreted as effects of AI alone. The technology enters a system already shaped by unresolved interpretive conditions and may reinforce, conceal or reduce them.

More measurement may reveal a pattern while adding further claims that people must interpret. An AI classification also reflects the examples and assumptions used to construct it. Training a system to detect presumed signs of debt and then treating its classifications as proof of debt would be circular. Independent records and alternative explanations remain necessary.

Fluent summaries introduce a related ambiguity. People may accept the same account while holding different assumptions about evidence or responsibility. Researchers need to examine whether participants can question and revise the account, as well as whether they can retrieve it.

Human-in-the-loop usually means that a person participates in a particular decision; human-on-the-loop refers here to oversight of a process that can act without approval at every step. Either arrangement needs practical time, context, expertise and authority to support intervention. Responsibility on paper is insufficient. In a distributed system, that capacity may be shared across roles without any participant understanding every internal operation.

Recursive self-improvement sharpens these questions without resolving them. Bounded coding results cannot establish an organisational accumulation mechanism, and performance improvement need not increase interpretive burden. The relevant history may be explained by technical maintenance, resource shortages or changing goals. The debt account requires additional evidence that unresolved interpretive obligations persist and constrain later correction. It should not depend on a forecast of unrestricted self-improvement or catastrophic loss of control.

Finally, Sensemaking Debt may prove not to be one unitary latent condition. Empirical work may reveal several related but distinct latent conditions, such as fragmented representations, loss of organisational memory, dependence upon compensatory translation, identity-based filtering and reduced capacity to revise governing assumptions. These conditions may interact without being reducible to one common quantity.

In that case, Sensemaking Debt may function more effectively as a family resemblance concept or a theoretical bridge. Such an outcome would preserve a possible integrative value while requiring the claims about a coherent latent condition to be narrowed. The present report does not claim a validated state variable.

The research may identify a distinct condition, several related constraints or a useful way to organise historical inquiry. Each result would require different claims. What must survive comparison is the proposed distinction between a present difficulty and the accumulated history that restricts later understanding and change.

Practical use should remain proportionate to that uncertainty. The concept can guide questions about lost context and recovery before it supports any diagnosis or numerical assessment.

The intertemporal valuation developed in this report adds further limitations. It does not make Sensemaking Debt directly measurable in money. It compares outcomes under constructed action trajectories and therefore depends upon causal assumptions about how collective understanding changes attention, readiness, authority and implementation. These mediating effects must be observed rather than presumed.

Counterfactual models can also create an illusion of precision. Long-horizon estimates are sensitive to discounting, damage functions, distributional weights, tail risks and the treatment of irreversible or non-market loss. A single expected value can conceal ethically and empirically consequential differences among populations and generations. The appropriate response is transparent scenario and sensitivity analysis, not the claim that uncertainty makes comparison impossible.

Boundary selection remains decisive. A locally negative business case and a positive system value may both be correct within their respective accounting frames. The analytical contribution lies in exposing that divergence and the institutional conditions that sustain it. Arithmetic alone cannot decide which perimeter is legitimate, who should bear present costs or how authority should be constituted.

The climate illustration is consequently not an empirical test of Sensemaking Debt. It demonstrates how a known aggregate trajectory can fail to trigger timely action when observation, incentives and authority are decomposed. Establishing that unrecovered interpretive obligations caused a particular delay would require longitudinal evidence capable of excluding alternative explanations such as resource scarcity, strategic conflict, vested interest or deliberate non-compliance.

The educational extension adds a further level-of-analysis limitation. Organisational Sensemaking Debt is not a diagnosis of a child, a synonym for low attainment or an explanation of every difficulty in learning. A developmental understanding debt would have to be demonstrated through a history of missed or displaced learning opportunities and their later consequences.

PISA compares populations and cohorts under an assessment framework; it does not observe the proposed accumulation process in an individual. New associations involving AI use are especially vulnerable to reverse causality, differences in purpose of use and unequal access to support. Neither initial reporting of PISA 2025 nor an adult essay-writing experiment establishes that generative AI caused population-wide educational decline.

Conclusion

An organisation can possess accurate information while struggling to connect what its participants understand. Sensemaking Debt is a provisional account of the history that may make this difficulty persist. A gap concerns a present mismatch; debt concerns unresolved obligations repeatedly carried forward until they restrict later interpretation and change. Those obligations may remain in rules, technologies and relationships. State, epistemic and intent continuity direct attention to the context needed to examine them: how an arrangement emerged, what informed it and which purpose it served.

