Human Capital Ontology Overview

• Human Capital Ontology (HCO) is a formal framework representing U.S. federal human-capital standards and labor market structures with a modular, layered design.
• HCO uses nested skill dependencies and quantitative measures to map career trajectories, offering insights into occupation complexity and wage implications.
• Extensions of HCO incorporate AI-augmented components by decomposing human capital into physical, routine-cognitive, and augmentable cognitive skills to assess wage effects.

Searching arXiv for the specified papers to ground the article in the cited sources. arXiv search query: Human Capital Ontology Human Capital Ontology (HCO) denotes, in the literature represented here, both a concrete ontology for U.S. federal human-capital standards and a broader ontology design pattern for representing skills, occupations, persons, career trajectories, and quantitative labor-market annotations. In its explicit arXiv formulation, HCO represents data standards maintained and employed by the Office of Personnel Management (OPM) to represent Human Capital Operations, classify job positions, encode Nature of Action (NOA) codes, and provide crosswalks between OPM Occupational Series and BLS Standard Occupational Classification codes, while extending the Common Core Ontologies (CCO) and the Basic Formal Ontology (BFO) (Babcock et al., 26 Jul 2025). In adjacent work, HCO is also used as the target schema for formalizing nested skill dependencies, occupational complexity, AI-conditioned complementarity and substitution, and career mobility using O*NET, OEWS, CPS, resume, and household-survey data (Hosseinioun et al., 2023, Lee et al., 15 Jun 2025, Maya, 1 Apr 2026).

1. Institutional scope and upper-level commitments

The explicit HCO described by Babcock, Farrington, and Gugliotti is an extension of BFO and CCO. Its purpose is to represent OPM data standards used to classify and manage U.S. Federal Government positions and to represent human resource personnel actions. The ontology covers OPM Nature of Action codes and their identifiers, Occupational Groups and Job Families, the Occupational Series into which they subdivide, classification and grading for both white- and blue-collar jobs, and crosswalks between OPM Occupational Series and BLS Standard Occupational Classification codes (Babcock et al., 26 Jul 2025).

Its principal modeling commitments are strongly realist in upper-level alignment but information-centric in the representation of administrative artifacts. Position is modeled as a subclass of CCO’s Directive Information Content Entity, WorkDuty as a subclass of Directive Information Content Entity, AggregateOfPositions as an information content entity that has Positions as parts, and OccupationalSeries as a subclass of AggregateOfPositions. PersonnelAction is modeled as a subclass of CCO’s Planned Act. NatureOfActionIdentifier is a subclass of CCO’s Designative Information Content Entity, and NatureOfActionCode is a subclass of CCO’s Code Identifier. HCO also imports CCO Document Acts to represent deontic roles, and BFO’s continuant part of relation is used to link instances of Position to the corresponding instance of Occupational Series (Babcock et al., 26 Jul 2025).

These choices have direct semantic consequences. Treating Position as a Directive Information Content Entity preserves the ability to represent vacancies and the prescriptive nature of work duties. Treating personnel actions as Planned Acts supports a taxonomy of action types such as appointment actions, separations, conversions, and pay adjustments. A common simplification is to treat HCO as merely an occupational code list; the published ontology instead distinguishes positions, duties, personnel actions, identifiers, and codes as separate ontological categories. At the same time, the paper is explicit about what it does not specify: namespaces, ontology IRI and version IRI, licensing, domain and range constraints, cardinalities, SHACL or ShEx validation, and detailed modeling of grades, pay plans, and white-/blue-collar classification are not provided (Babcock et al., 26 Jul 2025).

2. Nested human capital and prerequisite structure

A second line of work uses HCO as the target representation for a cumulative and sequential view of human capital. In that formulation, human capital is operationalized as a network of interconnected skills, knowledge, and abilities embodied in occupations and individuals. Its structure is cumulative and sequential: specific, advanced capabilities are often contingent on broader, more general skills acquired earlier in education or on the job. The central structural concept is nestedness, a hierarchical pattern in the skill–occupation bipartite system in which specific skills tend to appear in occupations that also require general skills, but not symmetrically (Hosseinioun et al., 2023).

