Essence: Core Constructs Across Disciplines

• Essence is a multifaceted concept that denotes the invariant, fundamental core of a system, seen in software kernels, constraint models, and cosmological field theories.
• In software engineering, the Essence Theory defines core alphas and practices, enabling method-agnostic and adaptive approaches in startups and mature organizations.
• Across domains like machine learning, logic, and cosmology, variations of Essence capture the central semantic or kinetic structure, guiding model innovation and analysis.

Essence, in the cited research literatures, does not denote a single doctrine or artifact. It names several technically distinct constructs: a method-agnostic framework for software engineering methods; a high-level constraint modelling language and its tooling ecosystem; a family of logical and metaphysical distinctions concerning essence, accident, and existence; a label for selective or semantic transfer in machine learning and program analysis; and, in cosmology, a class of non-canonical field models such as $k$-essence, $f$-essence, and $g$-essence. This suggests a recurring concern with what is treated as fundamental, invariant, or semantically central when peripheral structure varies.

1. Terminological range

In the cited work, “Essence” appears in at least five major research contexts.

Domain	Use of “Essence”	Representative papers
Software engineering	Kernel-and-language framework for methods and practices	(Kemell et al., 2021, Evensen et al., 2018)
Constraint programming	High-level modelling language above solver-level decisions	(Nightingale et al., 2016, Akgün et al., 2021)
Logic and metaphysics	Essence/accident and essence/existence distinctions	(Fan, 2015, Al-Fedaghi, 2024)
Machine learning and code analysis	High-level semantic residue or core logic	(Yang et al., 2019, Chefer et al., 2021, Zhao et al., 26 Feb 2025)
Cosmology	Non-canonical field models such as $k$-essence	(Hussain et al., 2024, Myrzakulov, 2010)

The term therefore functions less as a unified theory label than as a cross-domain marker for what a field takes to be indispensable structure. In some cases this indispensable structure is formalized as a kernel; in others it is a sparse set of outputs, a high-level semantic direction, a weighted core of program logic, or a distinction between what something is and that it exists.

2. Essence Theory of Software Engineering

In software engineering, the Essence Theory of Software Engineering is presented as a framework with a “kernel” of three main object types: alphas, activities, and competencies. Its seven core alphas—Stakeholders, Opportunity, Requirements, Software System, Team, Way of Working, and Work—are grouped into three areas of concern: customer, solution, and endeavor (Kemell et al., 2021). Related work describes Essence as a modular, method-agnostic framework comprising a kernel and a language, with the kernel containing foundational elements integral to all software engineering methods and the language providing notational elements for specifying items not always present in every method (Evensen et al., 2018).

The startup study by Kemell et al. applies Essence as an analytic and categorization lens to empirical data from 13 software startups, using 63 previously identified startup practices plus 13 new ones from their cases (Kemell et al., 2021). The reported differences from mature organizations include the integration of business and development; rare adoption of entire formal development methodologies; evolving requirements and fluid, multi-skilled roles; emphasis on early validation and MVPs; and resource constraints combined with adaptive tooling. Examples of mappings into existing alphas include “create an MVP,” “test features with customers,” and “prioritize features” under Requirements; “tailor agile practices,” “frequent meetings,” and “use efficient communication tools” under Way of Working; and “flat organization” and “multi-skilled developers” under Team (Kemell et al., 2021).

A central finding is that some startup practices could not be satisfactorily categorized under the existing seven alphas. The paper identifies business model development, funding, and marketing as recurring concerns that are ill-served by forcing them into Stakeholders or Opportunity. It therefore proposes a fourth area of concern, Business, with at least three new alphas: Funding, Business Model, and Marketing (Kemell et al., 2021). In this formulation, iteratively developing a business model is treated as a core activity analogous in importance to software-system implementation.

The significance of this line of work lies in its shift from method labels to atomic practices. Rather than asking whether startups “use Scrum” or another named method, it decomposes work into practices and then asks whether the Essence kernel can describe them. The result is not a rejection of Essence, but a domain-specific expansion of it.

3. Extension, adoption, and organizational mappings

Several papers examine how Essence is taught, tooled, or repurposed beyond its original software-engineering setting. Essencery is a Ruby on Rails web tool for “essentializing” software engineering methods and practices using the Essence graphical syntax (Evensen et al., 2018). Its evaluation used a qualitative, quasi-formal experiment with 16 IT/SE students, analyzed through the Technology Acceptance Model. Fifteen of the 16 participants found the tool intuitive and easy to use, and all 16 reports indicated that they were able to construct the graphs they intended (Evensen et al., 2018). The tool was explicitly motivated by a tooling gap: Essence had been described as suffering from low practitioner adoption, partly because of a lack of proper tooling.

Large-scale classroom evidence points in a similar direction. In a semester-long project-based course at NTNU, over 450 second-year BSc students were organized into 102 project teams and introduced to Essence, especially the seven kernel alphas and later the practice library and practice cards (Kemell et al., 2018). Nearly all teams found Essence useful in some respect, while fewer than 10 of 102 found it outright negative or useless. The most prominent barrier was that Essence was “difficult to learn,” especially from resources perceived as long, dense, or abstract; the paper therefore argues for better introductory guides and for the pedagogical value of practice cards, progress control, and reflective method tailoring (Kemell et al., 2018).

