An Addendum to Formal Methods and Architecture, Part 4
Part 4 asked when a formally established claim survives refinement, realization, deployment, and situated use.
While writing that post, I recognized that a seemingly small notation problem exposed a much larger issue.
The draft introduced an abstract specification
, a more concrete specification
, and a refinement relation written as
![]()
The notation was locally defined. It was mathematically permissible. In a behavior-oriented reading, it indicated that the concrete specification admitted no mapped behavior outside those permitted by the abstract specification
![]()
But I had already assigned meaning within the larger architecture framework developed in this series to
as a causal precedence between events. This is a legitimate usage of the symbol within causal set theory upon which this architectural framework is based. The problem is that now the same symbol carried two different relational meanings. The subscript
distinguished them formally, but it did not prevent the established causal interpretation from being carried into the new expression.
The problem was not that the refinement relation was undefined.
The problem was that its local definition did not erase the surrounding semantic history of the notation.
The expression was mathematically permissible yet representationally inadequate.
This is a small example of how formal work can go wrong without containing a false theorem, an invalid derivation, or a defective tool. A notation can be legal inside one formal fragment while becoming ambiguous inside the larger body of reasoning in which that fragment is used.
Going in a different direction, it also suggests something important about digital threads and digital twins.
Had every version of the draft been stored, linked, timestamped, and made traceable to the earlier posts, the digital record could have been complete. The symbol, equation, file history, and provenance could all have survived intact. The thread would have been digitally continuous.
The meaning would not.
What the Digital Thread Already Does Well
The concept of digital thread addresses a very real and difficult systems problem.
Engineering information is distributed across requirements tools, architecture models, simulations, product-lifecycle systems, source repositories, manufacturing records, test environments, configuration systems, operational data, and sustainment organizations. The continuity among these sources is often weak. Formats differ. Identifiers change. Ownership changes. Lifecycle stages become organizational silos.
The U.S. Department of Defense describes digital engineering as an integrated approach using authoritative sources of system data and models as a continuum across disciplines and lifecycle activities [Off18]. In manufacturing, Hedberg et al. describe an integrated lifecycle information framework intended to connect heterogeneous systems and support decision making [HFHC17]. Monnier et al. likewise treat the digital thread as information flow across product-lifecycle stages and identify interoperability, reuse, and the integration of heterogeneous standards as central problems [MSF22].
These are not minor improvements. A system whose engineering information cannot be found, related, versioned, or exchanged has little chance of preserving architectural reasoning across its lifecycle.
The point of this addendum is therefore not that digital thread work is misguided. It is that information continuity is necessary without being sufficient.
A digital thread may establish that two artifacts are connected. It may show that one was derived from another, that a requirement is allocated to a model element, that a model generated code, that a test refers to a requirement, or that operational data came from a particular configured asset.
None of those links, by itself, establishes that the meaning required by an architectural claim survived the connection.
The Nodes Are Local for a Reason
The heterogeneity of the nodes in a digital thread should not be treated as a defect to be removed.
A requirements model, control model, software design, hazard analysis, manufacturing plan, test record, and operational display are not incomplete versions of one universal representation. Each is constructed for a local purpose. Each makes some distinctions visible, supports some forms of inference, and omits detail that would obstruct the work performed there.
The nodes are therefore localized representations, appropriately adapted to their disciplinary, lifecycle, and decision contexts. The semantic difficulty appears when information must move between them.
Consider the term behavior. In robotics, behavior may refer to an embodied, closed-loop interaction involving sensing, dynamics, actuation, and an environment. In software engineering, behavior may refer to state-transition logic, interface responses, or externally observable execution.
Neither usage is defective. Both can be precise within their local theories. But they are not interchangeable.
A transport mechanism that maps one element named behavior to another element named behavior may preserve the label while collapsing the distinction between physical interaction and software execution. The source and destination nodes can both remain locally adequate while the transition between them is semantically inadequate.
The central difficulty of a digital thread is therefore not the existence or heterogeneity of its representations. It is the semantic adequacy of the transport and translation mechanisms between them.
Connected Is Not Preserved
Consider an apparently complete engineering chain
![]()
- Every artifact can be present
- Every arrow can be implemented as a navigable link
- The configuration history can be complete
- The source and destination formats can be machine-readable.
- The entire chain can be visible through an authoritative digital environment
The claim can still be lost.
An example in the Part 4 article used the term safe. Across the chain, that term might refer to
- Membership in a state set in a formal model
- Completion of a software transition
- Assignment of a Boolean variable
- Transmission of an actuator command
- Measured movement of the actuator
- Attainment of the physical condition that stakeholders regard as safe
A thread can link all six representations without establishing that the six propositions are equivalent.
The same failure can occur with time. A requirement may demand a physical response within 100 milliseconds.
- The formal model may measure from a hazard-detection event
- The software test may measure from task release
- The operational monitor may measure from a logged message
- Each artifact can retain the number 100 and the unit ms while silently changing the event from which the interval begins and the condition at which it ends
Traceability remains intact at the level of identifiers. The temporal claim does not.
This yields a distinction that ordinary connectivity cannot resolve
Digital continuity is not semantic continuity, and semantic continuity is not assurance continuity.
- Digital continuity means that the relevant representations, data, versions, and relationships remain accessible across the lifecycle.