The empirical substrate is O*NET 2019 for main results and O*NET 2005 for temporal comparisons, with BLS crosswalk alignment from 8-digit SOC to 6-digit SOC where needed. Occupations are characterized by quantitative skill requirements using O*NET skill level and importance, with $Level_{o,s} \in [0,7]$ and importance on a $1$–$5$ scale. Continuous $Level_{o,s}$ values are transformed to binary skill presence $m_{os}$ through a disparity filter that preserves the rank of skills and occupation skill-intensity. From the resulting binary skill–occupation matrix, asymmetrical conditional probabilities are computed:

$$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$

Directionality is then induced through either log-lift

$$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$

or the asymmetry index

$A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$

A dependency edge $s_j \rightarrow s_i$ is created when co-occurrence is statistically significant after filtering and asymmetry exceeds a pair-specific threshold, encoding that acquiring or applying $s_i$ is more likely contingent on $1$0 than vice versa (Hosseinioun et al., 2023).

The paper reports examples such as Mathematics $1$1 Programming and Oral expression $1$2 Negotiation, while Math $1$3 Dynamic flexibility is filtered out by low prevalence and independence. Generality is measured both substantively and network-theoretically: general skills are widely required across occupations at level $1$4, specific skills are needed at elevated levels in few occupations, and local reaching centrality in the dependency network provides a network-based generality measure that correlates with median level across occupations with $1$5. Demand-profile clustering yields 31 general, 43 intermediate, and 46 specific skills (Hosseinioun et al., 2023).

Nestedness is evaluated with multiple metrics—NODF, matrix temperature, checkerboard score, and overlap index $1$6—and all point in the same direction. Between 2005 and 2019, the checkerboard score declines from $1$7 to $1$8, temperature declines from $1$9 to $5$0, NODF rises from $5$1 to $5$2, and overlap index $5$3 rises from $5$4 to $5$5, indicating that the nested structure became more pronounced over time. At the skill level, alignment is measured by

$5$6

with $5$7 counterfactual matrices per skill preserving degree. At the occupation level, nestedness alignment is defined as

$5$8

These quantities connect ontology design directly to labor-market outcomes: higher $5$9 correlates with higher required education and wages and with lower automation risk; occupations with higher $Level_{o,s}$0 tend to have higher education requirements; and resume sequences show strong early accumulation of general and nested-specific skills in the first five transitions, with randomized sequences yielding $Level_{o,s}$1 skills $Level_{o,s}$2 (Hosseinioun et al., 2023).

Within HCO design, this framework yields a directed Skill DAG with object properties such as requiresSkill, prerequisiteOf, buildsUpon, applicableToOccupation, and transitionTo, and data properties such as skillLevelRequired, skillImportanceRequired, conditionalDependencyWeight, asymmetry, log-lift, alignmentScore, nestednessContribution, nestednessLevel, localReachingCentrality, wagePremiumEstimate, educationYearsRequired, automationRisk, demographicDisparityAnnotations, timePeriod, provenance, and confidenceInterval. Acyclicity, transitive closure, and inference from prerequisite edges to inferred occupational requirements are explicit design recommendations rather than merely descriptive annotations (Hosseinioun et al., 2023).

3. Complexity indices, communities, and the centrality of general skills

A third formulation of HCO is based on economic complexity analysis of the occupation–skill network. Using O*NET 28.0 and OEWS May 2023, the underlying dataset covers 872 occupations at 8-digit O*NET-SOC, aggregated to 6-digit SOC for regression analysis, and 120 O*NET KSA variables constituting the skill set. O*NET importance ratings on a 1–5 Likert scale are binarized at importance $Level_{o,s}$3 to produce a bipartite adjacency matrix $Level_{o,s}$4, with occupation diversity $Level_{o,s}$5 and skill ubiquity $Level_{o,s}$6 (Lee et al., 15 Jun 2025).

