open-problem

Status: open (2026-11-09). Area: type theory, logical frameworks · Prerequisites: sophialang core (SPEC-003), annotations and checkers (SPEC-001 §8) · Related: SR-154, SR-53, SPEC-002 Sophialang Principles §3, Computational Trinitarianism §2.3, §3.

Sources: Harper, Honsell, Plotkin, A Framework for Defining Logics, JACM 40(1), 1993 (LF); Pfenning, Schürmann, System Description: Twelf, CADE 1999; Cousineau, Dowek, Embedding Pure Type Systems in the Lambda-Pi-Calculus Modulo, TLCA 2007; Assaf et al., Dedukti: a Logical Framework based on the λΠ-Calculus Modulo Theory, arXiv:2311.07185, 2023. The question comes from Daniel’s message of 2026-11-09 (SR-154) and from the external reviews of that message (Review 2026-11-09 Scope (external)).

Question

sophialang is meant to have a third part: a language in which arbitrary type systems can be stated, beside the programming and query parts (SR-154). Today the specifications have one fixed dependent core (SPEC-003) and “exchangeable checkers” (SPEC-002 §3), but a checker is an external tool with a version. Nothing says how its type system is written down. Can a logical framework inside sophialang state a type system’s judgments and rules as content-addressed definitions? Can checking under that system then be a versioned derivation that Sophia can check, without the core learning each system? And what does the fixed core have to contain for that to work?

Setting

  • Representing a system: a type system is given by its syntax, its judgment forms (such as Γ ⊢ e : A) and its inference rules. In a logical framework these become signatures: constants whose types encode the rules, so that a derivation is a term (LF, judgments as types).
  • Checking: a claim “e : A under system T” is an annotation whose evidence is a derivation in T’s signature. Checking it is type-checking in the framework, which stays one small checker whatever T is.
  • Computation: for systems whose rules compute (definitional equality, as in dependent or cubical types), a framework needs rewriting. λΠ-modulo (Dedukti) adds rewrite rules to LF for this.

What is known

  • LF and Twelf encode many logics and type systems as signatures and check derivations with one checker (Harper, Honsell, Plotkin; Pfenning, Schürmann).
  • λΠ-modulo embeds every functional pure type system (Cousineau, Dowek). Dedukti checks proofs translated from several provers in one framework (Assaf et al.).
  • Not checked here: whether sophialang’s planned core (SPEC-003: Π, universes, Prop, inductive types, quantities) is enough to serve as such a framework, or whether it needs rewrite rules as in λΠ-modulo.
  • Open in the design: SPEC-002 treats checkers as tools, so this question could also be answered “keep them as tools, and record only their rules as documentation”. That would be a choice, not a result.

What would count as an answer

  • A design that states at least two different systems in one framework, for example simply typed λ-calculus and a linear type system, with derivations checked by one framework checker, and with stored ids. It should also say what the core needs.
  • Or an argument that a framework inside sophialang costs more than it gains, compared with external checkers plus documentation, with the reason.