- Semantic continuity means that the referents, distinctions, units, interpretations, and relevant inferential structure remain sufficiently preserved across their transformations.
- Assurance continuity means that the claims, assumptions, evidence, scope, and remaining obligations needed for justified confidence also survive.
An organization may possess the first without the second or third.
The Hard Problem Is Semantic Transport
The word transport can conceal several different achievements. A successful digital-thread connection
may move the bits and parse the receiving format while failing to transport the meaning or the warranted inference.
A useful progression is
![]()
Success at one stage does not entail success at the next. A valid exchange file establishes that the receiving tool can read the artifact. It does not establish that the receiving discipline interprets the artifact as the originating discipline did or intended.
The problem is consequently located in the edges, but not because the nodes have been neglected.
The edge is where two locally adequate semantic contexts meet. The edges are also where local adequate semantic contexts are expanded to include both nodes semantic contexts.
Let a representational chain be
![]()
The existence of the edge
tells us that some relationship or transformation connects
to
. It does not yet tell us what the transformation preserves.
For assurance purposes, an edge may need to carry at least
![]()
where
states how the two representations or realized entities correspond
states how the relevant claim is translated into the vocabulary and semantics of the downstream representation
identifies the class of claims that the transformation has been shown to preserve
records the evidence supporting preservation
records the assumptions, boundary, and applicability conditions under which that evidence is valid
records inherited or newly created obligations that remain unresolved after the transformation
This tuple is schematic. A practical digital environment might distribute its contents across models, assurance cases, configuration records, test evidence, calibration data, interface contracts, and governance decisions. The architectural requirement is not that every program adopt one data structure. The requirement is that the information not disappear merely because an artifact link exists.
This helps us understand the emphasis on ontology in digital-thread work. Ontologies can supply persistent identities, semantic types, units, namespaces, subsumption relations, and explicit correspondence rules. They provide a way to say that two artifacts refer to the same entity, that one concept is narrower than another, or that two similarly named concepts should remain distinct.
That concentration is revealing. If the problem were only bit movement or schema conversion, ontology would be unnecessary. Its presence recognizes that format mismatch is the surface expression of a deeper semantic impedance.
That is an important response.
But ontology does not make the impedance disappear. A single canonical concept can produce semantic compression if it forces genuinely different local concepts into one category. The robotics and software-engineering concepts of behavior may require a qualified, partial mapping rather than a declaration that they are simply the same concept.
Nor does an ontological correspondence alone establish preservation of inference. Schematically, if a source model
satisfies a claim
, the stronger transport obligation has the form
![]()
where
transports or transforms the model and
translates the claim into the target semantic environment. Ontology can help construct
and \tau_i
R_{i+1}
R_i
\preceq$ survived perfectly. Its semantic projection changed. The failure was not missing data. It was an unmarked transition between meanings.
Closing
Digital threads promise continuity across the lifecycle. Digital twins promise a continuing relationship between representations and the entities or processes they represent.
Both promises are valuable. Part 4 adds a condition to them.
A connected sequence of locally appropriate artifacts is not a chain of preservation. A synchronized representation is not yet an applicable assurance claim. The meaning, assumptions, distinctions, evidence, and residual obligations must also travel, and every transformation at which they can change must be made visible.
The notation collision showed the problem in miniature. Everything needed to preserve the expression was present. What failed to survive was the thing the expression was supposed to mean.
The digital thread is not the architecture. It is infrastructure through which localized architectural representations move and are translated.
The architectural question is not whether every node speaks the same language. It is whether the transport between them preserved what needed to survive.
References
[Dig20] Digital Twin Consortium. Digital Twin Consortium Defines Digital Twin. Digital Twin Consortium, December 2020. Available at https://www.digitaltwinconsortium.org/ 2020/12/digital-twin-consortium-defines-digital-twin/.
[HFHC17] Thomas Hedberg, Jr., Allison Barnard Feeney, Moneer Helu, and Jaime A. Camelio. Toward a Lifecycle Information Framework and Technology in Manufacturing. Journal of Computing and Information Science in Engineering, 17(2):021010, 2017.
[Int21] International Organization for Standardization. ISO 23247-1:2021: Automation systems and integration — digital twin framework for manufacturing — part 1: Overview and general principles. Technical report, International Organization for Standardization, Geneva, Switzerland, 2021.
[LSM13] Timothy Lebo, Satya Sahoo, and Deborah McGuinness. PROV-O: The PROV Ontology. W3C recommendation, World Wide Web Consortium, April 2013. Available at https://www.w3.org/TR/2013/REC-prov-o-20130430/.
[MSF22] Laetitia V. Monnier, Guodong Shao, and Sebti Foufou. A Methodology for Digital Twins of Product Lifecycle Supported by Digital Thread. In Proceedings of the ASME 2022 International Mechanical Engineering Congress and Exposition, Columbus, Ohio, 2022. American Society of Mechanical Engineers. Paper IMECE2022-95182. Available at https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=934911.
[Off18] Office of the Deputy Assistant Secretary of Defense for Systems Engineering. Department of defense digital engineering strategy. Technical report, U.S. Department of Defense, Washington, DC, June 2018. Available at https://ac.cto.mil/wp-content/uploads/2019/06/2018-Digital-Engineering-Strategy_Approved_PrintVersion.pdf.
Leave a Reply