The Principle of Adaptive Balance proposes responsive renewal of understanding as one enabling condition of viability. The task depends on which perspectives must connect and which consequences make their differences important. Reorganisation may help when it restores usable context and changes the conditions that made obligations persist. AI can likewise support renewal through sensing, translation and accessible memory, while its recommendations and actions may create further dependencies to understand. Their effects require examination across the relevant sensemaking perimeter.

Recursive self-improvement makes continuity part of evaluating improvement itself. A system may change its capacity to act and the means used to judge those changes. The proposed requirement is to preserve the ability to examine consequential differences and revise the developing arrangement while meaningful alternatives remain available. Where improvement strengthens that ability, it may support renewal; where later development repeatedly inherits unresolved obligations, it may contribute to the condition described here. Demonstrated self-modification motivates this comparison but does not validate the debt construct.

The consequences of earlier understanding require a separate analysis. Comparing plausible paths of earlier, delayed and absent action can reveal benefits and losses beyond the organisation asked to pay. Climate change illustrates why a locally reasonable decision may leave a wider consequence insufficiently addressed. Such comparisons must include implementation, uncertainty, distribution and the options that delay may remove. They value changes in outcomes; they do not price understanding itself or validate a debt measure.

Understanding also has to be formed. The developmental extension asks what happens when learning opportunities repeatedly remain unfulfilled while later institutions assume the capability exists. PISA provides context for this inquiry, but cannot measure that accumulated history. Assisted performance, durable learning and collective adaptation remain distinct outcomes. Across both organisational and educational settings, evidence must follow obligations and their later effects, examine recovery and test alternative explanations. The construct may prove distinct, require narrower claims or yield to a more economical established account.

The practical concern is whether people can reconstruct and question consequential arrangements while they still have meaningful opportunities to change them.

Authors' Reflection

This inquiry began with recurring practical difficulties in reconstructing why organisations were arranged as they were, how consequential decisions had been formed and which purposes inherited constraints were still serving. The concept of Sensemaking Debt emerged as a way to ask whether those difficulties shared a temporal structure. It should remain answerable to the situations that prompted it and to evidence that may challenge its explanatory value. The present formulation has developed through engagement with organisational sensemaking, learning, cybernetics, resilience, boundary inquiry and human–AI research. Its proposed contribution lies in connecting momentary gaps, carried-forward interpretive obligations and later constraints. The authors claim neither ownership of the phrase nor discovery of all the phenomena it brings together.

The FlowSentience programme’s earlier publications provide an intellectual context for this work (Windahl & Johansson, 2025, 2026a, 2026b). They are not independent empirical tests of the construct. Similarly, the valuation model and the developmental extension introduced here are proposals for investigation. They should be revised or abandoned where more established explanations account for the evidence.

The educational extension makes that responsibility particularly clear. A language intended to reveal unfinished institutional obligations must not become a convenient way to blame young people for the conditions under which they learn. The value of the concept would lie in making those conditions more open to inquiry, support and renewal. The next stage belongs to a wider research community. Comparative cases, disagreement, replication and careful observation may establish a distinct accumulated condition, a family of related constraints or a useful bridge among existing theories. Each outcome would be more informative than protecting the original formulation from criticism.

Appendix A: An Initial Heuristic Formalisation

This appendix gives the formal detail behind three questions: what is carried forward, when effective action occurs, and how its consequences compare across time. The main argument does not require the notation. These are bounded analytical devices with specified assumptions, not validated measurements of Sensemaking Debt.

A Bounded Representation of Carry-Forward

Let gₜ be a context-specific, non-negative representation of consequential interpretive mismatch. Its definition must account for the form of connectivity required: a gap can arise through insufficient connection, concealed divergence or premature convergence. A simple deficit between required and sustained amounts is therefore inadequate as a general definition. Any scalar used in a model requires an explicit, locally justified mapping from observations.

$$D_{t + 1}\ = \ \text{max}(0,\ D_{t}\ + \ a_{t}\ - \ \rho_{t})$$

Dₜ is a heuristic representation of accumulated constraint, aₜ the amount carried forward during the interval, and ρₜ recovery during that interval. A bounded model might specify aₜ = κₜgₜ, with κₜ representing carry-forward propensity. The terms must have compatible units within that model; no such units are validated for the theoretical construct. The non-negative boundary prevents the heuristic from producing negative debt. It does not identify a debt-free real-world organisation or establish that constraint is one-dimensional.

For a self-improving system, a bounded model would also need a version history and a record of evaluation and adoption decisions. An update may alter the gap, the propensity to carry obligations forward or the means of recovery. Treating those influences as fixed would be an assumption to test. The present notation introduces no universal recursion coefficient and implies no exponential law of accumulation.