Essence has also been mapped onto larger organizational and engineering frameworks. A study on TOGAF represents each TOGAF phase as an Essence Practice, each step as an Activity Space, specific actions as Activities, outputs as Work Products, and roles/qualifications as Competencies (Múnera et al., 2019). That work extends SEMAT with a new Governance competency, because enterprise contexts required decision-making authority not covered by the standard competence set. The mapping is used diagnostically as well as descriptively: it exposes places where TOGAF lacks precision about which activities produce which artifacts, or which roles own specific activities (Múnera et al., 2019).

For systems engineering, another line of work preserves the Essence Language unchanged but modifies the Kernel only within the engineering solution area of concern (Levenchuk, 2015). The software-centric alphas “Requirements” and “Software System” are replaced by System Definition and System Realization. System Definition contains Requirements, Architecture, and Non-architectural Design, while System Realization contains Component, Module, and (Al)Location. The relation is summarized as

$$\text{System Realization} \xleftarrow{\text{is defined by}} \text{System Definition}.$$

This adaptation is presented as a way to harmonize Essence with systems engineering standards while retaining the Language as a situational method engineering foundation (Levenchuk, 2015).

Across these studies, Essence functions not only as a theory of software engineering, but also as a meta-model for educational scaffolding, method tailoring, enterprise architecture description, and systems engineering generalization.

4. Essence as a constraint modelling language

A distinct use of the term appears in constraint programming. Here, Essence is a declarative constraint modelling language for encoding classes of constraint satisfaction problems, while Savile Row is an implementation that translates Essence models to lower-level solver input (Nightingale et al., 2016). Essence specifications typically include a header, parameter declarations using given, constants with letting, decision variables with find, optional where clauses, optional objectives, and a such that block for constraints. The language supports atomic types int and bool, compound matrix types, quantifiers such as forAll, exists, and sum, global constraints such as allDiff, and relational semantics for undefinedness, under which any boolean expression containing an undefined sub-expression evaluates to false (Nightingale et al., 2016).

A later robustness study focuses on Essence specifications refined into constraint models using Conjure (Akgün et al., 2021). Because users may write logically equivalent specifications in very different ways, Conjure can miss modelling opportunities when important domain attributes or abstract types are omitted. The paper introduces two classes of reformulation rules. Domain attribute recovery infers attributes such as size, maxOccur, total, injective, or surjective from constraints. Type strengthening rewrites more general types into more specific ones, such as mset to set when maxOccur 1 is implied, or relation to function when functional constraints are detected (Akgün et al., 2021). In one example, reformulating a relation variable into a function reduces the domain size from $2^{3^3} = 134{,}217{,}728$ to $3^9 = 19{,}683$ (Akgün et al., 2021).

Automatic feature learning extends the same ecosystem into algorithm selection. Using the Essence modelling language and a car sequencing case study, one paper learns instance features directly from high-level Essence instances with a LLM rather than from hand-crafted low-level features (Pellegrino et al., 2024). The study uses 10,214 car sequencing instances and a portfolio of 12 model–solver pairs, defined as 3 Essence Prime models combined with Kissat, Chuffed, CPLEX, and OR-Tools. Its feature extractor is an 8-bit quantized Longformer, and the learned features are passed either directly to a neural selector or to downstream selectors such as AutoFolio or $k$-means (Pellegrino et al., 2024). Feature extraction is markedly faster than fzn2feat: the neural approach reports median and mean extraction times of $0.02$ s, compared with $6.71$ s median and $f$0 s mean for fzn2feat, with maxima of $f$1 s and $f$2 s respectively (Pellegrino et al., 2024).

In this literature, Essence denotes abstraction above solver-level commitment. The language is valuable precisely because it defers modelling and representation choices, making later reformulation, refinement, and even learned algorithm selection possible.

5. Philosophical, logical, and metaphysical formulations

In formal logic, the logic of essence and accident treats essence de dicto. Its primitive essence operator satisfies

$f$3

and the accident operator is defined by $f$4 (Fan, 2015). The resulting logic is less expressive than modal logic on non-reflexive models, but equally expressive on reflexive models. The paper also develops a suitable $f$5-bisimulation, characterizes the expressive power of the logic within modal and first-order logic, and axiomatizes it over various frame classes. Among the paper’s stated contributions is settling previously open definability questions, including the symmetric case (Fan, 2015).

A different but related metaphysical tradition is developed through Avicenna’s distinction between essence and existence. In the “thinging machines” framework, essence or “thingness” is mapped to a static Region, while actual existence is mapped to a dynamic Event, summarized by

$f$6

The transition from subsistence to actuality is modeled by a Create action, and the paper introduces the “exicon” as an existence container, written $f$7 for an event occupying an existential slot (Al-Fedaghi, 2024). This account explicitly separates what something is from the fact that it exists, and contrasts that separation with Leibnizian monads, which are described as always actual existents (Al-Fedaghi, 2024).

