Repository navigation
Architecture Placement
This is the operational placement test for new declarations and structural moves. Apply it with the subject/application boundaries in mathematical-ownership.md. Tier 1 follows an identified Mathlib API owner; deciding whether a declaration extends that API or introduces a new subject still requires mathematical review.
A direct extension of a Mathlib API follows that API's definition site at the
repository's pinned revision, under DerivedAlgGeo/, in that API's namespace.
Using a Mathlib type does not by itself make a new concept an extension of its
elementary API. Follow Tier 2 for a new subject. This alignment aids navigation
and future upstream work; matching directories is not required to import Mathlib.
| Concept the declaration extends | Mathlib path | Repository path |
|---|---|---|
DerivedCategory C, Ext, K-projectives, its t-structure, Bounded
|
Algebra/Homology/DerivedCategory/ |
Algebra/Homology/DerivedCategory/ |
HomotopyCategory, HomComplex, bounded and plus variants |
Algebra/Homology/HomotopyCategory/ |
Algebra/Homology/HomotopyCategory/ |
HomologicalComplex.HomologySequence short-exact comparison lemmas |
Algebra/Homology/HomologySequenceLemmas.lean |
Algebra/Homology/HomologySequenceLemmas.lean |
CochainComplex.truncLE, ιTruncLE, and truncLEMap: the canonical increasing good-truncation tower and its colimit |
Algebra/Homology/Embedding/CochainComplex.lean |
Algebra/Homology/Embedding/CochainComplex.lean |
HomologicalComplex.stupidTrunc inclusions between integer degree-at-least truncations |
Algebra/Homology/Embedding/StupidTrunc.lean |
Algebra/Homology/Embedding/StupidTruncGE.lean |
HomologicalComplex.stupidTruncXIso normalized retained-component comparison and its bicomplex specialization |
Algebra/Homology/Embedding/StupidTrunc.lean |
Algebra/Homology/Embedding/StupidTruncGE.lean |
| Spectral sequences | Algebra/Homology/SpectralSequence/ |
Algebra/Homology/SpectralSequence/ |
| Coproducts indexed by a family with finite nonzero support | CategoryTheory/Limits/Shapes/FiniteProducts.lean |
CategoryTheory/Limits/Shapes/FiniteProducts.lean |
Sigma.π finite-support projection identity for a supplied coproduct |
CategoryTheory/Limits/Shapes/ZeroMorphisms.lean |
CategoryTheory/Limits/Shapes/ZeroMorphisms.lean |
HomologicalComplex₂.totalFunctor and direct-sum totalization |
Algebra/Homology/TotalComplex.lean |
Algebra/Homology/TotalComplex.lean |
HomologicalComplex.mapBifunctorMap and its finite-support quasi-isomorphism criteria |
Algebra/Homology/Bifunctor.lean |
Algebra/Homology/Bifunctor.lean |
SheafOfModules and PresheafOfModules
|
Algebra/Category/ModuleCat/{Sheaf,Presheaf}/ |
Algebra/Category/ModuleCat/{Sheaf,Presheaf}/ |
ModuleCat, Grp
|
Algebra/Category/{ModuleCat,Grp}/ |
Algebra/Category/{ModuleCat,Grp}/ |
Effective descent for modules along faithfully flat ring maps (Mathlib's comonadicExtendScalars and its descent TODO) |
Algebra/Category/ModuleCat/Descent.lean |
Algebra/Category/ModuleCat/Descent/: target (STK2.3, #1644); no module yet; do not import (ADR-0014 §8) |
ObjectProperty, the full subcategory it cuts out, and functor lifts into it |
CategoryTheory/ObjectProperty/ |
CategoryTheory/ObjectProperty/ |
| Sites, sheaves, sheaf cohomology, descent, stacks | CategoryTheory/Sites/ |
CategoryTheory/Sites/, with Čech theory under SheafCohomology/Cech/ and stacks under Descent/, where StackInGroupoids/ holds the one stack carrier (ADR-0014 §1) with its morphisms, discrete and representable stacks; 2-fibre products, representable morphisms and substacks belong there when built |