Skill proximity is defined by co-requirement across occupations:

$Level_{o,s}$7

Louvain community detection is run 100 times on the skill network, yielding three stable communities: general, cognitive, and physical skills. General skills correspond to Basic and Social skills such as Oral Expression, Active Listening, and Coordination; cognitive skills correspond to Knowledge and Cognitive Abilities such as Economics and Accounting, Science, and Medicine and Dentistry; physical skills correspond to Physical, Sensory, and Psychomotor Abilities such as Manual Dexterity, Static Strength, and Mechanical. The general skill community is characterized by high ubiquity and centrality in nested structure, while specialized skills split into two distinct clusters aligned with occupational specialization (Lee et al., 15 Jun 2025).

The Occupational Complexity Index (OCI) and Skill Complexity Index (SCI) are derived through the Method of Reflections, whose high-order values stabilize and correspond to the eigenvector associated with the second-largest eigenvalue of the network matrix. Standardized high-order values define OCI for occupations and SCI for skills. The resulting structure places general skills at the nested core, while cognitive and physical skills diverge in opposite directions. Skill-level contribution scores are obtained through null-model reshuffling with 1,000 simulations:

$Level_{o,s}$8

General skills have the highest ubiquity and the largest $Level_{o,s}$9 values. Cognitive and physical skills contribute equally to modularity through $m_{os}$0, but their interactions with general skills differ: cognitive skills are more integrated with general skills, whereas physical skills are more isolated (Lee et al., 15 Jun 2025).

This structure is tied to wage formation through community-weighted skill shares. SCI is min–max normalized to $m_{os}$1, weighted by O*NET importance to form community totals $m_{os}$2, and normalized to obtain

$m_{os}$3

In the wage specification,

$m_{os}$4

general skills have a weaker direct effect than a moderating role. In Model (4), the main effect of cognitive skills is $m_{os}$5, $m_{os}$6; cognitive $m_{os}$7 general is $m_{os}$8, $m_{os}$9; the main effect of physical skills is $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$0, $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$1; and physical $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$2 general is $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$3, $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$4. The model with only general skills has lower explanatory power, with $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$5. For HCO, this motivates explicit classes for SkillCommunity, ComplexityIndex, and StructuralContribution, and datatype properties for OCI, SCI, ubiquity, diversity, modularity contribution, nestedness contribution, and community-specific skill shares (Lee et al., 15 Jun 2025).

4. AI-augmented human capital and institutional conditioning

A fourth strand generalizes HCO to AI-augmented economies by decomposing human capital into three orthogonal components:

$$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$6

Here $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$7 is physical–manual, $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$8 is routine–cognitive, and $$P(s_i) = \frac{n(s_i)}{N}, \qquad P(s_i \mid s_j) = \frac{n(s_i \wedge s_j)}{n(s_j)}.$$9 is augmentable–cognitive. The framework assumes $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$0, $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$1, and $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$2, with disjoint task partitions and standardization used to enforce orthogonality at the worker or occupation level depending on measurement granularity. AI capital interacts asymmetrically with these components: AI substitutes for routine cognitive work and complements augmentable cognitive work through an amplification function $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$3, where $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$4 denotes AI adoption intensity (Maya, 1 Apr 2026).

The adoption variable $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$5 is operationalized at sector $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$6 occupation-group cells as a standardized composite of formality rate, mean education, mean income, and large-firm share, with weights $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$7, $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$8, $$D_{ij} = \log \left[\frac{P(s_i \mid s_j)}{P(s_i)}\right]$$9, and $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$0, respectively. The amplification function is increasing and concave, with $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$1 and a finite upper bound:

$A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$2

The corresponding wage equation augments the standard Mincer specification with cognitive components, AI interactions, and institutional conditioning:

$A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$3

Its signature predictions are $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$4, $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$5, and $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$6, and the paper argues that the standard Mincer equation is misspecified in AI-augmented economies because it omits human-capital composition and AI-conditioned returns (Maya, 1 Apr 2026).

