C-K theory (Concept-Knowledge design theory)

Status: Documented, not shipped · Evidence: C · Family: Divergent ideation and idea expansion · Verdict: reject (2026-06-11)

What it is

C-K theory (Concept-Knowledge design theory) is a formal theory of design reasoning developed by Armand Hatchuel and Benoit Weil at Mines ParisTech - sketched by Hatchuel in the mid-1990s, introduced to the engineering-design community at ICED 2003, and given its consolidated formulation in Research in Engineering Design in 2009. Its single founding move is a partition of design work into two spaces with different logics:

• K space (Knowledge): propositions that have a logical status - given what the designer currently knows, each is decidably true or false.
• C space (Concepts): propositions of the form “there exists an object x with properties p1…pn” that are undecidable in current K - present knowledge can neither prove nor refute them (“a chair with no legs,” “an engine that breathes the Martian atmosphere”). A concept is, by definition, neither true nor false yet.

Design, in this theory, is the process that transforms an undecidable concept into a decidable definition in K (a “conjunction” - the finished design), and it proceeds by iterating four operators: C->C (partition the concept by adding attributes; a restrictive partition stays within the object’s known identity, an expansive partition deliberately breaks it), C->K (test or develop a concept against knowledge, possibly terminating in a conjunction), K->C (use knowledge to inject new attributes or disjunctions into the concept tree), and K->K (expand knowledge itself - learn, experiment, acquire new science). The signature claims: holding concepts exempt from truth evaluation protects wild ideas from premature feasibility-kill, and expansive partitions - the attribute that breaks the object’s identity - are the engine of genuinely novel design. The canonical worked example in the 2009 paper is the design of Magnesium-CO2 engines for Mars exploration.

Two derivatives matter for grading. The descriptive derivative is C-K’s account of fixation: ideation defaults to the most accessible knowledge, which yields only restrictive partitions. The operational derivative is the KCP method (Knowledge-Concepts-Proposals), a multi-week facilitated workshop process for organizations, deployed at RATP and Thales among others. The candidate vetted here is the thinking move - “draw the two spaces and iterate the operators” - not the workshop.

When it helps / when it misleads

It helps in three honest ways. As a lens, it explains why brainstorms fixate and gives a vocabulary for the fix: the restrictive-vs-expansive distinction is the part of the theory with real experimental backing, and it is genuinely useful for classifying stimuli and prompts. As a program structure, the KCP derivative gives large organizations a way to run months-long innovative-design explorations whose deliverable includes a map of what was explored and what knowledge is missing - that is its actual, documented industrial use, with trained facilitators. As a boundary marker, its own scope statement is valuable: C-K applies when the desired object’s very definition is unknown, which is exactly when optimization and selection methods mislead.

It misleads when treated as a solo step-by-step procedure, and this is the decisive problem:

• No next-action criterion. The theory supplies no rule for which operator to apply next, which property to introduce, whether to prefer a restrictive or an expansive partition, or when to force a conjunction. Its named critics state this directly (Choulier, Coatanea and Forest, below). What remains is notation, and notation does not steer.
• A C-K map is unfalsifiable as an artifact. Any brainstorm can be retro-fitted into a C-K tree after the fact; nothing in the formalism distinguishes a good map from a bad one. Contrast morphological analysis, whose cross-consistency pruning yields a checkable output, or an issue tree, whose MECE property can be audited.
• Notation gets mistaken for thinking. Producing the diagram feels like progress and can certify exploration that never happened.
• Agents degrade inside it. The one direct LLM-agent study (Ma, Brubaker and Goucher-Lambert, below) found agents scripted to follow C-K collapse into predominantly C->C transitions with diversity trending downward - the missing action-selection criterion made concrete.

Do not reach for it at all when the problem is decidable - choosing among known options, optimizing a known design, estimating a known quantity. The theory itself says so.

What the evidence says

The honest grade is C (conceptually plausible but under-tested), confirming the registry’s preliminary grade. The record contains real science, but almost none of it lands on the candidate move.

The theory and its formalizations (conceptual only). Hatchuel and Weil (ICED 2003; Research in Engineering Design 2009) define the spaces and operators rigorously, and Hendriks and Kazakci (DESIGN 2010) give a first-order-logic account of the dual expansion of concepts and knowledge. This establishes internal coherence, not effectiveness. Formal rigor is not outcome evidence.

One strong experimental line - adjacent to the theory, not of it (transferred, capped). Agogue, Kazakci and colleagues (2014, Journal of Creative Behavior) ran the “egg task” experiments: participants generate ideas to ensure a hen’s egg dropped from 10 m does not break, after exposure to different example types. Restrictive examples (solutions from the most accessible knowledge) constrained originality; expansive examples (solutions that modify the object’s identity) reduced fixation and stimulated originality. This is a real controlled result, and C-K supplied the classification of the examples. But it tests example types as stimuli, with C-K as the explanatory lens; nobody in the study drew a C-K map or ran the four operators. By this library’s conservative rule, evidence for a theory’s classifier cannot be laundered into evidence for practicing the method. It supports C (“the conceptual core has predictive content”) and nothing higher.

