Skip to content

Decision ADR 0014 Stacks And Moduli Ownership

Chris Dare edited this page Nov 15, 2026 · 1 revision

ADR-0014 — Stacks, algebraic spaces and moduli: one carrier, natural representability, a stability-free core

  • Status: accepted
  • Date: 2026-09-27 (America/New_York)
  • Decider: Chris Dare
  • Recorded: 2026-11-09 by MO2.1 (#1624). Mathlib paths and line numbers are at the revision pinned in lake-manifest.json (v4.35.0, 56a005855ea1); repository paths, line numbers and states are at main@7c1c3725.
  • Supersedes: the path and ownership decisions of ADR-0013: its Decision list and the SF8/SF9 ownership lines under its Consequences. ADR-0013's choice of the monorepo, and its final trust clause, stay in force; the clause is restated in §7.
  • Decision: every stack is a CategoryTheory.StackInGroupoids; representability is natural in the test object; a moduli problem is a prestack in groupoids cut out of a stack of categories; site-generic stacks, geometric stacks, algebraic spaces and moduli problems reach no stability condition; stability consumers are separate subcomponents.

Context

ADR-0013 (2026-08-17) brought Quot schemes, stacks and good moduli into the monorepo and named paths for them. Six weeks of construction showed that both the paths and several constructions need revising.

  • The paths were not used. Generic stacks in groupoids moved to CategoryTheory/Sites/Descent/StackInGroupoids/ on 2026-09-01 (commit 7d862cee), Tier 1 beside Mathlib's CategoryTheory/Sites/Descent/. Quot lives at AlgebraicGeometry/Moduli/Quot/, not AlgebraicGeometry/Quot. AlgebraicGeometry/Moduli/GoodModuli does not exist. The geometric stability consumers sit in AlgebraicGeometry/Moduli/{Semistability,HarderNarasimhan}/, and AlgebraicGeometry/Moduli.lean imports both, so importing the moduli umbrella imports Bridgeland stability.
  • Representability is pointwise (finding d1). CategoryTheory.StackMorphism.FiberRepresentation (DerivedAlgGeo/CategoryTheory/Sites/Descent/StackInGroupoids/Morphism.lean:76) gives each test object T its own equivalence Discrete (T ⟶ representing) ≌ f.FiberCategory y T, with nothing relating the equivalences along a map of test objects. Every representability, property, diagonal and presentation notion in AlgebraicGeometry/Stacks/{Algebraic,AlgebraicStack}.lean and AlgebraicGeometry/Moduli/PerfectComplex/Algebraicity.lean inherits this. For an infinite field k, X = Spec k[ε]/(ε²) ×ₖ 𝔸¹ and R = 𝔸¹ have Hom-sets of the same cardinality from every test scheme, so StackMorphism.IsSmoothSurjective holds for the morphism of representable stacks induced by X ⟶ Spec k, which is not smooth.
  • The boundedness witness does not bound (finding d2). In AlgebraicGeometry.FiniteTypeBoundednessWitness (AlgebraicGeometry/Moduli/PerfectComplex/Boundedness.lean:103), geometricFiber x is unrelated to universalFamily. The structure does not assert boundedness in the sense of arXiv:1902.08184v4, Definition 9.4, which requires the selected objects to be fibres of one family on a finite-type scheme.
  • Universes are tied (finding d3). StackInGroupoids fixes Cat.{w, w} (StackInGroupoids.lean:38), while Mathlib's Pseudofunctor.IsStack (Mathlib/CategoryTheory/Sites/Descent/IsStack.lean:49) allows Cat.{v', u'}.
  • Finite-type openness is vacuous. AlgebraicGeometry/Moduli/Semistability/FiniteType.lean:80 and the geometric base-change witness it consumes assume [∀ T : SchemeBaseChange S, IsLocallyNoetherian T.left], which is false for every nonempty S.
  • Some capability classes are unpinned. HasDerivedPushforward, HasDerivedTensor and HasCoherentDerivedTensor ask only for some exact functor, and the zero functor satisfies them. They are contracts, not constructions.