The measurement pipeline is explicitly LLM-based. A total of 18,796 O*NET task statements were mapped to 440 Colombian occupations using chained SOC $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$7 ISCO-08 $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$8 CIUO-08 AC crosswalks covering 99.9% of weighted employment. Claude Haiku 4.5 scored each task for augmentation potential $A_{ij} = P(s_i \mid s_j) - P(s_j \mid s_i).$9, substitution risk $s_j \rightarrow s_i$0, and augmentation type. Occupation-level $s_j \rightarrow s_i$1 and $s_j \rightarrow s_i$2 are weighted averages of these scores using O*NET task importance, standardized to mean 0 and variance 1 across occupations. Validation is reported through convergent, discriminant, and reliability statistics: $s_j \rightarrow s_i$3, $s_j \rightarrow s_i$4, $s_j \rightarrow s_i$5, $s_j \rightarrow s_i$6, and on a 20% rescored subsample Claude Sonnet 4 yields $s_j \rightarrow s_i$7, $s_j \rightarrow s_i$8, and Krippendorff’s $s_j \rightarrow s_i$9 after level-bias adjustment (Maya, 1 Apr 2026).

Empirically, the wage return to $s_i$0 rises with AI adoption in the formal sector, with $s_i$1 and $s_i$2, while informal workers cannot capture augmentation rents, with $s_i$3 and $s_i$4. The triple interaction $s_i$5, $s_i$6, identifies formality as the binding mechanism. The augmentation premium is strongest for ages 46–65, with $s_i$7, $s_i$8, negligible for ages 18–30, and especially large in Health, with $s_i$9, $1$00, and Education, with $1$01, $1$02, while Agriculture and Manufacturing are negative or null. For HCO, this implies dedicated classes for HumanCapitalComponent, AI_Capital, AI_Adoption, AmplificationFunction, Institution, and FormalityStatus, together with formal inference rules linking wage effects to $1$03, $1$04, $1$05, and formality (Maya, 1 Apr 2026).

5. Schema patterns, inference layers, and queryable representations

Across these strands, HCO functions less as a single immutable schema than as a layered representational architecture. The concrete OPM ontology contributes administrative and classificatory entities; the nestedness and complexity blueprints contribute occupation–skill network structure and temporal annotations; and the AI-augmented blueprint contributes orthogonal human-capital components, AI adoption, and institution-conditioned wage semantics. This suggests a modular HCO with upper-level ontological commitment, labor-market graph analytics, and econometric annotation layers (Babcock et al., 26 Jul 2025, Hosseinioun et al., 2023, Lee et al., 15 Jun 2025, Maya, 1 Apr 2026).

Source	Core classes	Distinctive semantics
(Babcock et al., 26 Jul 2025)	Position, WorkDuty, OccupationalSeries, PersonnelAction, NatureOfActionIdentifier, NatureOfActionCode	OPM classification, NOA designation chains, BFO/CCO alignment
(Hosseinioun et al., 2023)	Skill, Occupation, Person, TimeSlice	Directed prerequisite graph, nestedness, alignment, transitions
(Lee et al., 15 Jun 2025)	SkillCommunity, ComplexityIndex, StructuralContribution, NetworkEdge	OCI/SCI, modularity, nestedness contribution, community-weighted shares
(Maya, 1 Apr 2026)	HumanCapitalComponent, AI_Capital, AI_Adoption, AmplificationFunction, Institution, FormalityStatus	$1$06 decomposition, $1$07, corrected wage semantics

The design recommendations are concrete. The nestedness-based blueprint specifies Skill, Occupation, Person, and TimeSlice as core classes, with requiresSkill, prerequisiteOf, buildsUpon, applicableToOccupation, and transitionTo as object properties, and with quantitative annotations such as skillLevelRequired, skillImportanceRequired, asymmetry, log-lift, conditionalDependencyWeight, nestednessContribution, alignmentScore, localReachingCentrality, wagePremiumEstimate, educationYearsRequired, automationRisk, timePeriod, provenance, and confidenceInterval. It also recommends representing skill prerequisites as a DAG, enforcing acyclicity, computing transitive closure, and inferring occupational requirements when a required skill has upstream prerequisites above threshold (Hosseinioun et al., 2023).

The complexity-based blueprint adds SkillCommunity instances for General, Cognitive, and Physical; ComplexityIndex instances for OCI and SCI; reified occupation–skill edges storing both the binary requirement and the original O*NET importance; and skill–skill co-occurrence edges with proximity weights. It further recommends maintaining O*NET KSA identifiers, 8-digit O*NET-SOC occupation identifiers, 6-digit SOC crosswalks for wage data, and provenance for O*NET v28.0 and OEWS May 2023 (Lee et al., 15 Jun 2025).

