6.1 KiB
Conceptual Modeling Is the Context — Provenance Note
Status: SOURCE_REVIEWED_CONTENT_LEVEL_REFERENCE
Source URL: https://metadataweekly.substack.com/p/conceptual-modeling-is-the-context
Captured for: OTOM / Mass-Number Lens / Admissibility Forest
Article: Conceptual Modeling Is the Context Engineering Nobody Is Doing
Author: Juha Korpela
Author attribution: Written by Juha Korpela, independent consultant at Datakor Consulting, co-founder of Helsinki Data Week, and author of the Common Sense Data Substack.
Author/source link: https://metadataweekly.substack.com/p/conceptual-modeling-is-the-context
Publication: Context & Chaos / Metadata Weekly mirror
Date: 2026-04-15
Direct attribution
This note directly attributes the referenced article and its conceptual-modeling argument to:
Juha Korpela
Independent consultant, Datakor Consulting
Co-founder, Helsinki Data Week
Author, Common Sense Data Substack
Article: Conceptual Modeling Is the Context Engineering Nobody Is Doing
Source link: https://metadataweekly.substack.com/p/conceptual-modeling-is-the-context
The provided article text mentioned an author LinkedIn prompt, but no actual LinkedIn URL was present in the pasted source. Until a distinct author profile URL is available, the article/source URL above is the citation link used for attribution.
Provenance boundary
The article text was provided by the user as local pasted source material. This note now records content-level provenance and extracts project-relevant claims. Any direct quote or detailed claim should remain traceable to the user-provided article text.
Source thesis
The article argues that AI teams talk about context engineering through retrieval pipelines, semantic layers, and prompt design, while overlooking conceptual modeling: an older discipline that captured business meaning before implementation. The abandoned conceptual/logical modeling layer is exactly what AI agents now need in order to understand organizational data in context.
Key source claims
1. LLMs make context obligatory because they lack knowledge of a specific organization's idiosyncratic business semantics.
2. Conceptual modeling disappeared when Big Data-era storage systems made physical implementation design feel less necessary.
3. Data warehouses expose technical metadata, but not the business semantics of entities such as Customer, Contract, and Order.
4. Semantics belongs in a knowledge plane, separate from but linked to the data plane.
5. Without a shared knowledge plane, AI either guesses from generic averages or gets trapped in solution-level semantic silos.
6. Conceptual models can seed ontologies, knowledge graphs, context graphs, and cross-solution semantic repositories.
Relevance to OTOM
This directly supports the Mass-Number Lens doctrine that no candidate is evaluated frame-free.
No mass-number assignment is frame-free.
Every mass-number is indexed by a domain D and conceptual frame R.
In OTOM terms, conceptual modeling supplies the operational frame R that decides:
what counts as a candidate,
what counts as signal,
what counts as noise,
what counts as typed residual,
what counts as admissible alignment,
what counts as closure.
Data plane vs knowledge plane mapping
The article's data-plane / knowledge-plane split maps cleanly to the Mass-Number Lens:
DataPlane
= raw tables, files, streams, embeddings, graphs, fields, observations
KnowledgePlane
= conceptual model, ontology, glossary, semantic relationships, domain frame R
LensLink
= mapping from data-plane object to knowledge-plane concept
Mass-number evaluation should therefore be indexed as:
M_D,R(x)
where:
D = domain / data-plane substrate
R = conceptual model / knowledge-plane frame
x = candidate object or residual component
Failure modes imported into closure doctrine
The article identifies two predictable failures when the knowledge plane is missing. These become closure failure modes:
1. Missing semantics
No knowledge-plane context exists.
The system guesses from generic priors.
In OTOM: residual drift is untyped and likely hallucinated.
Closure status:
UntypedResidual | CategoryMisplaced | Quarantined
2. Semantic silos
Definitions exist only locally at the solution or dataset level.
Cross-functional workflows encounter inconsistent meanings.
In OTOM: multiple local frames R_i conflict without an adapter bridge.
Closure status:
CategoryMisplaced | TypedResidual | AdapterRequired
Integration pipeline
ObservedField
-> DataPlane object
-> ConceptualModel / KnowledgePlane frame R
-> LensLink mapping
-> Residual extraction
-> Deterministic stochastic coarse-graining
-> Invariant-energy decomposition
-> Mass Number Foci
-> Admissibility Forest
-> Closure ledger
Doctrine refinement
Conceptual modeling is the context frame that decides which invariants are visible, which residuals are typed, and which foci are admissible.
Expanded:
The knowledge plane supplies the alignment map.
Without it, the system mistakes business semantics for noise or invents average-case meaning.
Mass-Number Lens implication
A Mass Number Focus is only meaningful relative to a conceptual frame:
Focus_D,R(x)
The same data-plane structure may unfold into different foci under different conceptual models. This is not a contradiction; it is frame-indexed semantics.
Adapter theorem targets
theorem massNumber_frame_indexed :
MassNumber D R x -> FrameIndexed D R x := by
sorry
theorem missingKnowledgePlane_requiresResidualTyping :
MissingKnowledgePlane D -> RequiresResidualTyping D := by
sorry
theorem semanticSilo_requiresAdapterBridge :
SemanticSilo R₁ R₂ -> RequiresAdapterBridge R₁ R₂ := by
sorry
Guardrail
This provenance note uses Juha Korpela's article as conceptual support, not as proof of the Mass-Number Lens. The article validates the engineering need for explicit context/semantics; the OTOM closure machinery remains a separate formalization target.