Current organizations operate in contexts where performance is no longer measured solely by the amount of resources implemented. It also depends on the quality of interactions, the reliability of commitments, cooperation, adaptability and risk management. Classical economic models are well suited to representing resource allocation and constrained optimization, but they are less able to account for how relational and behavioral variables actually influence organizational processes.

It is in this articulation zone that the CS-LR model was built. It is based on four pillars: Confidence (C), Satisfaction (S), Loyalty (L) and Risks or Hazards (R) ([5]; [6]). Already, the initial model is no longer limited to the linear relationship C → S → L: it proposes a dynamic loop in which experiments, relationships and environmental disturbances can successively modify behavioral states.

However, there remains a methodological difficulty: a variable such as trust can be declared by an actor without this declaration corresponding perfectly to his practices. An actor can declare himself confident while strengthening his controls, declare himself satisfied while multiplying complaints, or declare himself loyal while effectively reducing his cooperation. Conversely, some observable behaviors may reflect a relational stability that questionnaires imperfectly capture.

This is where the major CS-LR mutation comes in. Observation is no longer an adjunct to further research. It becomes a second knowledge register integrated into the construction, measurement and interpretation of the variables of the model. This development is consistent with the orientation already present in the first version, which identified the declarative-observational combination as a means of overcoming the bias of the declarative alone.

In the previous version, the observation was expressed as a methodological perspective. In the new version, it becomes a constituent dimension of the model: the CS-LR is now designed to link declared states, observed behaviors, organizational events and performance.

This transformation leads to a reformulation of the scientific question. It is no longer just a question of whether C, S and L influence R or efficiency. We must also ask ourselves how these constructs translate into behaviors and how the confrontation between perception and observation makes it possible to improve their measurement.

So the crux of the problem is this: by integrating both reported perceptions and observable manifestations of trust, satisfaction, loyalty, and risk, how does such modeling enable a more rigorous representation of the behavioral dynamics of organizational efficiency?

The article has four objectives: 1) to reconstruct the theoretical and empirical path of the model; 2) to explain its observational mutation; 3) to propose a measurement architecture integrating two data sources; 4) to show how this architecture is articulated with microeconomic optimization and systemic modeling.

The model can be represented by a main loop:

Perceptions (C, S, L) → relational behaviors → coordination → performance/risk events → new perceptions (C, S, L).

R acts both as a disruptor and as a variable that can be influenced by organizational behaviors. The loop becomes:

C_t + S_t + L_t → behaviors_t → θ_t → performance_t; R_t → performance_t and C_{t+1}, S_{t+1}, L_{t+1}.

This representation makes it possible to distinguish the latent states from the observable manifestations. Variables C, S and L are not directly visible as such; they are inferred from declarative and behavioral indicators. R may, depending on its nature, be directly observed by events or approached by perceptions of probability and severity.

Table 1 becomes the methodological core of the CS-LR: a variable is no longer considered sufficiently documented by a single statement when a behavioral or organizational manifestation can be observed ([5]).

Data convergence and divergence

Triangulation should not be seen as an obligation of perfect convergence. Three configurations are possible. Firstly, convergence: the actor declares himself confident and the behavior observed actually shows stable cooperation and limited control. Secondly, the divergence: the actor declares a strong confidence but the practices reveal a multiplication of controls. Thirdly, the lack of systematic correspondence: the two sources can measure different dimensions of the same construct.

The model must therefore avoid mechanically merging the two sources into a single score. A more rigorous strategy is first to keep the perceptual and behavioral indicators separately and then to study their relationship. Aggregation should only take place after verification of their theoretical and empirical consistency.

Table 1. Multimodal operationalization.

To operationalize the mathematical formulations established in the previous section, this chapter constructs a rigorous empirical framework designed to test the model’s validity. This process is divided into two operational stages. First, Section 6.1 delineates the core observational CS-LR assumptions required to ground the analysis. Second, Section 6.2 expounds upon the formal validation methodology used to evaluate the model’s empirical robustness.