Industrial adoption (practitioner-case evidence, on a different unit of analysis). Hatchuel, Le Masson and Weil (DESIGN 2004) report early industrial applications; Agogue and Kazakci (2014) survey ten years of academic and industrial impact; the KCP method became a working method at RATP and then Thales, and in 2009 five companies (Dassault Systemes, RATP, Renault, Thales, Vallourec, later joined by SNCF and STMicroelectronics) funded the Design Theory and Methods for Innovation chair at Mines ParisTech. This is genuine adoption - but of a multi-week, expert-facilitated organizational process, reported as cases by the theory’s own school, with no controlled comparison anywhere in the record. Adoption facts, not outcome evidence, and on the workshop, not the solo move.

The direct agent test cuts the other way. Ma, Brubaker and Goucher-Lambert (ASME IDETC 2025; journal version Journal of Mechanical Design 148(7), 2026) built LLM agents that adhere to C-K theory’s wording and framework and evaluated them with mixed methods across simulation experiments plus a content analysis of the agents’ transition rationales. Headline observations: the C->C transition was the predominant operation, and diversity trended downward. The study’s purpose was theory development (using simulation to find gaps in the theory), not validation, and it includes no baseline outcome comparison against unstructured ideation - so it cannot raise the grade; as the only agent-context evidence, its degenerate operator distribution is consistent with the critics’ operationalization complaint rather than with a working agent artifact.

The named critique. Choulier, Coatanea and Forest (ICED 2011, “CK, an engineering design theory? Contributions, limits and proposals,” with the companion paper “The Engineering Design CK Theory: Contributions and Limits”) conclude that despite many practical applications the theory “has not been operationalized,” that its real operability scope “is probably limited compared to the claims” of its creators, and that there is an “absence of any criteria allowing the designer to decide on the next action to be taken” - which property to introduce, restrictive or expansive partition, or when to force a conjunction. They also note that core engineering-design notions (function, structure) remain outside C-K’s perimeter.

What does not exist. No controlled trial, on humans or agents, of the candidate move itself - “those who draw C-K maps and iterate the operators design better than those who do not” - appears anywhere in the record reviewed. No unsourced statistics were encountered that needed excluding; the adoption claims on the theory’s promotional site (ck-theory.org) were counted as adoption facts only.

Net. Coherent formalism, one good experiment on an adjacent claim, school-reported industrial cases of a different (organizational) derivative, an unflattering agent simulation, and a direct published critique on exactly the operationalizability question. That profile is C - and it is C with the doubt pointed at the very thing a skill would need to be.

Why it is / is not a skill here

Verdict: Reject (status: excl). This confirms the wave-3 preliminary verdict (cand / reject / C, 2026-06-11) and its stated ground - operationalizability - now with the named sources behind it. The re-vet found no reason to overturn; it found the critique published in the theory’s own venue.

Distinctness is not the problem - and saying so honestly matters. Both wave-3 external runs found the mechanism distinct, and this run agrees: no shipped skill maintains a dual-space ledger in which propositions carry an explicit undecidability status that exempts them from truth evaluation. Morphological analysis recombines within known parameters; frame-creation reframes one problem; nothing shipped does the C/K bookkeeping. If C-K fails here, it does not fail on overlap.

It fails the artifact commitment (selection-bar commitment 3: artifact, not prose). A skill here must emit a named, structured, reproducible and checkable deliverable. The case that C-K cannot:

1. Its own critical literature says the theory has not been operationalized and supplies no next-action criteria (Choulier, Coatanea and Forest 2011). A procedure with no selection rule between four operators at every node is not a procedure; it is a notation plus taste.
2. The form of C-K that is operational - KCP - is a multi-week facilitated organizational workshop, which is the wrong unit for a single-agent skill (and the registry’s facilitation wall already declines that genre as a build target).
3. The only agent-context study shows what happens when an LLM is scripted into the formalism anyway: predominant C->C churn and declining diversity, with no outcome baseline. The doubt the preliminary entry recorded (“a reproducible agent-executable artifact is unproven”) is now a published observation, not just a doubt.
4. A C-K map has no internal quality check. Two honest runs on the same input produce arbitrarily different trees with no way to say which explored better - which means eval cases for such a skill could not distinguish a good artifact from a confabulated one.

The agent-reachable fragments are already shipped. Each piece of C-K an agent could actually execute is the core of an existing skill: the expansive partition (break the object’s identity by negating a defining attribute) is assumption-reversal’s move; the C->K probe (“what knowledge would make this concept decidable, and what must we learn?”) is what-would-have-to-be-true plus question-burst; the parameterized layout of a design space is morphological-analysis; the reconception of what the object is is frame-creation. A user who wants C-K’s practical effects gets them, with checkable artifacts, from those four.

