3.1The primary text
The text of Leviathan is that of the 1651 edition in its original spelling, as given by Project Gutenberg eBook 3207 and fetched on 2026-10-05 for the study from which this article is drawn. That edition's marginal headings stand in the copy as lines of their own, and they are quoted as headings where they matter. The copy holds chapters XIII, XVII and XVIII, with the section of chapter XXI headed "In What Cases Subjects Absolved Of Their Obedience To Their Soveraign" and the sections of chapter XXIX on private judgement and erroneous conscience. The article reads chapters XIII, XVII and XXI as its cases and draws two short passages from XVIII and XXIX where they answer a question the three chapters raise. Every quotation is copied from that file character for character, with the 1651 spelling and punctuation kept.
3.2The record
ARCHON publishes small scenes drawn from working records, anonymised and summarised; the raw records are never published. Each scene is read through eight lenses. Each lens poses a question, offers three options with what each gains and risks, and gives a judgment that is dated and marked as a revisable interpretation, with a condition for revisiting it and a list of what is missing. The first lens is marked as inspired by the record; the other seven are marked as hypothetical, which means they suppose what could follow and never report what happened.
The scene read here is titled, in English, "The person who hesitates to leave even after call forwarding". Its setting is dated 2020, and its readings 2026-09-20. It is published in Korean, which is the original, and is read in an English translation that carries the hash of the Korean file. The scenes are synthetic in the sense that matters for evidence: they are modern summaries written for a platform, with roles in place of people, and their readings are the platform's interpretations. They are not archival documents and cannot show what any organisation did. The judgments of the seven hypothetical lenses follow one template across the ten scenes the study reads closely, with only the roles and the missing element changed. That repetition is a property of the records, not a finding about this scene, and Section 7 quotes from those lenses only what differs or what bears directly on visibility.
Record quotations are given in block quotes with a source line: the scene's short title, the lens, and the field.
3.3The ledger
The study, What Withholdeth: Restraint in Ordinary Organisations, keeps a decision ledger over 186 of ARCHON's cases, of which 21 have settings dated 2020. It was made in two steps. Code first collected what the records state directly, such as whether a condition for revisiting the judgment is present. Then one automated reader was run twice, independently, over every record, answering questions registered in advance on 2026-10-05: whether the record shows a restraint, where it comes from, whether the scene's words say why it exists, whether a condition for its removal is named, whether those who wait under it could invoke that condition themselves, and whether they can see what it protects and how far it reaches. A cell counts as settled only when the two codings agree. A strict reading counts only what both found stated in the words; a lenient one also counts what they inferred from the setting.
Agreement is modest on the two cells this article uses most. On whether those who wait can see the restraint's boundary, the two codings agreed on 117 of 186 cases (63%); on whether they could invoke the removal condition themselves, on 115 of 186 (62%). On a sample of 40 cases the agreement on the boundary cell was 60%, κ 0.30. The study's first coding of these fields included the studio's own frame in the coders' material and was superseded; its figures are not used here. Section 7 reports the ledger's reading of the scene in plain words and treats it as a reading. The cells and figures used appear in the Appendix.
3.4How the cases are used
The three chapters and the scene are used as heuristic cases, which use cases to find new variables, paths and questions, and the comparison as a plausibility probe, a preliminary study that asks whether an untested idea is worth fuller testing (George and Bennett 2005). Flyvbjerg answers the misunderstandings that one cannot generalise from a single case and that case studies only generate hypotheses (Flyvbjerg 2006); the claim here is smaller. One book and one synthetic scene can show that a variable is present and describable, not how often it matters.