Action Timing, Including No Action

For policy π, let τ be the first effective intervention period. τ may take any value from 0 to T, or ∞ when intervention has not occurred within the modelled horizon. Hₜ denotes relevant history and environmental information available at t. Define the conditional action hazard as:

$$h_{t}^{\pi}\ = \ P(\tau\ = \ t\ |\ \tau\ \geq \ t,\ H_{t},\ \pi)$$

$$p^{\pi}(t)\ = \ h_{t}^{\pi}\ \prod_{s = 0}^{t - 1}\ (1\ - \ h_{s}^{\pi});\ \ \ \ \ p^{\pi}(\infty)\ = \ \prod_{s = 0}^{T}\ (1\ - \ h_{s}^{\pi})$$

With the empty product equal to one, these probabilities sum to one for a specified hazard sequence. If the hazard depends on stochastic histories, the products are evaluated along those histories and integrated over their distribution. A single product of unconditioned average hazards generally cannot substitute for that calculation. Later intervention beyond T belongs in a continuation or terminal-value model rather than being silently assigned zero consequence.

The visibility–understanding–readiness–mandate–implementation chain in the main text decomposes one specified joint pathway conditional on history and on no earlier action. It is not automatically the overall action hazard: action may arise through alternative pathways, and readiness and understanding may develop recursively. Empirical models must specify how such pathways combine and avoid counting overlapping routes twice.

Consequence Values and Cost Accounting

$$\text{ENPV}(\pi)\ = \ E\lbrack\ \Sigma_{t = 0}^{T}\ \delta_{t}\ \Sigma_{i \in \mathcal{P}}\ w_{it}\ (B_{it}^{\pi}\ - \ C_{it}^{\pi}\ - \ L_{it}^{\pi})\ \rbrack$$

𝒫 is the specified consequence perimeter, δₜ a discount factor, and wᵢₜ explicit distributional weights. B, C and L denote non-overlapping benefits, costs and losses. If this all-inclusive expression contains renewal costs in C, those costs must not be subtracted again. A monetary calculation requires a stated price basis and dimensionless weights; a utility-based welfare model requires a separately specified utility function and consistent units.

$$\delta_{0}\ = \ 1;\ \ \ \ \ \delta_{t}\ = \ \prod_{s = 1}^{t}\ (1\ + \ r_{s})^{- 1}$$

The discount schedule rₛ, horizon, residual value, distributional weights and treatment of uncertainty are assumptions to disclose and test. A changing rate schedule should not be presented as equivalent to a constant rate. Values remain conditional on the normative and empirical choices defining the appraisal.

An equivalent accounting arrangement can separate the renewal programme. Let the policy-specific conditional value v be the expected discounted consequence value when action occurs at τ, including intervention and other outcome costs but excluding the renewal programme. Let the policy-specific quantity K be the expected discounted cost of that programme, under the same perimeter and valuation convention. Then:

$$V(\pi)\ = \ \Sigma_{\tau \in \{ 0,\ldots,T,\infty\}}\ p^{\pi}(\tau)v^{\pi}(\tau)\ - \ K^{\pi}$$

$$\Delta V\ = \ V(\pi_{1})\ - \ V(\pi_{0})$$

In words, the comparison adds the consequences possible under each action time, weighted by its chance of occurring, and subtracts the renewal programme's cost. If a developing AI system changes which actions are possible or how well they work, those consequences must also differ between policies. A timing-only comparison would then be inadequate.

The policy superscript on v matters. Stronger renewal may change action quality, the available options or the environmental histories associated with a given action time, as well as changing its probability. Only when those conditional values are demonstrably invariant can the comparison be simplified to a difference in timing probabilities multiplied by a common v(τ), minus the additional programme cost. Table 1 deliberately assumes that restricted case.

No equation here maps D directly into money or presumes that every improvement in interpretation changes action. Causal estimates are required for the mediating links. Uncertainty, alternative explanations and non-monetary consequences should remain visible alongside the arithmetic.

Appendix B: Conceptual Genealogy and Claim Status

This appendix is also a reference for readers returning to an unfamiliar term. The plain explanations below aid interpretation; the formal definitions preserve the scope and provisional status of the concepts. No entry should be read as a validated diagnostic category.

Sensemaking Debt

Preferred term: Sensemaking Debt. Status: provisional theoretical construct. Analysis level: organisation or wider socio-technical system. Temporal character: longitudinal and path-dependent. First FlowSentience development: emerged during the 2026 research process leading to the present paper.