Why reject rather than fold or recipe. A fold needs one shipped skill that mechanically subsumes the candidate; here the overlap is diffuse across four skills and the whole (the dual-space ledger) is subsumed by none, so there is no honest fold target. A recipe would chain assumption-reversal -> question-burst -> what-would-have-to-be-true -> morphological-analysis; that chain reproduces the useful fragments but not the undecidability bookkeeping - and since there is no evidence the bookkeeping itself adds outcome value, shipping the chain as C-K would ship a brand name around moves the library already has. Reject on the merits; the dossier is the product.

The learning value of this decision. C-K theory is arguably the most influential design theory of the last quarter century, with a research chair, industrial adoption, and a genuinely distinct formal idea. It still does not convert into a skill, because a theory’s job (model the reasoning) and a skill’s job (emit a checkable artifact a cold agent can reproduce) are different jobs. Prestige and distinctness do not waive the artifact commitment - and the gap between “true theory” and “usable procedure” is exactly what its own critics, and the one agent simulation, document.

Lineage and who to read

The theory belongs to Armand Hatchuel and Benoit Weil (Mines ParisTech, now Mines Paris - PSL), with the surrounding school - Pascal Le Masson, Akin Kazakci, Marine Agogue - extending it into teaching, formalization, experiment, and the KCP practice. Hatchuel sketched the two-space idea in the mid-1990s; the ICED 2003 paper introduced it publicly; the 2009 Research in Engineering Design paper is the consolidated formulation. Industrial backing centers on the 2009 Design Theory and Methods for Innovation chair (Dassault Systemes, RATP, Renault, Thales, Vallourec; later SNCF and STMicroelectronics), with KCP deployed as a working method at RATP and Thales. For the textbook treatment, see the Le Masson, Weil and Hatchuel design-theory line at Springer (“Designing in an Innovative Design Regime - Introduction to C-K Design Theory”). For the empirical fragment, read the Agogue fixation experiments; for the limits, read Choulier, Coatanea and Forest; for the agent-era test, read Ma, Brubaker and Goucher-Lambert. “C-K theory” is an academic theory with no trademark; this entry is documented descriptively and is not flagged as branded.

Named sources

• Armand Hatchuel and Benoit Weil, “A New Approach of Innovative Design: An Introduction to C-K Theory,” Proceedings of ICED’03 (Stockholm, 2003). The founding statement of the two spaces and four operators. (Foundational - conceptual)
• Armand Hatchuel and Benoit Weil, “C-K design theory: an advanced formulation,” Research in Engineering Design (2009), doi 10.1007/s00163-008-0043-4. The consolidated formalism; the Magnesium-CO2 Mars engine example. (Foundational - conceptual)
• Armand Hatchuel, Pascal Le Masson and Benoit Weil, “C-K Theory in Practice: Lessons from Industrial Applications,” DS 32: Proceedings of DESIGN 2004 (Dubrovnik, 2004), pp. 245-258. Early industrial cases, reported by the theory’s authors. (Practitioner cases - school-reported)
• Lex Hendriks and Akin Kazakci, “A Formal Account of the Dual Extension of Knowledge and Concept in C-K Design Theory,” Proceedings of DESIGN 2010 (Dubrovnik, 2010). First-order-logic formalization of the dual expansion. (Formalization - conceptual)
• Marine Agogue, Akin Kazakci and colleagues, “The Impact of Type of Examples on Originality: Explaining Fixation and Stimulation Effects,” Journal of Creative Behavior (2014), doi 10.1002/jocb.37. The egg-task experiments: restrictive examples constrain originality, expansive examples stimulate it. Controlled - but it validates C-K’s example classifier, not the practice of C-K mapping. (Experimental - adjacent claim, transferred)
• Marine Agogue and Akin Kazakci, “10 Years of C-K Theory: A Survey on the Academic and Industrial Impacts of a Design Theory,” in An Anthology of Theories and Models of Design (Springer, 2014), pp. 219-236. The impact survey: dissemination, teaching, industrial uptake. (Survey - adoption facts, not outcomes)
• Denis Choulier, Eric Coatanea and Joelle Forest, “CK, an Engineering Design Theory? Contributions, Limits and Proposals,” Proceedings of ICED’11 (2011); companion paper “The Engineering Design CK Theory: Contributions and Limits” (HAL, halshs-00705402). The operationalization critique: not operationalized, no next-action criteria, function and structure outside the perimeter. (Critical literature)
• Kevin Ma, Eric Brubaker and Kosa Goucher-Lambert, “Simulating Design Theory Using LLM Agents: A Case Study of C-K Theory,” Proceedings of the ASME IDETC (2025); journal version “Developing Design Theory Using Large Language Model Agents: A Case Study of C-K Theory,” Journal of Mechanical Design 148(7) (2026). LLM agents adhering to C-K: C->C predominant, diversity trending downward; theory-development simulation with no outcome baseline. (Agent simulation - unflattering for operationalizability)