Bicategories, pseudofunctors, Pseudofunctor.ObjectProperty
|
CategoryTheory/Bicategory/ |
CategoryTheory/Bicategory/, with Cat-valued pseudofunctor loci, cores and transport under Functor/Cat/ (ADR-0014 §5) |
Categories fibred over a base: Functor.IsFibered, HasFibers, Pseudofunctor.CoGrothendieck
|
CategoryTheory/FiberedCategory/ |
CategoryTheory/FiberedCategory/: target (STK1.6, #1638); no module yet; do not import. StackInGroupoids reaches it only by a proved comparison, never as a second stack root (ADR-0014 §1) |
| Abelian categories, Serre classes | CategoryTheory/Abelian/ |
CategoryTheory/Abelian/ |
| Quasi-abelian categories and strict morphisms |
CategoryTheory/Abelian/ (precedent Abelian/NonPreadditive.lean) |
CategoryTheory/Abelian/QuasiAbelian.lean |
| Linear categories and opposite linearity | CategoryTheory/Linear/ |
CategoryTheory/Linear/ |
| Compact objects and coproduct-preserving functors on a preadditive category | CategoryTheory/Preadditive/ |
CategoryTheory/Preadditive/CompactObject.lean |
SmallShiftedHom in a localization |
CategoryTheory/Localization/ |
CategoryTheory/Localization/ |
| The opposite of a (pre)triangulated category and its shift | CategoryTheory/Triangulated/Opposite/ |
CategoryTheory/Triangulated/Opposite/ |
HomComplex and its cohomology classes |
Algebra/Homology/HomotopyCategory/ |
Algebra/Homology/HomotopyCategory/ |
| Pretriangulated and triangulated categories, t-structures | CategoryTheory/Triangulated/ |
CategoryTheory/Triangulated/ |
| Monoidal categories and their compatibility with other structure | CategoryTheory/Monoidal/ |
CategoryTheory/Monoidal/ |
| Simplicial objects and face-map complexes | AlgebraicTopology/ |
AlgebraicTopology/ |
| Sheaves on a topological space, stalks, the category of opens |
Topology/Sheaves/, Topology/Category/TopCat/
|
Topology/Sheaves/, Topology/Category/TopCat/
|
PrimeSpectrum and its topology |
RingTheory/Spectrum/Prime/ |
RingTheory/Spectrum/Prime/ |
HomogeneousLocalization and graded algebras |
RingTheory/GradedAlgebra/ |
RingTheory/GradedAlgebra/ |
LocalSubring and its domination order |
RingTheory/LocalRing/ |
RingTheory/LocalRing/ |
| Descent of finite presentation and of flatness along faithfully flat ring maps |
RingTheory/Finiteness/Descent.lean, RingTheory/Flat/FaithfullyFlat/
|
RingTheory/Finiteness/Descent.lean, RingTheory/Flat/FaithfullyFlat/Descent.lean
|
| Modules, localization, graded modules, polynomials |
Algebra/Module/, Algebra/MvPolynomial/
|
the same |
| Lattices, bilinear forms, exterior powers, finite-dimensional lemmas | LinearAlgebra/ |
LinearAlgebra/ |
Upper half-planes of ℂ, argument bounds, polygonal-path length and perimeter |
Analysis/Complex/, Analysis/Convex/ (no Mathlib owner: the pinned revision has no chain length and no perimeter) |
Analysis/Complex/, Analysis/Convex/ComplexPolygonalPath/
|
| Euler characteristics of graded objects | Algebra/Homology/EulerCharacteristic.lean |
Algebra/Homology/EulerCharacteristic.lean |
| Alternating sums along finite exact sequences | Algebra/Exact/Sequence.lean |
Algebra/Exact/Sequence.lean |
Schemes, X.Modules, Proj, morphism properties |
AlgebraicGeometry/ |
AlgebraicGeometry/, with ProjectiveSpectrum/ under Mathlib's name |
If two rows seem to apply, identify the API actually being extended, rather
than choosing the weakest type appearing in the statement.