Sensemaking Debt is a proposed latent and path-dependent organisational condition produced when consequential interpretive obligations are repeatedly carried forward without sufficient renewal and begin constraining future interpretation, coordination and adaptation.

Neighbouring concepts include technical debt, cognitive debt, intent debt, knowledge debt, organisational memory loss, common-ground breakdown, resilience erosion and institutional complexity. The claimed contribution is the temporal Gap-Debt distinction, organisational analysis level, accumulation-recovery dynamic and integration with organisational continuity. Exclusive ownership of the phrase or discovery of all underlying phenomena is not claimed.

In practical terms, the question is whether unfinished work of understanding has become an inherited restriction. Latent means inferred from evidence rather than directly visible. Path-dependent means that earlier responses affect what can be understood and changed now.

Sensemaking Gap

Status: provisional analytical construct. Analysis level: a defined situation within a relevant sensemaking perimeter. Temporal character: momentary.

A Sensemaking Gap is a context-specific mismatch, at a particular point in time, between the form and degree of interpretive connectivity required for consequentially interdependent action and what the relevant adaptive system can currently sustain.

The gap can involve insufficient connection, concealed divergence or premature convergence.

A gap is a present mismatch, such as two services using 'ready' differently when a safe handover depends on recognising the difference. Its existence does not establish debt.

Sufficiently Shared Understanding

Sufficiently shared understanding is the context-appropriate degree and form of interpretive connectivity required for consequentially interdependent action to remain coherent, contestable and adaptive. Status: authors' synthesis grounded in established research.

It is not equivalent to consensus or identical mental models.

Interpretive connectivity means that relevant differences in meaning can be recognised and addressed. People can have sufficiently shared understanding while disagreeing, provided that their disagreement remains intelligible and available to inform action.

Principle of Adaptive Balance

For an adaptive system, keeping the renewal of context-appropriate interpretive connectivity and continuity sufficiently responsive to emerging obligations is proposed as one enabling condition of continued viability. Status: authors' theoretical synthesis.

This formulation identifies a proposed enabling condition. It does not establish a sufficient condition for survival, a universal law, or a numerical equilibrium. Material resources, ecological conditions, power, legitimacy and external shocks can determine outcomes even when interpretive renewal is strong. Renewal itself includes the formation of new capability as well as the repair of inherited discontinuity.

Interpretive Demand and Renewal Capacity

Interpretive demand is the differentiation, connection and revision of meanings required for coordinated action. Renewal capacity is the ability to reconstruct sufficiently shared understanding as circumstances change. These are analytical descriptions of a relationship, with no validated common units. Demand concerns the work required; capacity concerns the means available to undertake it.

Interpretive Obligations Carry Forward and Recovery

An interpretive obligation is work needed to establish or renew a consequential meaning, dependency, evidential basis or purpose. Carry-forward occurs when later activity inherits the unresolved obligation. Compensation keeps current activity workable despite it. Recovery changes the conditions that make it persist, including retiring an obsolete obligation. These terms describe candidate processes; they are not separately validated variables.

Effective Sensemaking Scale

The number, heterogeneity and consequential interdependence of interpretations that must remain sufficiently connected for coordinated action. Status: provisional analytical concept.

No universal numerical threshold is claimed.

Scale concerns how many different perspectives need to connect and how much their differences matter to one another's actions. A small cross-professional team can therefore face a demanding sensemaking task.

Sensemaking Perimeter

The consequence-dependent boundary across which interpretations materially affect the adaptive outcome being coordinated. Status: provisional operational label grounded in established boundary inquiry.

The report does not claim discovery of the general problem of system boundaries.

The perimeter asks whose knowledge and consequences must be considered for a specified outcome. It can cross organisational boundaries and must be justified. It is not a requirement that every participant share one comprehensive model.

State, Epistemic and Intent Continuity

Status: provisional analytical distinction developed in this paper. State continuity is the capacity to reconstruct how the present configuration, dependencies, roles and constraints emerged. Epistemic continuity is the capacity to reconstruct what was known, assumed, uncertain or contested when earlier decisions were formed. Intent continuity is the capacity to reconstruct the purpose or protected value an inherited arrangement was created to serve. The three are not claimed as established independent variables or an exhaustive model of organisational continuity.

Constraint Accretion

Constraint accretion is the accumulation of organisational arrangements and constraints that remain consequential while becoming increasingly difficult to interpret, revise or remove. Status: candidate carrier mechanism.

It is not claimed as a complete explanation of Sensemaking Debt.