The evaluation of the empirical relevance of the model is based on two distinct fields. The methodology is based on a double evaluation loop for each variable: a transactional/declarative loop and a behavioral/observational loop. The triangulation of these two loops makes it possible to scientifically measure the gap between the perceived and the organizational experience.

1) Management of a service function in engineering school (Axis 1)

This first field examines a higher education and research structure understood as a complex decision-making system where multiple categories of actors coexist (students, teachers, researchers, management, administrators, and partners).

Data collection applies double loop and triangulation as follows:

The Satisfaction Loop: The declarative loop captures perceived satisfaction via conventional satisfaction surveys. The observational loop traces feedback behaviors (recurrence of complaints, abandonment rates, and requests for correction). Triangulation identifies the profiles of actors who are “silent but dissatisfied” or “chronically dissatisfied despite a compliant service”.The Confidence Loop: The declarative loop measures interpersonal and institutional confidence through attitude questionnaires. The observational loop documents the actual use of the delegations of power, the respect of commitments (deadlines for the rendering of notes, processing of files) and the sharing of information. The triangulation objective the level of real confidence by measuring whether the acts of management corroborate the discourses.The Risk Loop: The declarative loop evaluates the risk aversion of school executives. The observational loop counts actual incidents, process disruptions and the activation of post-clearance checks.

2) Functional inter-entity delegation (Axis 2)

The second area concerns an integrated industrial system composed of five autonomous entities linked by delegation mechanisms. To circumvent the small size of the basket of entities ($N = 5$), a longitudinal and repeated observation device is deployed, breaking down the follow-up into decision and transactional episodes according to a certain periodicity (for example: 24 months).

The double loop and triangulation are modeled as follows:

The inter-organizational satisfaction loop: It compares the satisfaction declared during the annual steering committees (declarative loop) with the real fluidity of industrial flows and operational continuity (observational loop). Triangulation can detect latent crises masked by managerial diplomacy.The Confidence and Control Loop: The reporting loop captures the confidence shown by the management of the parent company towards its subsidiaries. The observational loop quantifies the actual level of supervision (frequency of audits, volume of reporting procedures imposed). Triangulation highlights situations of “false trust” (benevolent discourse but suffocating micro-management) or “forced trust” for lack of means of control.The Risk and Take Back Loop: It makes managers’ perception of industrial risks (declarative) dialog with indicators of non-compliance, delivery delays and direct managerial take back decisions (observational).

Table 2. Triangulation et double boucle.

The exploitation of this Table 2 requires a rigorous analysis of the methods of statistical treatment of the deviations and triangulation resulting from the double loop.

The introduction of a double loop (declarative and observational) involves the statistical processing of heterogeneous data. The challenge is not only to validate each loop independently, but to scientifically analyze the divergence (the gap) between the declared perception and the observed behavior. Two complementary approaches are proposed to operationalize this treatment.

1) Residue Analysis and Deviation Scores (Gap Analysis)

To measure the alignment or decoupling within the same dimension, between the declared confidence \(\text{Trust}_{\text{decl},it}\) and the observed control \(\text{Control}_{\text{obs},it, the first step is to standardize the variables on a common scale and then model their deviation.

Two techniques are favored:

Calculation of direct deviation score: Determined by \(\Delta_{it} = X_{\text{obs}, it} - X_{\text{decl}, it}\) significantly away from zero reveals an organizational anomaly. For example, a strongly negative score in confidence (low real supervision despite a maximum declared confidence) objective a phenomenon of “false confidence” or managerial detachment.Regression Residue Analysis: By regressing the observational variable on the declarative variable (\(X_{\text{obs}, i} = \beta_0 + \beta_1 X_{\text{decl}, i} + \epsilon_i\)), the \(\epsilon _{i}\), the $\epsilon$ residue captures the part of the behavior that contradicts or escapes the declared perception. These residues are then reintroduced as explanatory variables to analyze the incidents or the breakdown of the cooperation.

2) Structural Equation Modeling (SEM) with Latent Variables

In order to simultaneously address the complexity of Axis 1 (multiple actors) and the longitudinal nature of Axis 2 (repeated data), the methodology relies on structural equation models (SEM), combining factor analysis and multiple regressions (Figure 1).