Within the spectral-sequence row, FilteredTotalComplexAdjacentCore.lean owns the
generic adjacent-column short exact sequence, its natural map, and the
single-column total-map quasi-isomorphism criterion.
FiniteStripTotal.lean consumes that core to prove the finite horizontal-strip
comparison for the actual HomologicalComplex₂.total.map. The spectral-page
and first-page files consume the core downstream; they do not define a second
adjacent sequence. The retained truncation-component iso, inclusion naturality,
and criterion that the inclusion is an isomorphism when discarded components
vanish are owned separately by Embedding/StupidTruncGE.lean. The
finite-support coproduct lemma used to construct total objects is owned by
CategoryTheory/Limits/Shapes/FiniteProducts.lean; it has no bicomplex premise.
PrimeSpectrum.basicOpen_prod_eq_pi is stated
in the lattice of opens of a prime spectrum, but PrimeSpectrum.basicOpen is defined in
RingTheory/Spectrum/Prime/Topology.lean, so it lives there and not in
Topology/ or Algebra/. The basic-open cover bridge
PrimeSpectrum.coversTop_basicOpen_of_span_eq_top also lives under
RingTheory/Spectrum/Prime/: it uses only a commutative semiring and
PrimeSpectrum.basicOpen, then composes the prime-spectrum supremum criterion with the
topological open-site cover theorem. Stalks of module presheaves are stated with
TopCat, germs, and stalk functors, which Mathlib defines in
Topology/Sheaves/, so they live there.
The local AlgebraicGeometry/Spec/ path is already retired by the #1607 cutover; the issue's
repository snapshot predates that move. The former folder held a scheme-namespaced convenience
theorem about basic opens, whose scheme-free owner is now RingTheory/Spectrum/Prime/. Mathlib
builds the spectrum as a locally ringed space in AlgebraicGeometry/Spec.lean and defines the
scheme-valued AlgebraicGeometry.Spec and its contravariant functor in
AlgebraicGeometry/Scheme.lean. Both use the
prime-spectrum carrier, which Mathlib defines separately from its topology. At the repository's
Mathlib v4.32.1 pin
(commit 520045ab14e26149ee970e2e617ca04b09bde5d6),
the underlying topological space of scheme Spec R is Mathlib's PrimeSpectrum R, and the
scheme API adds its structure sheaf and locally ringed-space structure. The ownership map is:
| Foundation | Pinned Mathlib owner and APIs | DerivedAlgGeo extensions |
|---|---|---|
| Prime-spectrum carrier, topology, and basic opens |
RingTheory/Spectrum/Prime/Defs.lean: PrimeSpectrum; Topology.lean: PrimeSpectrum.basicOpen; Topology.lean: PrimeSpectrum.iSup_basicOpen_eq_top_iff
|
RingTheory/Spectrum/Prime/, including PrimeSpectrum.coversTop_basicOpen_of_span_eq_top as the scheme-free basic-open covering bridge |
| Spectrum as a locally ringed space and as a scheme |
AlgebraicGeometry.Spec.topObj, AlgebraicGeometry.Spec.topObj_forget, and AlgebraicGeometry.Spec.toLocallyRingedSpace identify its topological and sheafed presentations; AlgebraicGeometry.Spec and AlgebraicGeometry.Scheme.Spec give the scheme and its contravariant functor |
Scheme-specific results stay in AlgebraicGeometry/; affine module-sheaf results stay in AlgebraicGeometry/Modules/Affine/