A rule may become harder to change as it is reused in contracts and software whose relationships are poorly understood. That build-up illustrates the proposed carrier mechanism. The mere presence of many rules does not establish it.

Reorganisation as Repayment or Refinancing

Status: analytical application of the theory. Repayment changes the conditions that make obligations persist and restores continuity. Refinancing changes visible structure while carrying underlying obligations into the new arrangement.

Decision Acceleration and Sensemaking Acceleration

Status: analytical AI distinction. Decision acceleration increases the speed at which representations become actions. Sensemaking acceleration increases the speed at which actors can recognise, connect and revise consequential interpretations. These are not claimed as mutually exclusive categories of AI system.

Recursive Self Improvement

An established AI research idea in which improvements also strengthen the capacity to produce further improvements. This report distinguishes bounded changes to an agent from changes to underlying models or evaluators. Its proposed organisational extension concerns whether successive improvements preserve the ability to reconstruct and correct consequential changes. Neither the general idea nor its technical demonstration is claimed as a FlowSentience contribution.

Provenance Evaluation and Oversight

Provenance traces a claim or arrangement to its sources and changes. An evaluator is a process used to judge a proposed result or modification. Human-on-the-loop refers here to oversight of an acting process, with the practical capacity to intervene. These supporting terms are established or ordinary descriptions. An accessible log, a high score or a named supervisor does not alone establish collective understanding.

Intertemporal Consequence Valuation and Decision Conversion

Status: analytical extension developed in this paper. Decision conversion describes the transitions through which a consequential signal becomes visible, understood, valued, authorised, resourced and implemented. Intertemporal consequence valuation compares expected outcomes across alternative timings and probabilities of action within a defined sensemaking perimeter. Neither construct measures Sensemaking Debt; both examine how a latent organisational condition may affect action timing and divergence between local and system-wide value.

Counterfactual means an explicitly constructed alternative path. Expected net present value compares uncertain future benefits, costs and losses on a common present-value basis. Discounting specifies how values at different times are compared; distributional weights specify how affected groups enter the calculation. These decision-analysis terms expose assumptions rather than determine them. A locally negative value can coexist with a positive wider value when benefits accrue elsewhere.

Developmental Understanding Debt

Developmental understanding debt is a provisional extension proposed for investigation in this revision. It refers to a path-dependent burden that may arise when consequential learning opportunities are repeatedly displaced or left unfulfilled and later activity presupposes the capabilities those opportunities were expected to form. It is not a validated learner attribute, a medical diagnosis, a PISA-derived index, or an interchangeable name for organisational Sensemaking Debt. Institutional provision and responsibility remain part of its analysis.

Evidence and Measurement

A proxy is an observable pattern used as indirect evidence, such as repeated requests to reconstruct a rationale. Operationalisation specifies how a concept will be examined in a bounded study. Longitudinal evidence follows events through time; a cross-sectional observation describes a snapshot. A heuristic is a device for organising reasoning. None of these descriptions makes an indicator diagnostic or a model empirically validated.

Appendix C: Evidence Status of the Research Update

The wider update is a focused synthesis through 9 September 2026. Version 1.3 adds a targeted check of recursive self-improvement sources on 11 September 2026; it does not extend every earlier search to that date. Theoretical mechanisms, illustrative cases and valuation assumptions are the authors' synthesis. The table records the different evidential roles of the sources.

The recursive self-improvement extension uses the DGM paper's methods and supplementary detector example, Google DeepMind's primary account of AlphaEvolve, and the abstract of Chen and colleagues' September survey revision. The first two describe bounded technical work; the survey supplies conceptual distinctions. Their organisational, valuation and educational implications in this report are proposed inferences. No source is treated as an empirical test of Sensemaking Debt.

Primary-source details could not be checked for the newly released PISA 2025 tables. The discussion therefore identifies its limited reliance on initial reporting and does not reproduce unverified new estimates. The complete report should be consulted before using country results or associations for policy appraisal. Preprints are identified as such; neither their availability nor conceptual proximity establishes independent validation of Sensemaking Debt.

