Glossary¶

Definitions of key computational, graph-theoretic, and epistemological terms used throughout the Episteme documentation and codebase.

A¶

ADU (Argumentative Discourse Unit)¶

A segment of text that functions as an atomic argumentative component, such as an empirical premise, theoretical claim, or deductive conclusion.

Approximative Empirical Adequacy (Approximative empirische Adäquatheit)¶

The degree of empirical agreement between observed data \(I\) and theoretical models \(M\) under real-world measurement tolerances. It is fulfilled if the deviation remains strictly within the boundaries of admissible blurs (\(\exists X \in \text{Cn}(K) \text{ s.t. } I \approx_A X\)) (Balzer et al., 1987).

APOC (Awesome Procedures On Cypher)¶

Neo4j plugin providing additional graph algorithms, data integration utilities, and procedural cypher extensions.

Argumentative Discrepancy (Argumentative Diskrepanz)¶

A tension between normative rational reconstruction and historical description where structurally foundational, high-centrality axioms (implicit core hypotheses) are barely mentioned in surface text. Reflects the gap between the context of discovery and the context of justification (Stegmüller, 1976).

Artifact¶

Persistent, content-addressed output from a pipeline phase (stored under .pipeline_artifacts/) for execution caching, resume workflows, and auditability.

B¶

Balzer-Gähde Invariance Criterion (Balzer-Gähde Invarianz-Kriterium)¶

A formal structuralist criterion stating that a concept \(t\) is non-theoretical with respect to a theory \(T\) if it remains invariant under arbitrary theoretical modifications as long as the fundamental laws are preserved (Balzer et al., 1987).

C¶

Centrality¶

A structural graph-theoretic metric quantifying the relative topological importance, influence, or prominence of a node within a network (Newman, 2018). See also Degree Centrality, Eigenvector Centrality, and PageRank.

Chunk¶

A segment of parsed document text (typically 500–1500 tokens) partitioned along rhetorical section boundaries, serving as the unit of ingestion in Layer 1.

Concept¶

An abstract theoretical entity representing an intellectual notion, paradigm construct, or philosophical category.

Confidence Score¶

A normalized numerical measure (\(c \in [0, 1]\)) indicating extraction certainty or relationship validity.

Constraint (\(C, CL\))¶

A cross-application relational constraint (\(C \subseteq \mathcal{P}(M_p)\)) demanding that intrinsic theoretical parameters (e.g., mass, charge, or utility preferences) maintain invariant, identical values when an entity appears across overlapping application domains (Stegmüller, 1976; Balzer et al., 1987).

D¶

Degree Centrality¶

A baseline centrality metric defined as the number of edges incident to a node. Distinguished in directed graphs as in-degree (incoming connections, indexing prestige or evidentiary vulnerability) and out-degree (outgoing connections, indexing deductive generative power) (Newman, 2018).

Degeneration Index (Lakatosianische Degeneration)¶

The ratio of ad-hoc resolved anomalies (Type-b failures) to the sum of verified empirical successes and honest direct refutations (\(D = |M_b| / (|E| + |M_a|)\)). A monotonically rising index indicates a degenerating research programme (Schurz, 2014; Lakatos, 1978).

E¶

Eigenvector Centrality¶

A spectral centrality metric where a node's score is proportional to the sum of the centralities of its neighbors (\(x_i = \kappa^{-1} \sum_j A_{ij} x_j\)), weighting connections by the relative prominence of adjacent nodes (Newman, 2018).

Empirical Focus (Empirischer Fokus)¶

The capacity of theoretical terms and their constraints to restrict the volume of admissible observable states, thereby increasing the theory's empirical content and refutability (Stegmüller, 1976).

Entity¶

A conceptual or real-world object extracted from text, typed according to an ontology (e.g., Concept, Person, Theory, Work).

Entity Linking¶

The algorithmic process of resolving surface text mentions across distinct chunks and documents to a stable canonical node ID using dense vector similarity and cross-encoder reranking.

Epistemic Cycle (Epistemischer Zyklus)¶

The methodological danger of circular justification arising when the centrality of a node is determined purely through a graph network whose own validity is grounded in that same node (Stegmüller, 1976; Balzer et al., 1987).

Event Emitter / Domain Event Bus¶

An asynchronous publish-subscribe subsystem broadcasting structured domain events during pipeline execution for telemetry, logging, and Langfuse tracing.

F¶

Faithfulness¶

An intrinsic evaluation metric measuring whether extracted entities and relations are strictly warranted by the source text without hallucination.

Fusion¶

Phase 5 of the pipeline, which clusters semantically equivalent argument components across documents and computes macroscopic theoretical communities via the Leiden algorithm.

G¶

Graph Schema¶

The formal specification of allowed vertex labels, relationship edge types, and property constraints within the Neo4j property graph.

Graph Store¶

The persistent graph database backend (Neo4j) accessed via an asynchronous Cypher driver.

H¶

Homogeneity (Homogenität)¶

The overarching requirement of structural, semantic, or methodological unity within a scientific theory.

• Methodological Homogeneity (Konstruktvalidität): Ensures empirical indicators strongly correlate because they measure a common theoretical latent cause.
• Semantic-Axiomatic Homogeneity: A formal non-factorizability criterion stating that an axiomatization \(Ax\) cannot be partitioned into disjoint subsets whose empirical consequences are mutually independent, preventing the Tacking Paradox (Schurz, 2014).

I¶

Intertheoretical Links / Community Bridges¶

Directed relational edges connecting distinct theory-elements across different disciplines or schools of thought. Genuine theoretical translations preserve global homogeneity, whereas ad-hoc bridge edges violate it (Balzer et al., 1987).