|
| Global-sections/Spec adjunction and affine schemes |
GammaSpecAdjunction.lean: AlgebraicGeometry.ΓSpec.adjunction; AffineScheme.lean: AlgebraicGeometry.AffineScheme.equivCommRingCat
|
These are the upstream Spec/affine-scheme comparison APIs, not local declarations |
| Associated modules, localization criterion, and affine quasi-coherent equivalence |
AlgebraicGeometry/Modules/Tilde.lean: AlgebraicGeometry.IsLocalizing; isIso_fromTildeΓ_iff_isLocalizing; isIso_fromTildeΓ_of_isQuasicoherent; and AlgebraicGeometry.tildeEquiv
|
Modules/Affine/ retains AlgebraicGeometry.Scheme.Modules.basicOpenRestriction and a local formulation of the upstream criterion, plus component, presentation-cover, and repository-specific bridges |
| Proj's standard affine charts, cover, and structure map |
ProjectiveSpectrum/Basic.lean: AlgebraicGeometry.Proj.basicOpenIsoSpec, AlgebraicGeometry.Proj.awayι, AlgebraicGeometry.Proj.affineOpenCover, AlgebraicGeometry.Proj.toSpecZero
|
AlgebraicGeometry/ProjectiveSpectrum/ develops further scheme and module-sheaf results on these existing charts |
| The chart ring used by Proj |
RingTheory/GradedAlgebra/HomogeneousLocalization.lean: HomogeneousLocalization.Away
|
AlgebraicGeometry.Proj.chartRing is only an abbrev bundling this ring as CommRingCat
|
These are the implemented owners and paths after the #1607 cutover; the map records no proposed directory move or unimplemented target.
The dependency is concrete and one-way: ProjectiveSpectrum/Modules/ChartExtension.lean imports
Modules/Affine/Extension.lean, which imports Mathlib's
AlgebraicGeometry/Modules/Tilde.lean. ChartExtension.lean defines
AlgebraicGeometry.Proj.chartRing as an abbrev for HomogeneousLocalization.Away and defines
AlgebraicGeometry.Proj.awayRestrict by restricting along Mathlib's existing open immersion
AlgebraicGeometry.Proj.awayι. It does not define another affine chart or spectrum.
For a graded ring 𝒜 and homogeneous f of positive degree, Mathlib's
AlgebraicGeometry.Proj.basicOpenIsoSpec gives the chart isomorphism
(AlgebraicGeometry.Proj.basicOpen 𝒜 f).toScheme ≅ AlgebraicGeometry.Spec (CommRingCat.of (HomogeneousLocalization.Away 𝒜 f)).
The chart ring is the degree-zero homogeneous localization usually written (A[f⁻¹])₀ or
A_{(f)}, not the full ordinary localization Localization.Away f.
AlgebraicGeometry.Proj.affineOpenCover packages these charts into an affine open cover, and
AlgebraicGeometry.Proj.toSpecZero is the structure map
AlgebraicGeometry.Proj 𝒜 ⟶ AlgebraicGeometry.Spec (CommRingCat.of (𝒜 0)).
AlgebraicGeometry.Proj.awayι_toSpecZero verifies that each chart immersion followed by this
structure map is the map induced from 𝒜 0 to the chart ring.
Keep the relationships distinct:
-
Ownership: ring-theoretic
PrimeSpectrumextensions followRingTheory/Spectrum/Prime/; genuine scheme results follow Mathlib'sAlgebraicGeometry/owner; affine module-sheaf extensions stay underModules/Affine/; Proj chart consumers stay underProjectiveSpectrum/. - Dependency: the local Proj module-chart extension imports the affine module extension, which uses Mathlib's module-sheaf API.