The stacks-and-moduli plan of 2026-09-27 answers these with milestones MO2 (this boundary), STK1–STK4 (stacks), ASP1 (algebraic spaces), DQ1 (derived pushforward and base change), GM1–GM2 (good moduli spaces), PJ1 (projective morphisms over a base), MOD1–MOD3 (moduli problems) and ART1 (Artin's axioms). SF9 is re-scoped to its stability consumers, and SF13 and SF14 take BLMNPS Part III and §§20–22. The plan, with the issue behind each milestone, is .claude/roadmap/stacks-and-moduli.yaml.

Decision

1. One carrier

The canonical stack is CategoryTheory.StackInGroupoids (DerivedAlgGeo/CategoryTheory/Sites/Descent/StackInGroupoids.lean:35): a groupoid-valued pseudofunctor on LocallyDiscrete Cᵒᵖ satisfying Mathlib's Pseudofunctor.IsStack.

  • Categories fibred over a base (Mathlib's CategoryTheory/FiberedCategory/: Functor.IsFibered, HasFibers, and the fibred Grothendieck construction Pseudofunctor.CoGrothendieck) are reached from the carrier only by a proved comparison (STK1.6, #1638), never by a second stack root.
  • A moduli problem is a prestack in groupoids. No ModuliProblem carrier is introduced. The existing problem records become abbrevs or specializations of the carrier when their consumers are next rebuilt.

2. Universes

Fibres live in universes independent of the site, Cat.{v', u'}, as in Mathlib's IsStack. The carrier's Cat.{w, w} is generalized (STK1.1, #1631). Representable stacks are ULifted into the fibre universe (STK1.7, #1637).

3. The base

A relative stack over a scheme S is a stack on Over S, the repository's SchemeBaseChange S := Over S (AlgebraicGeometry/DerivedCategory/Families/Scheme.lean:27). One proved comparison relates these to stacks on Scheme over the representable yoneda.obj S (STK2.6, #1645). No second base convention is introduced.

4. Representability is natural, and the pointwise notions say what they are

A morphism of stacks is representable when its 2-fibre product with every representable is representable, naturally in the test object (Stacks Project, Tag 02Y4, Section 4.42; Tag 02ZW, Definition 94.9.1, for algebraic spaces). On discrete stacks this agrees with Mathlib's Functor.relativelyRepresentable and MorphismProperty.presheaf (Mathlib/CategoryTheory/MorphismProperty/Representable.lean:103, :330).

  • The existing pointwise notions keep their definitions and gain the qualifier Pointwise (MO2.4, #1626). A type-valued structure takes it as a prefix (PointwiseFiberRepresentation); a predicate named Is… or Has… takes it after the verb (IsPointwiseRepresentable, HasPointwiseRepresentableProperty), as Mathlib writes HasPointwiseLeftKanExtension. The list: FiberRepresentation, IsRepresentable, FiberRepresentationWithProperty, HasRepresentableProperty, the stack-morphism abbreviations IsLocallyOfFinitePresentation, IsSmoothSurjective and IsOpenImmersion, DiagonalFiberRepresentation, HasRepresentableDiagonal, ZariskiStackPresentation(Over), FppfStackPresentation(Over) and the RelativePerfect…ZariskiPresentations, together with every declaration and field whose name spells one of them.
  • RepresentableQuotientProblem is an arbitrary represented presheaf, not a Quot functor; it becomes RepresentedPresheafOver. FiniteTypeBoundednessWitness does not bound; it becomes PointwiseFiniteTypeCover.
  • The natural notions later take the standard names (STK1.4, #1636; STK3.1, #1684), each with a proved projection to its Pointwise… predecessor. The old names are not kept as library aliases, because the natural notions need them.

5. Moduli of objects is one construction

A local, pullback-stable, isomorphism-closed object property of a stack of categories cuts out a substack, and the core of a stack of categories is a stack in groupoids (STK1.5, #1634; this is the TODO at Mathlib/CategoryTheory/Bicategory/Functor/Cat/ObjectProperty.lean:19). Coh_{X/S}, vector bundles, flat objects in a heart and M_σ(v) are instances.