The AI-augmented blueprint contributes data properties for augmentability_score, substitution_risk, physical_manual_index, AI_adoption_intensity, wage, experience_years, age_bin, education_years, education_level, formality, sector_code, occupation_code, and country_code, alongside inference rules such as: if formality is true and $1$08, then $1$09; if formality is false, $1$10 in the Colombian data; and if task type is routine_cognitive and $1$11 increases, expected $1$12 (Maya, 1 Apr 2026).

The explicit OPM ontology already demonstrates the query orientation of HCO. Its competency questions include: which positions belong to a given OPM Occupational Series; what PersonnelAction is designated by a particular NOA code; and what BLS SOC occupation corresponds to a given OPM Occupational Series. The research-oriented blueprints extend this query space to prerequisite discovery, high-risk low-alignment occupations, career moves that maximize $1$13, detection of high augmentability under formal AI adoption, and institutional comparisons of augmentation premiums. A common misconception is that ontology work ends at taxonomy construction; in these sources, ontology design is coupled to versioned data ingestion, quantitative annotation, and inferential services over labor-market structure (Babcock et al., 26 Jul 2025, Hosseinioun et al., 2023).

6. Limitations, scope conditions, and future extensions

The literature imposes clear scope conditions on HCO. The nestedness framework is built on U.S.-centric O*NET, CPS, and OEWS data, with proprietary resume data from Burning Glass/Lightcast; occupation coding changes require crosswalks; and conditional dependencies reflect co-requirements and acquisition plus firm-imposed seniority, so causal interpretations are limited even though robustness checks strengthen associations. The complexity framework is cross-sectional, sensitive to the binarization threshold at importance $1$14, and uses a nestedness proxy rather than full NODF. The AI-augmented framework depends on local crosswalks, LLM scoring, and a sector $1$15 occupation-group composite for $1$16; measurement error in $1$17 and $1$18 attenuates estimates, and endogeneity of $1$19 remains a concern, with panel data or instruments preferable for causal interpretation (Hosseinioun et al., 2023, Lee et al., 15 Jun 2025, Maya, 1 Apr 2026).

The concrete OPM ontology has a different limitation profile. It states scope over grades, pay plans, and white-/blue-collar classification, but specific modeling details are not provided. The paper also does not specify the property names for linking personnel actions to employees or the exact predicate for SOC crosswalks, and it does not provide domain/range or cardinality constraints, namespace governance, serialization choice, or validation artifacts. Its stated future work is further axiomatization of Occupational Series and continued application across the U.S. Government (Babcock et al., 26 Jul 2025).

For HCO development more broadly, the dominant extension themes are temporalization, interoperability, and fairness-aware mobility analysis. The nestedness blueprint recommends versioned, time-aware representation of O*NET releases, explicit storage of confidence intervals and provenance, regular recomputation of system nestedness $1$20, skill-level $1$21, and occupation-level $1$22, and fairness modules that attach group-specific ratios for skill levels, education, wages, and temporal trends. It also recommends aligning O*NET skills and SOC occupations to ESCO, ISCO, and other international taxonomies via owl:sameAs mappings. The complexity blueprint recommends TimeSlice and VersionedIndex classes, Credential, Course, and Program classes, and additional structural analytics such as core–periphery indices, k-core decompositions, and nestedness variants. The AI-augmented blueprint recommends periodic refresh of augmentability and substitution scores to track model drift and preserve institutional context in cross-country generalization (Hosseinioun et al., 2023, Lee et al., 15 Jun 2025, Maya, 1 Apr 2026).

Taken together, these sources define HCO not as a single settled ontology, but as a family of formally related representational systems for positions, duties, personnel actions, occupations, skills, complexity, AI augmentation, and mobility. The explicit federal ontology provides the administrative backbone; the nestedness, complexity, and AI-augmented blueprints provide quantitative semantics for cumulative skill acquisition, structural embeddedness, and institution-conditioned returns. A plausible implication is that future HCO work will be most effective when these layers remain distinct but interoperable: upper-level ontological rigor for administrative standards, graph-based semantics for skill structure, and versioned statistical annotations for labor-market dynamics.