Groupoid Categories with Families (GCwF)

• Groupoid Categories with Families (GCwF) are frameworks that generalize traditional CwFs by replacing set-valued type families with groupoid-valued ones, accommodating richer homotopical interpretations.
• The construction weakens only the type level by enforcing groupoid truncation and introducing coherence equations, thus allowing natural univalent interpretations without losing strict substitution.
• Formalized in Cubical Agda, GCwFs establish a coherence theorem that shows the intrinsic syntax ultimately collapses to set-level equality, bridging higher structure with conventional type theory.

Searching arXiv for recent and foundational papers on Groupoid Categories with Families and closely related categorical semantics. A Groupoid Category with Families (GCwF) is a variant of the category-with-families semantics for dependent type theory in which the usual set-valued family of types is replaced by a groupoid-valued one, while substitutions and terms remain set-truncated. It was introduced to address a specific obstruction in Homotopy Type Theory (HoTT): ordinary internalized CwFs force types to be set-truncated, which blocks direct interpretation into natural univalent models such as the standard set model. The defining idea is to weaken the standard CwF notion only at the level of types, adding explicit coherence equations so that reindexing of types is functorial up to specified paths rather than by strict equality, while preserving strict 1-categorical substitution structure and set-valued terms (Altenkirch et al., 18 Sep 2025).

1. Motivation and conceptual placement

A traditional Category with Families is presented with a category of contexts and substitutions, a set of types $\mathsf{Ty}(\Gamma)$ in each context, a set of terms $\mathsf{Tm}(\Gamma,A)$ for each type, and strict substitution and comprehension equations. When this is internalized in HoTT using quotient inductive-inductive types, the syntax forces types to be set-truncated. In the set-truncated syntax discussed in the source paper, this is imposed by a constructor

$\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$

This preserves ordinary intrinsic syntax and associated equality phenomena, but it is semantically too restrictive in univalent settings, because natural semantic collections of types are often not sets but only groupoids (Altenkirch et al., 18 Sep 2025).

The core obstruction is that a semantic interpretation of

$\mathsf{Ty}\,\Gamma$

typically looks like $\llbracket \Gamma \rrbracket \to \mathsf{Type}$, and $\mathsf{Type}$ is not a set in HoTT. The same difficulty appears for the intended standard set model when presented univalently: even if its semantic fibers are sets, the ambient collection of such sets is naturally a groupoid rather than a set. For this reason, traditional CwFs “exclude models based on univalent categories, such as the standard set model,” or force alternative encodings such as inductive-recursive universes (Altenkirch et al., 18 Sep 2025).

The opposite extreme is also unsatisfactory. If truncation is dropped entirely, types in syntax are no longer a set, and higher path structure becomes uncontrolled. The source paper proves this already for the “wild syntax,” where $\mathsf{Ty}$ is not a set via an interpretation into the circle $\mathbb S^1$. The GCwF construction is therefore positioned between two extremes: fully strict set-truncated syntax and completely untruncated higher syntax with all coherences. Its governing compromise is that substitutions and terms remain sets, while types are only truncated at the groupoid level (Altenkirch et al., 18 Sep 2025).

This placement aligns GCwFs with the internal shape of a 1-category in HoTT: objects need not form a set, but hom-types do. A plausible implication is that GCwFs isolate the minimum weakening needed to support univalent semantics without giving up ordinary substitution calculus.

2. Formal structure

A GCwF starts with a 1-category of contexts and substitutions. Concretely, it has a type of contexts

$\mathsf{Con} : \mathsf{Type},$

and for $\Gamma,\Delta : \mathsf{Con}$, a set of substitutions

$\mathsf{Tm}(\Gamma,A)$0

with composition and identities

$\mathsf{Tm}(\Gamma,A)$1

satisfying the ordinary category laws

$\mathsf{Tm}(\Gamma,A)$2

$\mathsf{Tm}(\Gamma,A)$3

The point is that $\mathsf{Tm}(\Gamma,A)$4 is set-valued, so the underlying context/substitution structure is a genuine internal 1-category (Altenkirch et al., 18 Sep 2025).