K¶

Katz Centrality¶

An extension of eigenvector centrality assigning a baseline score \(\beta\) to all nodes in addition to the weighted sum of neighbor centralities (\(x_i = \alpha \sum_j A_{ij} x_j + \beta\)), preventing vanishing scores in directed acyclic topologies (Newman, 2018).

Knowledge Graph (KG)¶

A traditional factual graph focusing on concrete real-world entities and deterministic relations, contrasted with dialectical Theory Graphs.

L¶

LiteLLM¶

A unified provider-agnostic abstraction layer standardizing API calls to OpenAI, Anthropic, Ollama, and local models.

LLM (Large Language Model)¶

Deep transformer models utilizing billions of parameters for contextual text understanding, extraction, and argument classification.

M¶

Failure Type a / Direct Contradiction (Misserfolg Typ a)¶

A well-established empirical observation that logically or probabilistically contradicts a theory version directly via Modus Tollens (Schurz, 2014).

Failure Type b / Immunized Failure (Misserfolg Typ b)¶

An empirical anomaly that contradicted an earlier theory version, but was neutralized in the current version via an ad-hoc auxiliary clause without generating verified novel predictions. Remains a latent failure in the Degeneration Index (Schurz, 2014).

N¶

NER (Named Entity Recognition)¶

The process of identifying and categorizing named entities and abstract concepts within text chunks.

Non-Statement View / Model-Theoretic Approach (Modell-Ansatz)¶

The structuralist philosophy of science perspective that analyzes theories as classes of mathematical model structures (\(M_p, M, M_{pp}\)) rather than flat sets of linguistic sentences (Stegmüller, 1976; Balzer et al., 1987).

P¶

PageRank¶

A variant of Katz centrality that scales a node's contributed centrality inversely by its out-degree, preventing nodes with broad broadcast connectivity from disproportionately inflating target scores (Newman, 2018).

Perspective / Historical Adequacy (Perspektive)¶

The degree of conceptual congruence between a formal reconstruction of a theory and the measurement apparatus and observational vocabulary historically available at time \(t\), avoiding presentism (Stegmüller, 1976).

Phase¶

A discrete stage of pipeline execution with strictly defined Pydantic input and output boundaries.

Prognostic Power (Prognostische Leistung)¶

The capacity of theoretical constraints crossing overlapping application domains to enable strict empirical predictions in previously unmeasured domains (Stegmüller, 1976).

Provenance¶

Deterministic tracking metadata (chunk ID, character offsets, source citation) connecting graph elements directly to source literature.

R¶

Ramsey-Sneed Claim (Ramsey-Sneed-Behauptung)¶

The structuralist method for eliminating circularity in theoretical terms by asserting that an empirically described situation (\(M_{pp}\)) can be enriched by theoretical functions such that the resulting structure constitutes a valid model (\(M\)) satisfying global constraints (Balzer et al., 1987).

Relationship¶

A directed edge connecting two entities or argument components, annotated with semantic type and confidence.

Retrospective Bias / Presentism (Retrospektive Verzerrung)¶

The historiographical flaw of analyzing historical scientific theories through modern concepts or measurement standards unavailable to the original authors (Schurz, 2014).

S¶

SAME_AS¶

A structural equivalence relationship indicating that two entity mentions or argument components across different texts denote the identical concept.

Sneed's Functional Criterion for T-Theoreticity (Sneed'sches Kriterium)¶

A criterion defining a term \(\tau\) as \(T\)-theoretical if every available method for determining its value presupposes the validity of the laws of \(T\) (Balzer et al., 1987).

Strong Empirical Adequacy (Starke empirische Adäquatheit)¶

The idealized condition that all intended applications \(I\) can be enriched into exact theoretical models without observational noise (\(I \in \text{Cn}(K)\)). In practice, replaced by Approximative Empirical Adequacy with admissible blurs (Balzer et al., 1987).

Statement View (Satz-Ansatz)¶

The traditional linguistic view of scientific theories as flat sets of verbal or axiomatic sentences, contrasted with the structuralist model-theoretic approach (Stegmüller, 1976).

T¶

Tacking Paradox (Klebeparadoxon)¶

The logical anomaly where conjoining an arbitrary irrelevant hypothesis \(H\) to a confirmed theory \(T\) technically increases its empirical content (\(E(T \wedge H) \supset E(T)\)). Prevented by demanding axiomatic homogeneity (Schurz, 2014).

Theory-Tree (Theorie-Baum)¶

A specialization hierarchy \((TN, \alpha)\) characterized by a singleton root element \(T_0\) (\(B(N) = \{T_0\}\)), ensuring that all specialized variants descend from a single fundamental core (Balzer et al., 1987).

Theory-Holon (Theorie-Holon)¶

The macroscopic global network of all scientific theories, interconnected via directed intertheoretical links and community bridges (Balzer et al., 1987).

Theory Graph (TheoryNet)¶

A layered property graph and argumentation network (\(TF = (At, R) \cong G = (V, E, \lambda_v, \lambda_e)\)) capturing conceptual abstractions, evidentiary dependencies, and defeasible arguments.

Triple¶

A subject-predicate-object semantic assertion with associated confidence and scope (local vs. global).

V¶

Vector Index¶

Neo4j or in-memory vector index enabling dense Maximum Inner Product Search (MIPS) over contextual embeddings.

W¶

Workflow¶

The coordinated execution sequence of pipeline phases transforming raw documents into materialized Theory Graphs.

Z¶

Admissible Blurs (Zulässige Unschärfen, \(\mathcal{A}\))¶

The family of acceptable empirical error tolerances within a uniform space \((M_p, \mathcal{U})\), enabling realistic evaluation of approximative adequacy while preventing vacuous immunization (Balzer et al., 1987).