References

Abrahamson, E. (1996). Management fashion. Academy of Management Review, 21(1), 254-285. https://doi.org/10.5465/AMR.1996.9602161572

Alvesson, M. (1993). Organizations as rhetoric: Knowledge-intensive firms and the struggle with ambiguity. Journal of Management Studies, 30(6), 997-1015. https://doi.org/10.1111/j.1467-6486.1993.tb00476.x

Alvesson, M. (2001). Knowledge work: Ambiguity, image and identity. Human Relations, 54(7), 863-886. https://doi.org/10.1177/0018726701547004

Alvesson, M., & Spicer, A. (2012). A stupidity-based theory of organizations. Journal of Management Studies, 49(7), 1194-1220. https://doi.org/10.1111/j.1467-6486.2012.01072.x

Alvesson, M., & Willmott, H. (2002). Identity regulation as organizational control: Producing the appropriate individual. Journal of Management Studies, 39(5), 619-644. https://doi.org/10.1111/1467-6486.00305

Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., Suh, J., Iqbal, S., Bennett, P. N., Inkpen, K., Teevan, J., Kikin-Gil, R., & Horvitz, E. (2019). Guidelines for human-AI interaction. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (pp. 1-13). Association for Computing Machinery. https://doi.org/10.1145/3290605.3300233

Andronis, L., & Barton, P. M. (2016). Adjusting estimates of the expected value of information for implementation: Theoretical framework and practical application. Medical Decision Making, 36(3), 296-307. https://doi.org/10.1177/0272989X15614814

Argyris, C., & Schön, D. A. (1978). Organizational learning: A theory of action perspective. Addison-Wesley.

Arrow, K. J., & Fisher, A. C. (1974). Environmental preservation, uncertainty, and irreversibility. Quarterly Journal of Economics, 88(2), 312-319. https://doi.org/10.2307/1883074

Arrow, K. J., Cropper, M., Gollier, C., Groom, B., Heal, G., Newell, R. G., Nordhaus, W. D., Pindyck, R., Pizer, W. A., Portney, P. R., Sterner, T., Tol, R. S. J., & Weitzman, M. L. (2013). Determining benefits and costs for future generations. Science, 341(6144), 349-350. https://doi.org/10.1126/science.1235665

Ashby, W. R. (1956). An introduction to cybernetics. Chapman & Hall.

Associated Press. (2026, September 9). Families’ interest in their kids’ education is falling around the globe as scores drop, test shows [Initial news reporting]. https://apnews.com/article/2c1c7b338e3d5b38031ef0eecef9ce5d

Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), 775-779. https://doi.org/10.1016/0005-1098(83)90046-8

Beer, S. (1972). Brain of the firm: A development in management cybernetics. Herder and Herder.

Beer, S. (1979). The heart of enterprise. Wiley.

Carlile, P. R. (2004). Transferring, translating, and transforming: An integrative framework for managing knowledge across boundaries. Organization Science, 15(5), 555-568. https://doi.org/10.1287/orsc.1040.0094

Chen, M., Wang, L., & Qu, B. (2026). Recursive self-improvement in AI: From bounded self-refinement to autonomous research loops (Version 2) [Preprint; abstract consulted]. arXiv. https://doi.org/10.48550/arXiv.2607.07663

Chowdhury, M. Z. U. S., & Khan, S. R. (2026, August 23). Evaluation in the age of AI: Output as evidence of learning [Preprint; abstract consulted]. arXiv. https://arxiv.org/abs/2608.22660

Comes, T. (2026). Sensemaking AI: Introducing a research and design agenda for human-AI networks. EPJ Data Science, 15, Article 42. https://doi.org/10.1140/epjds/s13688-026-00634-5

Conant, R. C., & Ashby, W. R. (1970). Every good regulator of a system must be a model of that system. International Journal of Systems Science, 1(2), 89-97. https://doi.org/10.1080/00207727008920220

Cunningham, W. (1992). The WyCash portfolio management system. OOPS Messenger, 4(2), 29-30. https://doi.org/10.1145/157710.157715

DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48(2), 147-160. https://doi.org/10.2307/2095101

Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. https://doi.org/10.1126/sciadv.adn5290

Dunbar, R. I. M. (1992). Neocortex size as a constraint on group size in primates. Journal of Human Evolution, 22(6), 469-493. https://doi.org/10.1016/0047-2484(92)90081-J

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

Eppler, M. J., & Mengis, J. (2004). The concept of information overload: A review of literature from organization science, accounting, marketing, MIS, and related disciplines. The Information Society, 20(5), 325-344. https://doi.org/10.1080/01972240490507974

European Commission. (2022). Opportunities and challenges of artificial intelligence technologies for the cultural and creative sectors. Publications Office of the European Union. https://doi.org/10.2759/144212

Faraj, S., Pachidi, S., & Sayegh, K. (2018). Working and organizing in the age of the learning algorithm. Information and Organization, 28(1), 62-70. https://doi.org/10.1016/j.infoandorg.2018.02.005