Over this 1-category, types form a groupoid-valued family

$\mathsf{Tm}(\Gamma,A)$5

Thus each $\mathsf{Tm}(\Gamma,A)$6 is 1-truncated rather than 0-truncated. The exact truncation constructor used in the syntax is

$\mathsf{Tm}(\Gamma,A)$7

Reindexing of types is contravariant,

$\mathsf{Tm}(\Gamma,A)$8

written $\mathsf{Tm}(\Gamma,A)$9, but unlike in an ordinary strict CwF, functoriality is not definitional. Instead, it is equipped with specified equalities

$\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$0

together with the standard 2-dimensional coherence laws $\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$1, $\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$2, and $\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$3 making $\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$4 into a 2-presheaf over the 1-category of contexts (Altenkirch et al., 18 Sep 2025).

Terms remain set-valued: $\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$5 They reindex by

$\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$6

with dependent laws

$\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$7

Because $\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$8 is set-valued, no higher coherence beyond propositional uniqueness is required (Altenkirch et al., 18 Sep 2025).

The comprehension structure is given in the Ehrhard presentation. For each context and type there are

$\mathsf{isSetTy} : (e,e' : A_0 = A_1)\to e = e'.$9

$\mathsf{Ty}\,\Gamma$0

These satisfy the usual comprehension equations, but because type equalities are only groupoid-truncated, several equations are heterogeneous equalities over equalities of types. The key ones include

$\mathsf{Ty}\,\Gamma$1

$\mathsf{Ty}\,\Gamma$2

$\mathsf{Ty}\,\Gamma$3

The GCwF is therefore a CwF-like structure in which weakening occurs only on the type side, not on substitutions (Altenkirch et al., 18 Sep 2025).

3. Intrinsic syntax and the initial GCwF

The principal syntactic object studied for GCwFs is the groupoid-syntax, defined as the initial GCwF for a type theory with a base family of sets and $\mathsf{Ty}\,\Gamma$4-types. The object language contains a base family specified semantically by

$\mathsf{Ty}\,\Gamma$5

together with a universe-like code type $\mathsf{Ty}\,\Gamma$6, a decoding operation $\mathsf{Ty}\,\Gamma$7, constants

$\mathsf{Ty}\,\Gamma$8

and dependent function types $\mathsf{Ty}\,\Gamma$9 (Altenkirch et al., 18 Sep 2025).

The corresponding syntax provides

$\llbracket \Gamma \rrbracket \to \mathsf{Type}$0

$\llbracket \Gamma \rrbracket \to \mathsf{Type}$1

together with

$\llbracket \Gamma \rrbracket \to \mathsf{Type}$2

and the equations

$\llbracket \Gamma \rrbracket \to \mathsf{Type}$3

The groupoid-syntax is a higher inductive-inductive definition with four mutually dependent sorts

$\llbracket \Gamma \rrbracket \to \mathsf{Type}$4

to which are added:

• set truncation for substitutions and terms,
• groupoid truncation for types,
• specific 2-dimensional coherence constructors for $\llbracket \Gamma \rrbracket \to \mathsf{Type}$5, $\llbracket \Gamma \rrbracket \to \mathsf{Type}$6, and $\llbracket \Gamma \rrbracket \to \mathsf{Type}$7 commuting with $\llbracket \Gamma \rrbracket \to \mathsf{Type}$8 and $\llbracket \Gamma \rrbracket \to \mathsf{Type}$9.

This HIIT is the initial GCwF with $\mathsf{Type}$0-types and a base family. The source describes it as the free GCwF generated by $\mathsf{Type}$1, $\mathsf{Type}$2, and the given type formers. Its significance lies in preserving intrinsic syntax—there are still only well-typed terms and substitutions, not raw syntax—while retaining one dimension of path structure at the level of types (Altenkirch et al., 18 Sep 2025).

The paper also establishes a comparison with ordinary CwF syntax: the set-syntax is isomorphic to the groupoid-syntax. One direction forgets the extra 2-cells and interprets them via set-truncation in the target; the converse uses the main theorem that $\mathsf{Type}$3 is a set. Thus GCwF syntax strictly generalizes ordinary CwF syntax, but in the initial case considered there is no loss of the usual intrinsic presentation (Altenkirch et al., 18 Sep 2025).