-
Specialization and comparison maps:
AlgebraicGeometry.Spec.topObj_forgetidentifies the topological carrier;AlgebraicGeometry.ΓSpec.adjunctionrelates global sections andSpec;AlgebraicGeometry.AffineScheme.equivCommRingCatidentifies affine schemes with opposite commutative rings;AlgebraicGeometry.IsLocalizingandAlgebraicGeometry.isIso_fromTildeΓ_iff_isLocalizinggive the general localization criterion;AlgebraicGeometry.Scheme.Modules.isIso_fromTildeΓ_iff_isLocalizedModulerestates it using the localAlgebraicGeometry.Scheme.Modules.basicOpenRestrictionmap.AlgebraicGeometry.tildeEquivandAlgebraicGeometry.Scheme.Modules.isIso_fromTildeΓ_of_isQuasicoherentgive the affine module comparison;AlgebraicGeometry.Proj.awayι_toSpecZerorelates each standard chart to the structure map.
The size of a local directory measures the additional DerivedAlgGeo extensions and proofs it contains, not how much affine or projective spectrum theory is available to consumers: the core scheme, module-sheaf, and Proj chart APIs listed above are already in Mathlib.
For the mathematical picture, see nLab on affine schemes,
nLab on projective schemes, and the Stacks
Project discussion of Proj, its affine charts, and the map to Spec A₀ (Tag
01M3).
Tier 1 decides where an extension of an existing Mathlib API lives. It does
not follow that everything built with a Mathlib API belongs at that API's path.
A new mathematical object that merely uses Module.Dual, a bilinear form, or
HomComplex is placed by its own subject and at an appropriately general root,
by Tier 2. Reading Tier 1 as a universal carrier rule is how a new subject ends
up filed under whichever upstream definition its implementation happened to
reach for, and how the same policy simultaneously recognized conceptual
ownership for weak charges while filing the analogous numerical constructions
under a consumer directory.
Two consequences, both load-bearing:
- A directory is an index; it is not the import graph and not the specialization graph. The three must agree where they overlap and cannot be identical. A comparison theorem between two constructions does not license the generic one to import the special one, and a shared directory does not make two constructions instances of each other.
- A narrowly imported module is not a full subject umbrella. Keep them distinct in prose and in the layering gate: an umbrella re-exports its children by construction, so a rule that holds for a leaf may fail for the umbrella above it for reasons that have nothing to do with the leaf's subject.
Proposed destinations in an architecture review are local recommendations, not promises that Mathlib will accept a module of that name.
docs/architecture/cutover-ledger.md carries the declaration-level owner map
agreed in MO1.01 (#1312) for the 2026-09-13 ownership review: for each
confirmed finding, the definition owner, the neutral core, the application
adapter and the comparison owner, together with the independent-consumer
justification for every proposed new carrier. Two of its standing decisions
apply to every structural change in the repository, not only to MO1:
- paths move and fully qualified declaration names do not, so a relocation never invalidates a historical review payload; and
- a proposed new carrier needs a consumer outside the module it was extracted
from, or it is replaced by a theorem or an
abbrev-- an extraction is not its own second consumer.
Read that map before proposing a destination for code it already covers.
| Situation | Mathlib precedent | Repository placement |
|---|---|---|
| A structure on an abstract triangulated category |
Triangulated/TStructure/, Subcategory, Orthogonal, LocalizingSubcategory, Generators
|
CategoryTheory/Triangulated/<Name>/ |
| A weakened or strengthened variant of a named concept |
Topology/MetricSpace/Pseudo/, Monoidal/Braided/, Monoidal/Closed/
|
a child directory named by the adjective, below the canonical concept |
| Compatibility between two independent structures |
Monoidal/Preadditive.lean, Monoidal/Linear.lean
|
CategoryTheory/Monoidal/<Other>.lean |
| A structure on an abstract abelian category |
Abelian/SerreClass/, Abelian/GrothendieckCategory/
|
CategoryTheory/Abelian/<Name>/ |
| A geometric realization of a categorical interface |
Algebra/Category/ModuleCat/Abelian.lean, AlgebraicGeometry/Modules/Sheaf.lean