A broader cognitive formulation appears in work on intelligence. There, abstraction is defined as the ability to capture the essence of something in space and/or time, and the intelligence process is described as realizing order by transforming signals through three levels: Physical, Information, and Abstract. At the highest level, essence is associated with “formless, timeless constructs—concepts, beliefs, theories, knowledge” (Yaworsky, 2018). This use is not a modal or ontological logic of essence, but it preserves the same emphasis on abstraction and invariance.

6. Computational and representational uses

In machine learning, “essence” often denotes a selective or high-level semantic residue. “Essence Knowledge Distillation” for speech recognition argues that not all teacher outputs should be distilled: only the top-$f$8 output probabilities are retained, the remainder are set to zero, and the result is renormalized (Yang et al., 2019). Student training combines these sparse soft targets with hard labels through

$f$9

On the 309-hour Switchboard task with output dimensionality 8912, a TDNN-LSTM student trained with top-5 outputs achieves total WER $g$0, outperforming the teacher ensemble at $g$1 and the best single model at $g$2; larger $g$3 values such as $g$4 or $g$5 do not improve results (Yang et al., 2019).

In image synthesis, “essence transfer” is distinguished from style transfer. Style transfer manipulates textures and colors, whereas essence transfer edits a source image to include the high-level semantic attributes of a target image while preserving source identity (Chefer et al., 2021). The method combines StyleGAN and CLIP under an additive structure in both latent spaces, with an optimization-based variant and an encoder-based variant. The stated goal is to guarantee identity preservation and high-level feature transfer without relying on a facial recognition network (Chefer et al., 2021).

In programming-language theory, the “essence of inheritance” is not subtyping or code reuse alone, but a communicative mechanism that mirrors the human progression from concrete examples to abstraction (Black et al., 2016). The paper calls this ex post facto parameterization: inheritance allows one to start with a concrete program and later treat a concrete element as if it were a parameter, without redesigning the code in advance (Black et al., 2016).

Program analysis introduces yet another technical usage. “Essence clones” are code pairs in which only a portion of the code is similar, but that portion embodies the core logic of each function (Zhao et al., 26 Feb 2025). ECScan detects such clones by assigning weights to code lines according to information content and then comparing weighted line vectors. Its reported average F1-score is $g$6, and the paper emphasizes that traditional syntactic methods often miss these clones because peripheral code differences dominate global similarity scores (Zhao et al., 26 Feb 2025).

These usages converge on a common operational motif: “essence” is what remains after deemphasizing low-information, stylistic, or peripheral variation.

7. Cosmological “essence” models

In cosmology, the suffix “-essence” names families of non-canonical field models. A recent observational study analyzes a $g$7-essence Lagrangian

$g$8

with inverse square and exponential potentials, constrained by Pantheon+SHOES, DES Type Ia supernovae, BAO data from SDSS and DESI, and cosmic chronometers (Hussain et al., 2024). The study uses Bayesian inference with priors informed by dynamical-system stability, and concludes that the $g$9-essence model fits well across all data combinations, although the BIC criterion slightly favors $k$0CDM (Hussain et al., 2024).

A two-field pure $k$1-essence model goes further by making the Lagrangian depend only on the covariant derivatives of two scalar fields and on a quotient of their kinetic energies, rather than on the fields themselves (Gao, 2024). The paper calls this “pure K-essence” and reports that the equation of state can be arbitrarily small and arbitrarily large. It further states that the model can play the role of inflation field, dark matter, and dark energy, and that the absence of the scalar fields themselves in the equations of motion makes the study considerably simple, allowing exact black hole solutions (Gao, 2024).

Another line of work derives $k$2-essence Lagrangians for relativistic barotropic fluids in three formulations: pressure as a function of energy density (Model I), rest-mass density (Model II), or pseudo rest-mass density in the Thomas-Fermi approximation for a complex scalar field (Model III) (Chavanis, 2021). For polytropic and logotropic unified dark matter/dark energy models, the paper recovers the Born-Infeld action of the Chaplygin gas in Models I and III and derives explicit reduced actions for the logotropic fluid in Models II and III (Chavanis, 2021).

The terminology broadens further in work on $k$3-essence and $k$4-essence, where non-canonical fermion fields and combined bosonic–fermionic models are studied as cosmological sources (Myrzakulov, 2010). That paper presents exact solutions, constructs fermionic and bosonic-fermionic DBI models as reductions of $k$5-essence and $k$6-essence, finds Chaplygin-gas counterparts, and proposes nonlinear models of bouncing and cyclic universes (Myrzakulov, 2010).

In cosmology, then, “essence” is not a theory of whatness in the philosophical sense. It is a naming convention for non-canonical effective-field constructions whose distinctive content lies in their kinetic structure and dynamical phenomenology.

Taken together, these literatures do not yield a single doctrine of essence. They do, however, repeatedly reserve the term for what is treated as central rather than peripheral: the kernel of a method, the abstract structure of a model, the invariant content of a proposition, the semantically informative part of a probability distribution or image embedding, the core logic of duplicated code, or the kinetic structure driving a cosmological scenario. This suggests that “essence” functions across disciplines as a technical shorthand for disciplined reduction to what a field regards as fundamental.