4. The coherence theorem: “the groupoid-syntax is a set”

The central result is the theorem

$\mathsf{Type}$4

This means that although the initial syntax was constructed with types only groupoid-truncated, the freely generated type theory has no nontrivial 1-dimensional path structure remaining. The slogan of the paper—“the groupoid-syntax of type theory is a set”—therefore expresses a coherence theorem: the syntax is weakened just enough to interpret richer models, but the free syntax collapses back to set-level equality (Altenkirch et al., 18 Sep 2025).

The proof is by $\mathsf{Type}$5-normalisation, where $\mathsf{Type}$6 denotes elimination of explicit instantiation from types using substitution laws. Normal forms are defined by mutual induction as

$\mathsf{Type}$7

with constructors

$\mathsf{Type}$8

$\mathsf{Type}$9

Normal types are therefore built from $\mathsf{Ty}$0, $\mathsf{Ty}$1 applied to arbitrary terms of $\mathsf{Ty}$2, and $\mathsf{Ty}$3, but not by explicit substitution (Altenkirch et al., 18 Sep 2025).

A key lemma proves that $\mathsf{Ty}$4 is a set. This is shown by an encode-decode argument using a recursively defined relation

$\mathsf{Ty}$5

with clauses

$\mathsf{Ty}$6

$\mathsf{Ty}$7

$\mathsf{Ty}$8

and $\mathsf{Ty}$9 in mixed cases. Since $\mathbb S^1$0 is propositional and equality in $\mathbb S^1$1 retracts onto it, normal forms are set-truncated (Altenkirch et al., 18 Sep 2025).

The next step defines substitution on normal forms

$\mathbb S^1$2

together with quotation naturality

$\mathbb S^1$3

The added coherence constructors $\mathbb S^1$4, $\mathbb S^1$5, $\mathbb S^1$6, $\mathbb S^1$7, $\mathbb S^1$8, and $\mathbb S^1$9 are crucial here, because they make quotation a 2-natural transformation from normal forms into syntax (Altenkirch et al., 18 Sep 2025).

Finally, the proof defines

$\mathsf{Con} : \mathsf{Type},$0

with the salient clauses

$\mathsf{Con} : \mathsf{Type},$1

$\mathsf{Con} : \mathsf{Type},$2

Because quotation is a retraction and $\mathsf{Con} : \mathsf{Type},$3 is a set, $\mathsf{Con} : \mathsf{Type},$4 is a set. The initial GCwF is therefore set-truncated in the expected sense, and the ordinary set-syntax is recovered (Altenkirch et al., 18 Sep 2025).

5. Semantics and relation to surrounding frameworks

The primary semantic consequence is that GCwFs allow direct interpretation of intrinsic syntax into the standard set semantics: $\mathsf{Con} : \mathsf{Type},$5 with contexts as sets, substitutions as functions, context extension as dependent sum, $\mathsf{Con} : \mathsf{Type},$6 as metatheoretic dependent function space, $\mathsf{Con} : \mathsf{Type},$7 by the set $\mathsf{Con} : \mathsf{Type},$8, and $\mathsf{Con} : \mathsf{Type},$9 by the family $\Gamma,\Delta : \mathsf{Con}$0. This works because the target of types is $\Gamma,\Delta : \mathsf{Con}$1, which is a groupoid, not necessarily a set; hence the constructor $\Gamma,\Delta : \mathsf{Con}$2 is interpretable (Altenkirch et al., 18 Sep 2025).

The GCwF framework is explicitly described as a natural extension of the CwF framework “when starting from a 1-category.” The weakening is deliberately conservative: $\Gamma,\Delta : \mathsf{Con}$3 remains strict as a category, $\Gamma,\Delta : \mathsf{Con}$4 remains set-valued, and only $\Gamma,\Delta : \mathsf{Con}$5 is weakened from a set-valued presheaf to a groupoid-valued 2-presheaf. This distinguishes GCwFs from broader weak semantics, such as comprehension categories or bicategorical approaches, in which substitution itself may be preserved only up to isomorphism (Altenkirch et al., 18 Sep 2025).