|
with the geometric object under AlgebraicGeometry/; the declaration may keep the interface's namespace for dot notation |
| A bespoke carrier built on a Mathlib API | definition site | beside that API |
A neutral predicate on pseudofunctors used by both geometry and stability, and the neutral family probes (ChargeProbe, OpenLocusProbe, GenericSemistabilityProbe, DedekindHNProblem, WeakDedekindHNProblem and their predicates) |
CategoryTheory/Bicategory/Functor/Cat/ |
CategoryTheory/Moduli/ until Mathlib has a home for it; the probes' file CategoryTheory/Moduli/Probes.lean is a target (MO2.5, #1627); no module yet; do not import. Until #1627 lands the probes sit in CategoryTheory/Triangulated/StabilityCondition/Weak/Families/Basic.lean
|
A theorem whose public type mentions a scheme, variety, Coh X, Dqc X, or a geometric morphism property |
AlgebraicGeometry/ |
AlgebraicGeometry/, organized by geometric object |
| Algebraic spaces over a scheme (ADR-0014 §6, §8) | one directory per geometric object: AlgebraicGeometry/ProjectiveSpectrum/, AlgebraicGeometry/EllipticCurve/
|
AlgebraicGeometry/AlgebraicSpace/: target (ASP1, #1651); no module yet; do not import |
| Algebraic stacks, their properties and valuative criteria | the same object-directory precedent |
AlgebraicGeometry/Stacks/; today it holds pointwise presentations only (ADR-0014 §4) |
| Good, coarse and categorical moduli spaces of stacks | inside an object directory, a child named by what it adds, as Mathlib's Algebra/Category/ModuleCat/Monoidal/
|
AlgebraicGeometry/Stacks/ModuliSpace/, never Moduli/GoodModuli: target (GM1); no module yet; do not import |
| Test stacks Θ, B𝔾_m, ST_R | the same |
AlgebraicGeometry/Stacks/Test/: target (STK4.4, #1699); no module yet; do not import |
| Moduli problems, versus their stability consumers | the repository's AlgebraicGeometry/DerivedCategory/ umbrella, which omits its Stability/ child |
moduli problems under AlgebraicGeometry/Moduli/; their stability consumers under AlgebraicGeometry/Moduli/StabilityCondition/, a target (MO2.3, #1628; no module yet; do not import) that the Moduli umbrella will omit. Until #1628 lands they sit at Moduli/{Semistability,HarderNarasimhan}/
|
Within Tier 2, identify the mathematical object and its independently useful theory. Use the weakest sufficient hypotheses to separate a reusable block from its application, not to rank all subjects or file every construction under its most elementary carrier. A stability function has a stability owner even though its charge is an additive map. A general bilinear identity has a linear-algebra owner even when first used by stability. This subject decision never relocates a direct Mathlib extension away from its API owner.
- The abstraction level of the statement. A derived category is a
triangulated category, and Mathlib still files it under
Algebra/Homology. - The weakest vocabulary in the signature, when the carrier has a definition site.
- The motivating theorem, the current filename, the first consumer, or the proof technique.
- The namespace. Declarations inside
LinearMap,Module,SheafOfModules, orChargeProbestill live at their carrier's path. - Which interface a geometric object happens to satisfy.
AlgebraicGeometry/is organized by object; inside an object directory the files are named by the structure they add.
Before adding or moving a public declaration:
- Search Mathlib and this repository for the carrier or concept.
- Name the canonical root module and the concrete consumer module.
- State the Lean relationship between them: direct reuse,
extends, an instance, anabbrev, or a proved comparison. - Verify that the root imports no specialization, consumer, paper-specific file, or geometric realization, including through umbrellas. Put comparison modules downstream of both presentations.
- Import the root directly from the consumer. Do not add a compatibility shim merely to preserve the old motivational path.
- Update the nearest umbrella, axiom audit, declaration baseline, layering gate, and this documentation in the same change.
If a file mixes a generic block, its application and their comparison, split at those declaration boundaries. Record hypotheses separately from subject ownership. Use the decision record to make the owner, imports and Lean specialization map reviewable.
