ifc-0313

20.2 What the present evidence supports

The book contains three kinds of named object: component systems, testbed ladders, and proposed integrative architectures. Each targets a characteristic artifact, but they do not carry the same empirical status. CLIC, OPTIC, and RELIC isolate component constructions; IPC, AF, and DIAL-URL supply controlled mathematical construction ladders; SGTE and CTTE organize scientific evidence sequences; AGENTIC specifies their composition; ARTISTIC supplies visual repair and assurance infrastructure; and DILATE proposes the stronger transition to artistic theory extension. The names make this division of labor traceable, provided they are not mistaken for equally mature algorithms.

Named object	Maintained package	Target artifact	Admission evidence
CLIC	Causal mechanisms and intervention operators	Latent alternative, confounder, or missing causal direction	Counter-witness interventions, uncertainty, and held-out mechanism tests
OPTIC	Skill vocabulary and its Lie-bracket organization	New reusable skill, edit, or composition rule	Transfer, cost reduction, closure, and negative controls
RELIC	Sequential model, schema, and decision interface	New state feature, regime, option, or transition schema	Return, sample efficiency, transport, and abstention
AGENTIC	Typed workflow connecting diagnosis, construction, and action	New workflow edge or allocation policy	End-to-end gain beyond isolated components and oracle comparisons

Named object	Maintained package	Target artifact	Admission evidence
IPC, AF, DIAL-URL	Finite presentations, symbolic grammars, and behavioral functors	Latent presentation, executable generator, or typed comparison cell	Hidden-point semantics, exact typing, transport certificates, and false-extension controls
SGTE	Simulator-grounded scientific theory	New mechanism or discriminating experiment	Simulator-relative intervention outcomes under a registered interface and held-out regimes
CTTE	Claims and relations extracted from a scientific corpus	New hypothesis and executable study design	Corpus-relative provenance, consistency, typed feasibility obligations, and an explicit external-admission interface
ARTISTIC	Scene grammar and image integrity constraints	Localized visual repair or enforcement operation	Independent structural observer, contract satisfaction, and preservation of unaffected content
DILATE	Visual techniques and diffusion-time controls	New reusable “paintbrush” acting on generation dynamics	Cross-prompt reuse, irreducibility, persistence, and critical value

. Evidence-status ledger. CLIC, OPTIC, and RELIC have controlled component realizations. AGENTIC has bounded workflow-composition demonstrations, while its full domain benchmarks remain specifications. SGTE and CTTE are registered testbed ladders with admitted intermediate rungs, not single finished systems. IPC, AF, and DIAL-URL reach variable-size presentations, typed generator synthesis, and finite comparison-generator recovery under supplied grammars; they do not invent an unrestricted primitive or verify a new distributive law. ARTISTIC has operational assurance and repair evidence under supplied or estimated declarations. DILATE has local bracket, paintbrush, and routing evidence, while construction and admission of a genuinely missing artistic generator remain open. No name in this ledger denotes a demonstrated general creative intelligence.

These distinctions determine which claims the following summary can support; in particular, ARTISTIC’s admitted repairs provide an assurance substrate for testing visual theory invention, not evidence that such invention has already occurred.

The experiments reported here provide bounded demonstrations at different levels of maturity. Under their registered contracts, the strongest bounded results show that explicit structural declarations can expose failures that a global loss or fluent generator leaves hidden; active counter-witnesses can improve diagnosis; selective systems can abstain when evidence is insufficient; and admitted repairs can improve planning, causal recovery, proposal construction, or visual coherence. The ARTISTIC studies make the point especially visible: a powerful image generator can produce a plausible cricket, baseball, or tennis scene while violating the rules that make it an instance of that game. A compiled scene theory turns those violations into localized obligations rather than an invitation to regenerate the entire image.

These results do not establish general autonomous creativity. Most experiments use controlled worlds, bounded proposal languages, registered observers, or human-curated doctrines. Even the most ambitious studies test whether the architecture can construct a missing reusable structure under a known evaluation contract. That is a necessary calibration for a theory of creative learning, not its completion.