Two neighboring bodies of work clarify the categorical environment in which GCwFs sit. First, strict presheaf-based CwF semantics interprets types over a context $\Gamma,\Delta : \mathsf{Con}$6 as presheaves on the category of elements $\Gamma,\Delta : \mathsf{Con}$7, terms as natural sections, and context comprehension as a dependent sum

$\Gamma,\Delta : \mathsf{Con}$8

That construction is entirely set-valued, but it provides the standard representability pattern on which GCwF-style family semantics is built (Hu, 2021). Second, groupoid-valued presheaf categories

$\Gamma,\Delta : \mathsf{Con}$9

supply a canonical ambient category of strict groupoidal families, together with Grothendieck constructions $\mathsf{Tm}(\Gamma,A)$00 that behave like comprehension projections and show that groupoid-valued presheaves model homotopy types for a broad class of bases $\mathsf{Tm}(\Gamma,A)$01 (Guetta, 2022).

From an algebraic viewpoint, the generalized-algebraic-theory treatment of CwFs shows how internal categories, internal CwFs, and CwFs with extra type structure can be presented by sorts, operators, and equations, all preserved strictly by cwf-morphisms. That perspective does not define GCwFs, but it suggests that a strict algebraic presentation of “groupoid with families” could be obtained by extending the internal-category signature with inverses and inverse equations (Bezem et al., 2020). This suggests a close relation between GCwFs as introduced in HoTT and broader algebraic approaches to family-based semantics.

Ordinary CwFs also form the strict baseline from which GCwFs depart. In the classical account, a CwF consists of a category of contexts and substitutions, a $\mathsf{Tm}(\Gamma,A)$02-valued presheaf of types and terms, and context comprehension satisfying strict equations. Later comparisons with semantic categories such as finitely complete categories and locally cartesian closed categories show that strict structure often has to be weakened to coherent isomorphism at the level of morphisms. A plausible implication is that GCwFs occupy a distinct position: they retain strict substitution internally while weakening only the homotopically forced part of the syntax, namely type equality (Castellan et al., 2019).

6. Formalization, admissible coherence, and significance

All constructions in the main GCwF development were formalised in Cubical Agda. The entire framework is presented internally using path types, dependent paths, transport, and cubical composition and filling. The formalization imposes nontrivial design constraints: higher inductive-inductive constructors for mutually dependent sorts cannot be interleaved directly, so the syntax is encoded using a code sort $\mathsf{Tm}(\Gamma,A)$03 and an element family $\mathsf{Tm}(\Gamma,A)$04; transported variants such as $\mathsf{Tm}(\Gamma,A)$05 are introduced to avoid difficulties with explicit transport and strict positivity; and some equality characterizations use the Martin-Löf identity type rather than cubical path equality because $\mathsf{Tm}(\Gamma,A)$06 computes definitionally on $\mathsf{Tm}(\Gamma,A)$07 (Altenkirch et al., 18 Sep 2025).

An important formal observation is that some higher coherences one might expect to be primitive in the syntax, such as the pentagon $\mathsf{Tm}(\Gamma,A)$08 and the identity triangles $\mathsf{Tm}(\Gamma,A)$09, $\mathsf{Tm}(\Gamma,A)$10, are not built into the groupoid-syntax. After proving that $\mathsf{Tm}(\Gamma,A)$11 is a set, they become admissible. This is a practical simplification: the syntax need only include the coherence constructors necessary for $\mathsf{Tm}(\Gamma,A)$12, $\mathsf{Tm}(\Gamma,A)$13, and $\mathsf{Tm}(\Gamma,A)$14, while the remaining 2-dimensional coherence becomes derivable once set-truncation is established (Altenkirch et al., 18 Sep 2025).

The principal significance of GCwFs is therefore twofold. As a semantic notion, a GCwF enlarges the class of admissible models to include univalent models in which the collection of types is naturally groupoidal. As a syntax/semantics interface, it allows the use of ordinary intrinsic syntax because the free GCwF with a base family of sets and $\mathsf{Tm}(\Gamma,A)$15-types is nevertheless set-truncated. The result is a deliberately 1-dimensional compromise: enough higher structure to interpret natural HoTT models, but not enough to make equality in syntax intractable (Altenkirch et al., 18 Sep 2025).

A common misconception is that allowing groupoid-level type equality necessarily makes free syntax irreducibly higher-dimensional. The main theorem shows the opposite in the case studied: groupoid-truncation is introduced as a semantic relaxation, but the initial syntax collapses back to a set. Another possible misconception is that GCwFs weaken substitution itself; in fact, the construction keeps substitutions as hom-sets of a strict 1-category and weakens only the type family. That asymmetric weakening is the central design principle of the framework (Altenkirch et al., 18 Sep 2025).