Methodology: The Chapter
12 entriesThe methodology chapter as continuous prose. Read these in order; they are the chapter itself.
- Warranting Without Vouching: The Problem This Chapter Solves Published on: The review closed by asking whether the missing composition can be built. An absent thing is simply a thing to build; what the review makes hard is the warranting. Its finding was that every surveyed field ends by handing the judgement to a human who can ask, so a discipline shown to work because its designer judged it works has placed a human at the very boundary the thesis says must be mechanised. The chapter's task: establish that a discipline for refusing unwarranted claims is itself warranted, without standing where the review said no-one is standing. Design science follows from the review rather than being chosen ahead of it; Peffers supplies the entry point and the artefact vocabulary.
- Typing the Artefact: What Is Being Built, Read Off the Review Published on: The review did not merely find an absence; it stated the shape of what is absent, in the same words, three times over. That statement is a requirements list derived by survey, and this section reads it off rather than proposing one. Its five parts are typed into design science's own categories of construct, model, method, and instantiation, and each is stated precisely enough to be argued with and to be built, because a basis model described as capturing relevant provenance factors is both weak prose and unimplementable. The section stops short of interfaces and type signatures, which belong to a later chapter and to a different kind of document.
- How the Artefact Is Built: The Method Governing Design Decisions Published on: Design science's central activity is making the artefact, and a methodology that describes only how the result is judged has described the smaller half of its subject. This section states the method the design work follows, and derives it from a property of the artefact: its parts are rigid against each other, so they must be fixed in dependency order, each minimal, each settled before the next is built on it. Five commitments follow: draft the smallest thing that could work; publish requirements before structure; name every fork and settle it on evidence rather than preference; read a neighbouring formalism only against a stated lack; and record what the reading returned, including when it contradicts the design. Two are inherited rather than invented. The section closes on what a design method can and cannot do: it makes decisions checkable, not correct.
- Demonstration: Running the Artefact on Instances of the Problem Published on: Running the artefact on instances of the problem, so evaluation has observations to measure. The library is run rather than the discipline applied, because the claim under test is that admissibility decides with no person present. It runs on the pre-registered scenarios, a case set it cannot revise. Four things are recorded per case, keeping what the artefact did separate from what the world did, so a reader can recompute a verdict without repeating the run. Cases the model cannot express are recorded as coverage findings and reported, not dropped.
- The Instantiation: Faithfulness, Interface, and an Unsolved Problem Published on: The typing section said what the instantiation is for and not how it is built or judged. Faithfulness is tested by recoverability: for each settled design decision there must be an observation distinguishing an implementation that honours it from one that does not, and a decision without such an observation is recorded as untestable. The library holds the model and the methods and must not accept a caller-supplied verdict or offer a bypass, since a library that did would enforce nothing. Revocation checking and transport stay outside. And it states an open problem: the model assumes records arrive formed, which is free for a demonstration and not for a tool, because nothing in the design says how a plant's existing signals become the fields a record requires.
- Implementation: Build Order, What Establishes It Works, and What the Setting Imposes Published on: How the library is produced and what would establish that it works, given that the artefact is to be usable by an operational-technology engineer rather than only runnable against the scenarios. One constraint is inherited from the review, that the library must decide without a person; the rest comes from the artefact's structure and its setting, and says so. Record construction is built first, inverting the expected order, because building the deciding parts first would produce a library that decides correctly about records nothing in a plant can produce. Three separate claims establish it works: faithfulness, correctness, and usability, the last being design work not yet done rather than testing not yet performed. Three constraints come from the setting: the record must arrive rather than be fetched, a refusal is itself an action, and the library sits alongside systems that will not be replaced.
- The Circularity Problem, Named Before Its Answer Published on: A discipline evaluated against cases chosen by the person who wrote its rules will pass, and the pass will mean nothing. That is the central methodological threat, and this section states it in the open before any of the safeguards that answer it, because a chapter presenting its defences without first naming what they defend against is asking to be taken on trust. Two commitments follow: scenarios fixed before the rules harden, and criteria inherited from outside the design. The first turns out to be the paradigm's own requirement rather than a scruple this work volunteered, which means a reader can hold the work to it without taking the author's word for why it matters.
- Criteria the Design Did Not Author Published on: The circularity problem needed criteria the designer did not write, and the review left them behind without meaning to. Every field it surveyed stated, in its own words and its own specification, the point at which it stops: the provenance standard locates assessment outside itself, the attestation framework declines to validate what it carries, the uncertainty survey calls its own estimates insufficient for safe decision-making. Quoted rather than paraphrased, these become a test authored by people who never heard of this work, and the test is falsifiable: if the basis model reduces to a provenance record that hands the judgement on, the provenance standard refutes it. The section closes on its own boundary, since these criteria test whether the missing composition was built and say nothing about how well it discriminates, which is a different question deferred to the evaluation.
- Success and Failure, Declared in Advance Published on: A discipline that rejects everything is trivially safe and practically worthless, and a chapter that only defines the failure of admitting False Determinism has written a test it cannot fail, because refusing everything passes it. So failure is declared here with two faces given equal weight: the false positive that lets an unwarranted claim through, and the false negative that rejects so much legitimate practice as to be unusable. Over-conservatism is named as a failure mode rather than a safe harbour, and both faces are fixed before the artefact exists to be flattered by them. The section declares what would count as failing; it does not measure it, because measurement is the discrimination question, and that is the evaluation's instrument, deferred and kept separate here on purpose.
- The Debt the Review Incurred Published on: The review conceded in public that several fields it did not survey may already hold semantics this work would otherwise reinvent, and it named the reading of them as owed before the basis model is fixed. A debt named and not scheduled is a debt disowned, so this section schedules it: the six fields, the sequence that reads them in the right order, and the trigger that fires the reading. The sequence is deliberate and runs draft the smallest model, identify its load-bearing semantics, pressure-test only those, then stabilise, because reading six literatures before knowing which semantics matter wastes months and freezing the model before the check risks reinventing something older than the problem. The section schedules the reading; it does not begin it.
- The Discrimination Problem, and a Route Through It Published on: The question the work is falsifiable on is whether the discipline rejects False Determinism without rejecting so much legitimate practice as to be unusable, and that is a claim about discrimination, which is only meaningful against a ground truth. The obvious sources of that ground truth fail: pre-registration fixes timing but not authorship, expert labels relocate the human vouch the work exists to remove, and incident reports document only what went wrong. The route through is a field the review already surveyed. Selective prediction evaluates systems whose job is to decline without labelling any case as one that should have been refused, using the coverage and risk relationship, and the ground truth it needs is outcome rather than opinion. One residual is real and stated: the scenario distribution stays the designer's even when the verdicts do not. The section records what nearly happened instead, which was calling the problem unsolvable and selling the gap as the finding.
- What Is Settled, What Is Open, and the Limitations That Remain Published on: The chapter set out to establish that a discipline for refusing unwarranted claims can itself be warranted without standing where the review said no-one stands, and this closing section takes stock of how far it got. What is firm is firm because the review forced it, not because it was chosen: the commission, the trap, the typing, the two-faced failure, and the criteria authored elsewhere. What is open is the discrimination problem, which has a route rather than a settled answer and is marked as such. The limitations are stated plainly rather than left for a reader to find: open-access sourcing, a single researcher, and Triton bounded to illustration. The chapter ends by handing the design chapter a specification it did not have to invent.