Galbraith, J. R. (1974). Organization design: An information processing view. Interfaces, 4(3), 28-36. https://doi.org/10.1287/inte.4.3.28

Google DeepMind. (2025, May 14). AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms [Research announcement]. https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/

Greve, H. R. (1998). Performance, aspirations, and risky organizational change. Administrative Science Quarterly, 43(1), 58-86. https://doi.org/10.2307/2393591

Hallegatte, S. (2012). A cost effective solution to reduce disaster losses in developing countries: Hydro-meteorological services, early warning, and evacuation (Policy Research Working Paper 6058). World Bank. https://documents.worldbank.org/en/publication/documents-reports/documentdetail/190261468181486694

Hannan, M. T., & Freeman, J. (1984). Structural inertia and organizational change. American Sociological Review, 49(2), 149-164. https://doi.org/10.2307/2095567

Hollnagel, E., Woods, D. D., & Leveson, N. (Eds.). (2006). Resilience engineering: Concepts and precepts. Ashgate. https://doi.org/10.1201/9781315605685

IPCC. (2022a). Climate change 2022: Impacts, adaptation and vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press. https://doi.org/10.1017/9781009325844

IPCC. (2022b). Climate change 2022: Mitigation of climate change. Contribution of Working Group III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press. https://doi.org/10.1017/9781009157926

Klein, G., Feltovich, P. J., Bradshaw, J. M., & Woods, D. D. (2005). Common ground and coordination in joint activity. In W. B. Rouse & K. R. Boff (Eds.), Organizational simulation (pp. 139-184). Wiley. https://doi.org/10.1002/0471739448.ch6

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task [Preprint, version 2, revised December 31]. arXiv. https://arxiv.org/abs/2506.08872v2

Kurtz, C. F., & Snowden, D. J. (2003). The new dynamics of strategy: Sense-making in a complex and complicated world. IBM Systems Journal, 42(3), 462-483. https://doi.org/10.1147/sj.423.0462

Lenton, T. M., Rockström, J., Gaffney, O., Rahmstorf, S., Richardson, K., Steffen, W., & Schellnhuber, H. J. (2019). Climate tipping points - too risky to bet against. Nature, 575, 592-595. https://doi.org/10.1038/d41586-019-03595-0

Lindenfors, P., Wartel, A., & Lind, J. (2021). 'Dunbar's number' deconstructed. Biology Letters, 17(5), Article 20210158. https://doi.org/10.1098/rsbl.2021.0158

March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2(1), 71-87. https://doi.org/10.1287/orsc.2.1.71

McGuigan, J. (2005). The cultural public sphere. European Journal of Cultural Studies, 8(4), 427-443. https://doi.org/10.1177/1367549405057827

Meyer, J. W., & Rowan, B. (1977). Institutionalized organizations: Formal structure as myth and ceremony. American Journal of Sociology, 83(2), 340-363. https://doi.org/10.1086/226550

National Academies of Sciences, Engineering, and Medicine. (2017). Valuing climate damages: Updating estimation of the social cost of carbon dioxide. National Academies Press. https://doi.org/10.17226/24651

Nieborg, D. B., & Poell, T. (2018). The platformization of cultural production: Theorizing the contingent cultural commodity. New Media & Society, 20(11), 4275-4292. https://doi.org/10.1177/1461444818769694

Nonaka, I., & Takeuchi, H. (1995). The knowledge-creating company: How Japanese companies create the dynamics of innovation. Oxford University Press.

Ocasio, W. (1997). Towards an attention-based view of the firm. Strategic Management Journal, 18(S1), 187-206. https://doi.org/10.1002/(SICI)1097-0266(199707)18:1+<187::AID-SMJ936>3.0.CO;2-K

OECD. (2023). PISA 2022 results (Volume I): The state of learning and equity in education. OECD Publishing. https://doi.org/10.1787/53f23881-en

OECD. (2026a). OECD Digital Education Outlook 2026: Exploring effective uses of generative AI in education. OECD Publishing. https://doi.org/10.1787/062a7394-en

OECD. (2026b). Programme for International Student Assessment (PISA) [Programme page and publication listing; accessed September 9, 2026]. https://www.oecd.org/en/about/programmes/pisa.html

Ostrom, E. (2010). Polycentric systems for coping with collective action and global environmental change. Global Environmental Change, 20(4), 550-557. https://doi.org/10.1016/j.gloenvcha.2010.07.004

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

Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation-augmentation paradox. Academy of Management Review, 46(1), 192-210. https://doi.org/10.5465/amr.2018.0072