  • Dqc is only a prestack.
  • The stack property of Lieblich's M_pug is its own theorem (Stacks Project, Tag 0DLH, Lemma 99.16.6; #1736), not an instance of the construction above.

6. Order

First come algebraic stacks whose diagonal is representable by schemes (STK3.1, #1684). They suffice for Coh_{X/S} (Tag 0DLY, Lemma 108.4.1) and for complexes (Tag 0DPW, Lemma 108.12.1). Algebraic spaces come in parallel: first as targets (Quot spaces and good moduli spaces), then with the standard definition (Tag 025Y, Definition 65.6.1; ASP1.1, #1651), and then algebraic stacks whose diagonal is representable by algebraic spaces (Tag 026O, Definition 94.12.1; STK3.6, #1690), with STK3.1 as a special case.

7. Hard theorems are typed hypotheses over genuine definitions

Algebraicity (Artin, Lieblich; Tag 0DLN, Theorem 99.16.12), the existence of good moduli spaces (Alper, Halpern-Leistner and Heinloth, arXiv:1812.01128v5, Theorem A) and the representability of Quot functors enter consumer theorems as explicit hypotheses whose statements use genuine definitions. The shape is (h : IsAlgebraic (Mpug X S)); both names are illustrative. At main@7c1c3725 there is no Mpug and no algebraicity predicate for stacks; the existing CategoryTheory.Triangulated.StabilityCondition.WithClassMap.IsAlgebraic (an algebraic stability condition) is unrelated.

  • Such a hypothesis is never a structure field, a typeclass instance, or an arbitrary Prop parameter. The former Theorem22_2DependencyContract was removed for exactly this reason (.claude/roadmap/theorem22-2-dependencies.yaml:41).
  • Discharging such a hypothesis later must not change any consumer's statement.
  • ADR-0013's clause stands: no structure field, typeclass instance or roadmap status may stand in for an unproved algebraicity, boundedness, properness or descent theorem.

8. Placement

A target is a confirmed destination with no module yet. It is never imported and never documented as implemented.

What Path State at main@7c1c3725
Site-generic stacks: the carrier, morphisms, 2-fibre products, representable morphisms, substacks cut out by object properties CategoryTheory/Sites/Descent/ (Tier 1 beside Mathlib's Sites/Descent/; namespace CategoryTheory) exists (StackInGroupoids/)
Pseudofunctor-level constructions: object-property loci, cores, transport CategoryTheory/Bicategory/Functor/Cat/ (Tier 1) exists
Comparison with categories fibred in groupoids CategoryTheory/FiberedCategory/ (Tier 1 beside Mathlib's) target (STK1.6)
Neutral moduli interfaces: the boundedness predicate and the family probes CategoryTheory/Moduli/ Boundedness.lean exists; the probes arrive with MO2.5 (#1627)
Topologies on schemes and scheme-site descent AlgebraicGeometry/Sites/, AlgebraicGeometry/Modules/… Sites/Comparison.lean exists
Descent for modules Algebra/Category/ModuleCat/Descent/ (Tier 1 beside Mathlib's Algebra/Category/ModuleCat/Descent.lean) target (STK2.3, #1644)
Algebraic spaces (a new object directory) AlgebraicGeometry/AlgebraicSpace/ target (ASP1)
Algebraic stacks, their properties, valuative criteria AlgebraicGeometry/Stacks/ exists (pointwise presentations only)
Test stacks Θ, B𝔾_m, ST_R AlgebraicGeometry/Stacks/Test/ target (STK4.4, #1699)
Good, coarse and categorical moduli spaces AlgebraicGeometry/Stacks/ModuliSpace/ (not Moduli/GoodModuli) target (GM1)
Moduli problems only: Coherent/, Quot/, PerfectComplex/, Picard/, Heart/, sheaf semistability AlgebraicGeometry/Moduli/ PerfectComplex/, Quot/ exist
Scheme realizations of the neutral family probes AlgebraicGeometry/Moduli/Probes.lean target (MO2.3, after MO2.5)
Stability consumers AlgebraicGeometry/Moduli/StabilityCondition/ (MO2.3, #1628, regroups Moduli/Semistability/ and Moduli/HarderNarasimhan/ there), AlgebraicGeometry/DerivedCategory/Stability/, CategoryTheory/Triangulated/StabilityCondition/Families/ the last two exist

