A CoursIA research series, ICT — “integrated causal trajectories” —, studies how structures organize, transform and repair themselves over time. Its sixth stratum takes discourse as its material, and it has laid down in advance the prediction that will decide whether it fails or holds: a “discourse irreversibility debt”, which the Argumentum corpus will be tasked with measuring.
Illustration: docs/publication/articles/illustrations/6c-deux-cartes-un-miroir.png, Argumentum repository, commit 981e5233 — a montage of the actual renderings of the “False Dilemma” card (row 5.1.2.2.4) and the “Middle Ground” card (row 3.3.1.2.2): the Mirrors link, declared in both directions by the taxonomy, lives in the graph of cross-links — distinct from the attack graph, as the article says.
Discourse as substrate
The series progresses through seven layers, from the self-organized sorting of cells to higher-order degrees of freedom, and its description sums up the sixth in a single sentence: whether a causal structure persists when what circulates is no longer matter or weights, but reasons. The substrate is discourse: belief trajectories over argument graphs — the very object that the philosopher Phan Minh Dung formalized in 1995, arguments some of which attack others.
A prediction made before the measurement
The layer does not begin with a conclusion but with a recorded wager: manipulative rhetorical regimes would leave a measurable trace — an argument structure that holds up while the belief trajectory grows impoverished, like a debt that discourse pays in irreversibility. Stating the prediction before the experiment is what makes it falsifiable: the series description says so itself, this prediction is stated and “not yet measured.” A negative result will still be a result.
The executed seed
The layer is “framed but not delivered,” yet its seed has been executed: three notebooks — documents mixing text and executable code — and their capstone.
Belief trajectories applies the series’ instrumentation to argumentative trajectories under synthetic regimes, a deliberative ring and a populist cascade. Its finding is blunt: Argumentum’s real corpus is not measured there, and the transposition remains open.
Acceptability read as quantified formulas rereads the two classic verdicts on attack graphs — credulous, skeptical — as quantified Boolean formulas, in which “for all” and “there exists” alternate: being credulous means looking for a witness; being skeptical means demanding that everything hold up.
Truth maintenance adds the revision layer: two classic engines, JTMS and ATMS — systems that withdraw the consequences of a repealed premise — ported from the Epita project with a line-by-line proof of fidelity, then a bridge to the semantics of the attack graph.
The capstone consumes all three without copying them. The one that is to measure the prediction on the real corpus, for its part, is still at the design-note stage.
What the game contributes
The real material is the corpus: the taxonomy of fallacies, 1,408 nodes, its AIF ontology — the formal description of arguments and their attacks — and the graph export derived from it. This export exists: a script in the game’s repository produces, from the sources, 145 typed attack edges, the relations between fallacies, the tree, and, on the virtues side, the “good standing” links. It is also honest about its limit: the attack graph is bipartite — each fully modeled fallacy attacks an abstract node type, not another fallacy; the relations between fallacies live in a separate graph, labeled as such.
The project’s role
The game’s owner puts this role in a word: the Argumentum project as steward, together with Epita, of the experimental distillation. Concretely, that means maintaining the reference corpus and its pinned versions — the dated export above names its sources file by file —; supplying the exports; reviewing the notebooks that consume this corpus; and bringing into CoursIA, with the Epita project, the proven mechanisms — the truth maintenance carried by the third notebook, the funnel for descending the taxonomic tree.
The status, stated plainly
The layer is framed, not delivered: the prediction is not measured, the two legs that are to frame it — rhetorical morphogenesis, the grammar of propagation — are not delivered, and the horizon that extends them, toward otherness and collaboration between agents, remains conditional. Nothing here is presented as achieved. What exists: a pre-registered prediction, a seed executed on synthetic regimes, and a real corpus waiting to be measured.
Sources: CoursIA repository, commit 7eea21eb (November 8, 2026): description of the ICT series, the four notebooks linked above, and the tracking: seed CoursIA #7289 (done), capstone on the real corpus CoursIA #23777 (in design), horizon CoursIA #7291. Argumentum repository, commit 32408fdd: graph export script and export folder — counts re-measured on the files committed on November 8, 2026.