Figure 1. Structural equation models (SEM).

Measurement model (Factor analysis): Each construct (Satisfaction, Confidence, Risk) is treated as a latent variable not directly observable. This variable is fed simultaneously by reflective indicators (responses to questionnaires in the declarative loop) and formative indicators (performance metrics and behavioral flows in the observational loop).Structural model (Causal relationships): Structural equations allow to test whether the impact of satisfaction or confidence on performance (or on the continuity of cooperation) is mediated by the accuracy of the alignment between the declarative and the observational. Specificity of Axis 2: Fixed-effect panel models.

For the industrial system where $N = 5$, the analysis of the classical structural equations is adapted in the form of a panel model with fixed effects. Splitting data into $T$ periods (months, quarters) or $E$ transactional episodes provides a sufficient sample ($N \times T$). The fixed “entity” effects make it possible to control the unobserved invariant heterogeneity (such as the culture specific to each of the 5 subsidiaries), concentrating the statistical estimation on the temporal dynamics of the interactions and on the evolution of the trade-offs between trust and real supervision.

The standardized and concise econometric formulation for the estimation of Axis 2 (Fixed Effects Panel) is as follows:

Deviation Model Specification (Axis 2)

To analyze the impact of the decoupling between discourse and facts within the $N = 5$ industrial entities observed over $T$ periods, the econometric model in panel data is written:

\(Y_{it}=\beta_{1}\Delta \text{Trust}_{it}+\mathbfit{\gamma }^{\prime }\mathbf{X}_{it}+\alpha _{i}+\epsilon _{it}\)

where:

$Y_{it}$ (or \(Y_{it}\):): The operating performance (or incident rate) of the entity $i$ at the period $t$ (or \(i\) at the period \(t\)).\(\Delta \text{Trust}_{it}\): The standardized deviation score, calculated by the difference between the declared managerial trust and the actual supervision level (\(\text{Control}_{\text{obs}, it} - \text{Trust}_{\text{decl}, it}\)).$X_{it}$: The vector of control variables (transaction size, industry, production load).$\alpha_i$: The individual fixed effect specific to each $i$ entity, capturing invariant heterogeneity (corporate culture, local industrial specificities).$\epsilon_{it}$: The residual error term.

Fixed-effect estimation neutralizes $\alpha_i$, ensuring that $\beta$ coefficients measure only the effect of temporal variations in behavioral deviations on the continuity of cooperation.

From these applications proposed below, it emerges that the empirical refoundation of the research axes, through the prism of the double loop and the methodological triangulation, endows the model with increased scientific rigor. By systematically comparing perceptual variables (declarative loop) with operational realities (observational loop), this device removes the traditional bias of self-declared data.

Whether it is to map the logics of actors within a complex decision-making system (Axis 1) or to model the confidence-control dynamics by a longitudinal approach in panel data (Axis 2), this architecture makes it possible to isolate and quantify scientifically the organizational cost of managerial distortions. The model is no longer limited to measuring isolated psychological states or industrial outcomes; it precisely objectifies the impact of behavioral alignment on the performance and sustainability of cooperative systems.

The first version of the model already proposed a mapping linking standard theory, extended standard theory, non-standard theory and experimental economics.

The present version adds a transversal dimension: observation becomes the methodological principle allowing perception, behavior and performance to dialog (see Table 3).

Table 3. Epistemological mapping.

The evaluation of the scientific contributions of the CS-LR model requires to put into perspective its theoretical contributions with its intrinsic limits, thus introducing a critical discussion on the real scope of the observational shift and on the methodological precautions required for its empirical deployment.

1) Discussion: what observational change really changes?

The CS-LR mutation should not be reduced to the addition of a collection technique. It modifies the very conception of the behavioral construct. In a strictly declarative model, behavioral information is mediated by the perception of the actor. In the observational CS-LR, the model distinguishes between the perceived state, the behavioral manifestation and the organizational consequence.