The charge ownership cutover is complete (#1313). General paired functionals,
quadratic continuity and coercivity, and abstract Hodge-signature input live
under LinearAlgebra/. Additive charge families and exponential, numerical and
divisorial constructors live under
CategoryTheory/Triangulated/StabilityCondition/CentralCharge/. Wall equations
import those constructors from StabilityCondition/Walls/; full support
predicates live under StabilityCondition/Support/; geometric Chern/Todd
realizations remain under AlgebraicGeometry/Numerical/. Rule 8 of
scripts/check_layering.py pins the owners and checks their transitive import
boundaries. Preserve the existing roots and comparisons instead of creating
another charge carrier or retired-path shim.
The numerical specialization cutover is complete (#1316). Generic
Mukai/SqrtTodd.lean, Mukai/VectorClass.lean, Stability/Slope.lean, and
Stability/PolarisedWallTransport.lean import no K3, named-surface, or
dimension-specific transport consumer. K3 consequences and cross-model or
surface/threefold comparisons live in explicit downstream modules.
The numerical model cutover is complete (#1317). A formal
rank-degree-coordinate model -- an intersection ring, its grading, a degree
map, Chern and Todd coefficients, Riemann--Roch -- belongs under
AlgebraicGeometry/Numerical/Models/, with the named surface models as
siblings over Models/Surface/RankOne.lean and the dimension-general ring at
Models/MonogenicRing.lean. A realization map into a real divisor space, a
charge, a wall, a slice or a region built on such a model is a demonstration
and belongs under AlgebraicGeometry/Numerical/Examples/, which imports
Models/ and is never imported by it. A K3-only adapter carries a K3
suffix. Name a polarization slice as a slice, and do not name a construction
after Picard rank one when what it uses is generation of the whole
intersection ring by one class. None of these models is a scheme: the
scheme-level object stays with its geometric owner, and a scheme-level
specialization of a stability construction stays under
DerivedCategory/Stability/ with a Scheme suffix. Rule 12 of
scripts/check_layering.py pins the owners, the sibling structure, and the
separation of the arbitrary-divisor-rank charge branch from the scalar
H-degree compression.
Orthogonal exceptional blocks and a chosen right adjoint to a residual
full-subcategory inclusion are structures on abstract (pre)triangulated
categories, so their canonical owner is
CategoryTheory/Triangulated/SemiorthogonalDecomposition/. Constructors
for objectwise mutation triangles and the same-projection theorem also live
there; neither requires scheme vocabulary. The k-linear Serre duality data,
its Hom-finiteness hypothesis, the Serre pairing and trace, and uniqueness of
the Serre functor mention no shift and no triangulation, so their owner is
CategoryTheory/Linear/SerreFunctor/, with the three representability helpers
that need no Serre datum at all in CategoryTheory/Linear/Yoneda.lean. Ext
profiles, transport of spherical and pseudoprojective objects, and
classification-induced matching are shift-dependent and stay in
CategoryTheory/Triangulated/SerreFunctor/, which imports the linear root.
Adjacent Ext shift rigidity and ordered
block-length comparison belong with the semiorthogonal root. One-step and
dependent finite kernel extension, including generation from right
admissibility, live beside the generic Fourier--Mukai API. Only the comparison
with an Enriques surface's ten line bundles, its already-defined residual
property, the numerical (-2)-curve chains, and the paper-specific
degree-three and ambient-equivalence conclusions belong under
AlgebraicGeometry/Surface/Enriques/; the dependency runs from that consumer
to the categorical roots.
An adjunction in its most general implemented form is
CategoryTheory.Bicategory.Adjunction. Mathlib identifies ordinary functor
adjunctions with its Cat specialization. A theorem that merely assumes an
ordinary adjunction is placed by the other vocabulary in its signature:
preservation results belong with limits, Ext comparisons with derived Ext,
and kernel packages with Fourier--Mukai theory.
