11.1 The CLIC target card

BRIDGE/SKFM estimates intervention-induced vector fields and uses Lie-bracket geometry to screen causal structure before downstream graph scoring. The corresponding CLIC target is

where \([G]_{\mathcal I}\) is an interventional equivalence class rather than necessarily one privileged graph. The declaration records observed and latent variable types, intervention targets, admissible mechanisms, support and factorization assumptions, and the observer making a closure defect visible.

Operationally, a CLIC run receives a versioned causal sketch, intervention records or estimated intervention operators with uncertainty, and a registered language of experiments and repairs. It returns a status-bearing causal packet: a repair proposal, a request for a discriminating intervention, or an abstention, together with the evidence used to reach that status. It does not take an observational data table and emit a certified ground-truth graph.

In the contract of Chapter 10, \(\mathbb A_{\mathrm{CLIC}}\) presents local intervention-field estimation and \(\mathbb B_{\mathrm{CLIC}}\) presents global causal assembly. Their computational realizations estimate the local operators and revise the factorization, respectively. The observer family records operator, support, and overlap evidence; \(\Theta _{\mathrm{CLIC}}\) is the registered closure or overlap defect; \(\gamma _{\mathrm{causal}}\) tests finite consistency; and \(\mathsf{Adm}_{\mathrm{CLIC}}\) assigns status using evidence withheld from proposal construction. Thus CLIC specializes the generic contract rather than introducing a second causal workflow beside it.

Lie-bracket nonclosure, failure of a latent basis, or an inconsistent overlap supplies a mixed witness. The localization may suggest a missing edge, latent common cause, incorrect latent dimension, nonlinear mechanism, or inadequate intervention. It does not identify one of these explanations by itself.

The CLIC creative mandate freezes the initial causal sketch, repair metalanguage, intervention library, and reference family of supported structures. A result is combinational when it assembles a new causal repair from that registered vocabulary and transformational only when it changes the registered package. Novelty is measured relative to this frozen reference; interventional value is tested separately.

DIAL-SKFM must therefore choose experiments as well as graphs. Its registered control suite \(\mathsf{Ctl}_{\mathrm{CLIC}}\) includes the current and no-change structures, stronger in-sketch factorization search, operator and support diagnostics, alternate latent dimensions, nonlinear or unmodeled intervention explanations, and equal-cost random probes. Observational equivalence calls for abstention or a discriminating intervention. A causal extension proposal is evaluated through held-out interventional prediction, equivalence-class coverage, and those registered controls. When the causal presentation changes, the presentation component is

of a package comparison \(\Upsilon _{\mathrm{CLIC}}:\mathfrak R_{\mathrm{CLIC}} \to \mathfrak R_{\mathrm{CLIC}}^+\). Changes to intervention semantics, observers, or the model doctrine occupy their own components rather than being hidden in \(J_{\mathrm{causal}}\). The package comparison remains a proposal until its added variable, mechanism, factorization, or observer makes interventional consequences that survive independent tests.

Before such a map can support admitted accommodation, a causal integration observer \(\gamma _{\mathrm{causal}}\) must test whether the candidate local repair persists across intervention targets, overlapping charts, and finite intervention sequences. Depending on the experiment, this observer may use exact closure, approximate holonomy, path consistency, or held-out multi-intervention prediction. Without a registered \(\gamma _{\mathrm{causal}}\), the output remains a localized causal-repair candidate rather than a coherent finite causal construction.

The experiment artifacts retain their original DIAL-SKFM identifiers. Within the book, the complete architecture is called CLIC.

In the dossier terminology of Section 10.4.1, the operational object is a causal repair, equivalence class, experiment request, or abstention. The structural delta is \(\Upsilon _{\mathrm{CLIC}}\), with typed causal map \(J_{\mathrm{causal}}\) when its presentation changes. Transport records which variables, mechanisms, and interventional evidence survive that comparison. The status-bearing admission record separately records held-out interventional consequences and unsupported alternatives.

The evidentiary status of the present evidence is narrower than that architectural target. Its repair families, unsupported alternatives, and intervention libraries are registered in advance. The experiments test typed diagnosis, active evidence acquisition, uncertainty propagation, and guarded admission inside that metalanguage; they do not yet show autonomous invention of a causal variable or mechanism type absent from it.

The four studies should therefore be read as one controlled decomposition of the causal problem. DIAL-SKFM–1A admits typed construction only with exact operators and a registered repair vocabulary. DIAL-SKFM–1B estimates both operators from finite noisy samples but abstains under its complete contract. DIAL-SKFM–1C shows that an actively chosen counter-witness is more informative than an equal-cost random probe, yet also abstains. Finally, DIAL-SKFM–1D represents unsupported alternatives explicitly and nearly eliminates unsafe accommodation, but remains too conservative to pass its overall gate. Thus the positive conclusion is not that CLIC has solved causal theory invention. It is that the successive burdens—typing, estimation, experiment selection, alternatives, and admission—can be measured separately rather than hidden inside graph accuracy.