This distinction partly solves a classic social-science problem: the supposed equivalence between what an individual says and what he does. The model does not claim that this equivalence is false in all cases; it makes it an empirical question.

The second consequence concerns causation. A C-S-L-R loop cannot be understood solely as a succession of responses to questionnaires. Observable events and behaviors provide temporal anchors. They examine whether a change in trust precedes a change in behavior or whether, on the contrary, the behavior produces the change in trust.

The third consequence concerns governance. Control can now be observed as organizational behavior. If an organization declares a strong trust but multiplies the controls, the contradiction becomes a scientific object. It may signal latent risk, low effective trust or an institutional policy of independent trust control.

The fourth consequence is efficiency. The θ multiplier ceases to be a simple behavioral abstraction: it can be linked to observable process indicators. This relationship remains to be estimated, but it becomes methodologically more defensible.

2) Scientific limitations and precautions

The integration of observation does not eliminate bias. The observer can influence the situation; some practices may be invisible; behaviors may vary according to the context; organizational data may be incomplete. The observation must therefore be protocolized.

Another limitation is that the observed behavior is not always a pure indicator of the latent construct. Control can result from mandatory regulation rather than from a lack of trust. A claim may be related to a one-time anomaly rather than structural dissatisfaction. The model must therefore incorporate context variables.

Finally, triangulation must not become rhetoric. To be scientifically useful, each source must be defined, coded and linked to a specific analytical question. The mere coexistence of questionnaires and observations does not in itself constitute validation.

3) Scientific contribution of CS-LR

The model’s contribution can be summarized in five levels.

A conceptual contribution: extension of the CS-L kernel by explicit integration of risks and hazards.

A dynamic contribution: transition from a linear relation to a retroactive loop.

A methodological contribution: transition from declarative alone to a declarative-observational architecture.

An economic contribution: integration of behavioral variables into an increased production function and into the primal and dual programs.

A systemic contribution: articulation of perceptions, behaviors, events, coordination, risks and performance.

The original character of this version therefore lies less in the addition of a collection instrument than in the transformation of the observation logic into a component of the model itself.

The CS-LR model has reached an important conceptual milestone. It can no longer be properly described as a mere declarative extension of the triptych Trust-Satisfaction-Loyalty. Its evolution leads to an architecture in which declared perceptions and observable manifestations are two complementary paths of access to behavioral dynamics.

This mutation responds directly to a fundamental difficulty of behavioral models: a declared perception is not necessarily equivalent to an observed behavior. CS-LR does not seek to choose between the two. He connects them.

The model retains the achievements of its previous trajectory. Empirical work on the Senegalese insurance market provided a foundation for the C-S-L core; the Extended CS-L proposal introduced dynamics and contingencies; the CS-LR now extends this architecture to organizational functioning and observational measurement. Elements related to this trajectory are documented in the materials of this revision.

From a microeconomic point of view, the function Q = θ·f(T, K), the primal program and the dual program remain formal translation instruments. Behaviorally, C, S, L and R become constructs whose measurement can combine perceptions and manifestations. At the systemic level, the time loop makes it possible to study the reciprocal effects between relationship, behavior, risk and performance.

The major scientific consequence is that the CS-LR should no longer be presented as a model that « proves » a priori that trust, satisfaction or loyalty maximize efficiency. It must be presented as a framework that makes these relationships observable, measurable and testable. Mathematical formalization establishes the logical consequences of certain assumptions; declarative and observational data must empirically establish whether these assumptions hold.

Thus conceived, the CS-LR becomes a true integrative research framework [it links perception and behavior, individual and organization, relationship and risk, microeconomics and management, optimization and observation. Its strength lies not in its claim to universality, but in its ability to organize multiple registers of evidence around the same systemic dynamic.]

The main scientific implication is that the CS-LR should no longer be offered as a model that “demonstrates” a priori that trust, satisfaction or loyalty are the factors that most enhance efficiency. It should be presented as a framework that allows these relationships to be observed, measured, and tested. Our results support the work of [10] by demonstrating that the combination of the variables CSLR prevents the failure of unilateral governance by promoting trust and reciprocity among stakeholders.