Rasmussen, J. (1997). Risk management in a dynamic society: A modelling problem. Safety Science, 27(2-3), 183-213. https://doi.org/10.1016/S0925-7535(97)00052-0

Ratanjee, V. (2026, March 17). Sensemaking debt: When AI answers every question and leaders stop asking. Forbes. https://www.forbes.com/sites/vibhasratanjee/2026/03/17/sensemaking-debt-when-ai-answers-every-question-and-leaders-stop-asking/

Roediger, H. L., III, & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255. https://doi.org/10.1111/j.1467-9280.2006.01693.x

Roetzel, P. G. (2019). Information overload in the information age: A review of the literature from business administration, business psychology, and related disciplines with a bibliometric approach and framework development. Business Research, 12(2), 479-522. https://doi.org/10.1007/s40685-018-0069-z

Seeber, I., Bittner, E., Briggs, R. O., de Vreede, T., de Vreede, G.-J., Elkins, A., Maier, R., Merz, A. B., Oeste-Reiß, S., Randrup, N., Schwabe, G., & Söllner, M. (2020). Machines as teammates: A research agenda on AI in team collaboration. Information & Management, 57(2), Article 103174. https://doi.org/10.1016/j.im.2019.103174

Senge, P. M. (1990). The fifth discipline: The art and practice of the learning organization. Doubleday/Currency.

Simon, H. A. (1957). Models of man: Social and rational. John Wiley & Sons.

Snowden, D. J. (2011). Naturalizing sensemaking. In K. L. Mosier & U. M. Fischer (Eds.), Informed by knowledge: Expert performance in complex situations (pp. 223-234). Psychology Press.

Snowden, D. J., & Boone, M. E. (2007). A leader's framework for decision making. Harvard Business Review, 85(11), 68-76.

Storey, M.-A. (2026). From technical debt to cognitive and intent debt: Rethinking software health in the age of AI [Preprint, version 4]. arXiv. https://arxiv.org/abs/2603.22106v4

Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257-285. https://doi.org/10.1207/s15516709cog1202_4

Teece, D. J., Pisano, G., & Shuen, A. (1997). Dynamic capabilities and strategic management. Strategic Management Journal, 18(7), 509-533. https://doi.org/10.1002/(SICI)1097-0266(199708)18:7<509::AID-SMJ882>3.0.CO;2-Z

The Verge. (2026, September 9). Students who use AI generally score worse at school [Initial news reporting]. https://www.theverge.com/ai-artificial-intelligence/991956/student-ai-use-scores-oecd-pisa

Tushman, M. L., & Nadler, D. A. (1978). Information processing as an integrating concept in organizational design. Academy of Management Review, 3(3), 613-624. https://doi.org/10.5465/AMR.1978.4305791

Ulrich, W. (1983). Critical heuristics of social planning: A new approach to practical philosophy. Haupt.

Weick, K. E. (1995). Sensemaking in organizations. Sage.

Weick, K. E., & Sutcliffe, K. M. (2007). Managing the unexpected: Resilient performance in an age of uncertainty (2nd ed.). Jossey-Bass.

Weiner, B. J. (2009). A theory of organizational readiness for change. Implementation Science, 4, Article 67. https://doi.org/10.1186/1748-5908-4-67

Windahl, J., & Johansson, D. (2025). Sentienta värdeströmmar: Lyhörda flöden för vård och omsorg [White paper]. Zenodo. https://doi.org/10.5281/zenodo.17142073

Windahl, J., & Johansson, D. (2026a). The Sentience Threshold: How civilisations learn to see their own consequences. BoD - Books on Demand.

Windahl, J., & Johansson, D. (2026b). The Sentient Value Stream Framework: How organisations learn to sense themselves. BoD - Books on Demand.

Woods, D. D. (2015). Four concepts for resilience and the implications for the future of resilience engineering. Reliability Engineering & System Safety, 141, 5-9. https://doi.org/10.1016/j.ress.2015.03.018

Woods, D. D. (2018). The theory of graceful extensibility: Basic rules that govern adaptive systems. Environment Systems and Decisions, 38(4), 433-457. https://doi.org/10.1007/s10669-018-9708-3

Zhang, J., Hu, S., Lu, C., Lange, R., & Clune, J. (2026). Darwin Gödel Machine: Open-ended evolution of self-improving agents (Version 3; first posted 2025) [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2505.22954

Zhu, L., & Cai, M. (2026, September 4). Artificial intelligence in equity and crypto markets: Progress, profitability evidence, and the limits of automated investing [Preprint]. arXiv. https://arxiv.org/abs/2609.04917