Dependencies run from site-generic stacks to geometric stacks and algebraic spaces, from those to moduli problems, and from moduli problems to stability consumers. Site-generic stacks, the neutral moduli interfaces, pseudofunctor constructions, geometric stacks, algebraic spaces and moduli problems reach neither the stability tree nor a stability-consuming subcomponent, and stacks and algebraic spaces reach no moduli problem (the MO2.2 layering rule, #1625). AlgebraicGeometry/Moduli.lean omits Moduli/StabilityCondition/, which the top-level AlgebraicGeometry umbrella imports directly, as it already does for DerivedCategory/Stability/ (MO2.3).

ADR-0013 said ADR-0014 says
relative-perfect objects and their moduli problem under AlgebraicGeometry/Moduli/PerfectComplex unchanged for the moduli problem; the relative-perfect predicates are at AlgebraicGeometry/DerivedCategory/Perfect/ since MO1.11
Quot and bounded parameter spaces under AlgebraicGeometry/Quot AlgebraicGeometry/Moduli/Quot/ (the problem); Quot spaces as algebraic spaces or schemes are MOD3 targets
stacks in groupoids and descent under AlgebraicGeometry/Stacks site-generic stacks under CategoryTheory/Sites/Descent/; algebraic stacks under AlgebraicGeometry/Stacks/
good moduli under AlgebraicGeometry/Moduli/GoodModuli AlgebraicGeometry/Stacks/ModuliSpace/
stability-specific use under CategoryTheory/Triangulated/StabilityCondition/Families the three stability-consumer paths in the table above

9. Adoption

Each new root names two independent consumers, and the first is not a Bridgeland stability condition. Vector bundles inside Coh_{X/S} test open substacks; affine GIT quotients test good moduli spaces; quotient stacks test Θ-reductivity and S-completeness.

Consequences

  • The SF8/SF9 ownership lines of ADR-0013 are replaced. STK1–STK4 own the stack substrate, ASP1 algebraic spaces, GM1–GM2 good moduli spaces, MOD1–MOD3 the moduli problems, DQ1 derived pushforward and base change on Dqc, PJ1 projective morphisms over a base, and ART1 the route to algebraicity. SF8 keeps relative-perfect objects and arbitrary derived pullback (#554). SF9 keeps only its stability consumers, and SF13 and SF14 own BLMNPS Part III and §§20–22.
  • ADR-0013's temporary edge is gone: AlgebraicGeometry/Moduli/PerfectComplex/ reaches no stability module at main@7c1c3725, and the MO1.11 rule holds the predicates at AlgebraicGeometry/DerivedCategory/Perfect/.
  • Paths move and fully qualified names do not (docs/architecture/placement.md, the agreed owner map), except for the owner-approved renames of §4. A historical spelling survives only where an immutable review payload needs it, through exe/RestateHistoricalNames.lean; none of the renamed names occurs in attest/review.yaml.
  • Every issue in the plan above cites this record for its carrier, universes, base, representability, placement and hypothesis policy instead of re-deciding them.

Non-claims

  • This record proves nothing. No stack, algebraic space, good moduli space or moduli problem named in §8 exists before its issue lands, and a target path is not an implemented import.
  • It does not assert that any existing presentation is an algebraic stack, that any finite-type cover bounds, or that any semistable locus of a genuine family is open.
  • It does not choose a proof route for algebraicity (ART1.1, #1752 decides) or for the existence of good moduli spaces (GM2.5, #1721 decides).

Exported from .claude/decisions/ADR-0014-stacks-and-moduli-ownership.md at the commit that removed it from the tree.

Home

Architecture

CI and runners

Decision records

Roadmaps

Clone this wiki locally