Process models representing knowledge for action: a revised quality framework

Krogstie, Sindre and Jørgensen's 2006 revision of the SEQUAL framework — the canonical semiotic framework for conceptual-model quality. Reframes SEQUAL for active process models (models that actively change the reality they describe) rather than purely descriptive conceptual models.

SEQUAL — original framework recap

Lindland, Sindre, Sølvberg (Lindland et al., IEEE Software 1994) define quality as set correspondences over:

G — modelling goals (what the model should achieve organisationally).
L — language extension (everything expressible in the chosen modelling language).
D — modelling domain (everything that can be said about the situation at hand).
M — externalised model (the actual model artefact).
K_s, K_m — explicit knowledge of social actors / modellers.
I — social actor interpretation of M.
T — technical actor interpretation of M (e.g., what software does with the model).

Quality types map to set correspondences (per Stamper's semiotic ladder):

Type	Correspondence	Goal
Physical	externalisation + internalisation of M	model is available/accessible
Empirical	readability of M	layout, comprehensibility heuristics
Syntactic	M ⊆ L	syntactic correctness
Semantic	M vs D	validity + completeness
Perceived semantic	I vs K_s	perceived validity + perceived completeness
Pragmatic	M vs (I, T)	comprehension by interpreters
Social	agreement among I's	feasible agreement
Organisational	M fulfilling G	fulfilment of modelling goals

Validity and completeness are typically infeasible in absolute terms — the framework introduces feasibility so that modelling can terminate when further effort yields less benefit than acceptance.

The three shortcomings (motivating the revision)

Static view of domain D: original framework treats D as a given fixed reality. Process models — especially active/interactive models — change D through enactment. The quality of an active model depends on the changes it induces, not only on its fit to a pre-existing domain.
Narrow pragmatic quality: defined per Morris as comprehension (model→audience). Modern semiotics emphasises impact on interpreter — what they do with their understanding. This is especially relevant for active models where understanding directly leads to action.
Knowledge-as-object framing: original definitions discuss "externalised knowledge" — models as knowledge. Polanyi (tacit knowledge) and Walsham (knowledge management critique) argue knowledge is in the human mind; models can contribute to knowledge but are not knowledge themselves. Hence the language is revised: models do not externalise knowledge; they support knowledge construction.

Active models and the articulation/activation dynamic

Active model = model that directly influences the reality it reflects. Three activation modes:

Automated: software component interprets and acts (e.g., workflow engine routes work via M).
Manual: human reads M and acts accordingly.
Interactive: mixed — formal parts interpreted automatically, ambiguous parts left to humans.

Two complementary processes (Figure 2):

Articulation (D → M): domain features captured in model. Manual articulation by humans; automatic articulation by sensors. Increases semantic quality of M.
Activation (M → D): model-guided actions transform domain. Manual activation by human reading model; automatic activation by software interpreting M; interactive activation mixed. Changes the domain D.

Plus model evolution (M → M), language evolution (L), knowledge change (K), interpretation change (I), and meta-modelling (K → L).

The interplay produces co-evolution: the model and the domain shape each other over time. An interactive model that does not change "will not be able to reflect aspects of the reality that changes, nor can it reflect evolution of a human actor's understanding ... it will deteriorate."

The revised framework (Figure 3)

Adds two new sets and reorganises semantic quality:

D^O — optimal domain (the situation the organisation would (or should) want).
K^N — knowledge need (knowledge required for the organisation to perform its tasks).

Three types of semantic quality:

Type	Correspondence	Use
Descriptive semantic quality	M vs D	validity (M ⊆ D, no false statements); completeness (D ⊆ M, no missing) — for as-is models.
Prescriptive semantic quality	M vs D^O	validity (M ⊆ D^O); completeness (D^O ⊆ M) — for to-be / target models.
Semantic quality (formerly "perceived semantic")	K vs M	model fit to actual stakeholder knowledge

Three new ovals capture model dynamics:

Learning (of domain) — pragmatic quality redefined: M → ΔK ∩ K^N (added knowledge that fills the knowledge gap).
Modelling (of domain) — articulation activity, ΔK^m influences M.
Action (in domain) — activation activity, M-driven changes ΔD intersected with the gap to D^O.

Pragmatic quality is now a more demanding criterion: comprehension is necessary but not sufficient — the model must enable learning (knowledge increase) and action (domain change toward D^O).

In the revised version, physical, empirical and syntactic levels are intentionally de-emphasised — they are not the most problematic levels for active-model evaluation (modern modelling tools largely solve them).

Maritime industry case (Section "Applicability")

Globally distributed certification offices each developed local procedures. A 2000 harmonisation project used active modelling:

Pre-project: workshops elicited improvement areas → modelled "ideal" work processes (this is D^O).
Harmonisation project: developed harmonised processes (compromise between D as-is and D^O ideal, accounting for local autonomy).
IS development: the harmonised models drove construction of a global software system to support them (activation: model → domain change).

Sedera/Rosemann/Doebli's success-model dimensions (use, satisfaction, impact, model quality) were partially supported by the original SEQUAL framework but the analysis "needed a more detailed model" — supplied by the revised framework's distinction between current K and knowledge-need K^N, and between as-is D and to-be D^O.

Limitations the authors acknowledge

Qualitative, not quantitative — the framework structures evaluation but does not yield a numeric score. "The revision does not fix the framework's disability to facilitate precise, quantitative evaluations of models — it is still focussed on qualitative evaluations."
Lower three levels (physical, empirical, syntactic) are demoted in the revised exposition because the authors regard them as less problematic for active models — they remain part of SEQUAL but are not the focus.
Operationalisation gap: only syntactic quality has any hope of objective measurement; semantic and pragmatic quality involve unknowable sets (full domain D, full knowledge K).
Tool support (e.g., METIS/Workware combination) is suggested as a direction to embed quality metrics into modelling environments, but not delivered.

Why this matters for BPM

The theoretical scaffolding for the syntactic/semantic/pragmatic tripartite that appears in sources/2018-dumas-fundamentals-of-bpm §5.4 — Dumas's three-quality-aspects formulation is the original 1994 SEQUAL distilled to three actionable buckets.
Foundation for distinguishing as-is vs to-be model quality — descriptive vs prescriptive semantic quality maps cleanly onto Dumas's discovery (as-is) vs redesign (to-be) phases.
Bridge to APM and active/agentic systems — the activation concept anticipates agentic-BPM execution where models are interpreted and enacted by software. The articulation/activation interplay is essentially what observability + feedback would call the read/write loop on a process model.

Connections

Concepts: concepts/process-model-quality · concepts/sequal-framework (new — extracted from this source) Sources: sources/2018-dumas-fundamentals-of-bpm (cites the tripartite Lindland-derived framing) · sources/2010-mendling-reijers-vanderaalst-7pmg (positions itself against SEQUAL's abstraction) Authors / hub entities: entities/john-krogstie (new) · entities/guttorm-sindre (new) · entities/havard-jorgensen (new) Adjacent: concepts/agentic-bpm (active models prefigure agentic execution) · concepts/agent-process-observability (articulation/activation = read/write loop)

Open follow-ups

Original Lindland Sindre Sølvberg 1994 paper — would supply the original SEQUAL formalisation (vs this revision); status: referenced-not-ingested.
Krogstie & Jørgensen 2002 Quality of interactive models — the intermediate revision — referenced-not-ingested.
Moody et al. 2002a/b/2003 — empirical evaluations of SEQUAL — referenced-not-ingested.
Krogstie's later work extending SEQUAL further (2012 Model-Based Development and Evolution of Information Systems).