Comparison data of type (DerivedCategory C)ᵒᵖ ≃ DerivedCategory Cᵒᵖ extends
Mathlib's derived category, so its owner is
Algebra/Homology/DerivedCategory/Opposite.lean. Algebraic linear duality on
ModuleCat is one subject at ModuleCat's own Mathlib path:
Algebra/Category/ModuleCat/LinearDual/Basic.lean bundles the contravariant
additive functor and LinearDual/Exact.lean proves it exact over a field.
Being category-theoretic is not a placement reason -- ModuleCat is defined in
Mathlib/Algebra/Category/ModuleCat/, and Tier 1 follows the definition site.
The derived lift stays at Algebra/Homology/DerivedCategory/LinearDual.lean,
because it extends the derived category rather than ModuleCat. A Serre-duality statement that mentions a scheme
imports those roots and the canonical coherent-derived specialization; its
geometric motivation does not move the generic functors into duality.
The comparison Dᵇ(Coh X) ≃ Dᵇ_coh(Dqc X) intrinsically mentions a scheme,
so its statement and conditional consumer API belong under
AlgebraicGeometry/DerivedCategory/Dqc/. The existence proposition is not an
instance: Dqc/Comparison.lean accepts it explicitly and produces a coherent
representative and comparison isomorphism. The same rule applies to the
conditional equality of the perfect and compact-object properties. A moduli
consumer adds its relative-perfect and bounded-cohomology hypotheses without
becoming the owner of either comparison.
The word "perfect" currently appears in three non-interchangeable APIs. The
paths below describe the present implementation. Moving foundational relative
perfectness out of the moduli consumer is tracked in #1322; it does not change
these mathematical comparisons. In particular, the current cohomological
schemePseudoCoherent criterion is not automatically standard pseudo-coherence
on arbitrary non-Noetherian bases; preserve the scope warning in #554.
The ambient categories and formal relationships are:
| Notion | Ambient object | Meaning and owner | Valid comparison |
|---|---|---|---|
schemePerfect X |
D(Coh X) |
Thick envelope of degree-zero finite locally free coherent sheaves; AlgebraicGeometry/DerivedCategory/Coherent.lean
|
perfectDerivedToDqc_obj_mem_schemePerfectInDqc maps it into the defining perfect essential image in Dqc(X)
|
schemeRelativePerfect p |
Dqc(X) for p : X ⟶ S
|
Pseudo-coherence plus local finite Tor amplitude over the chosen base; DerivedCategory/Perfect/Relative.lean
|
It implies schemePseudoCoherent; it is not identified with absolute perfection without an additional geometric theorem |
Coh.TwoTermPerfectDeterminantData F |
a coherent sheaf F plus presentation data |
An explicit two-term finite locally free resolution used by determinant and Chern-class consumers; Divisors/Determinant.lean
|
Moduli/PerfectComplex/Comparison.lean forgets it to an absolute perfect degree-zero object and then to schemePerfectInDqc
|
schemePerfectInDqc X is the bridge, not a fourth competing definition: it
is the essential image of SchemePerfectDerivedCategory X under the concrete
coherent-derived functor. The general equality with compact objects remains
explicit evidence. Neither relative perfection nor two-term amplitude is
promoted to an equivalence with the full absolute perfect locus.
The canonical zero of SchemeQuasicoherentDerivedCategory X belongs to
Dqc.lean and is available for every scheme. Perfect-moduli files may prove
that this root object satisfies their additional predicates, but may not
reconstruct or rename the ambient zero.
ADR-0014
fixes the carrier and the placement table for stacks, algebraic spaces and
moduli problems. Every stack is a CategoryTheory.StackInGroupoids; a moduli
problem is a prestack in groupoids, with no separate ModuliProblem carrier.
Moduli of objects is to be one construction (STK1.5, #1634; not yet built):
Pseudofunctor.ObjectProperty cuts the substack out of a stack of categories,
and the core makes it a stack in groupoids.
CategoryTheory.Moduli.BoundednessProblem is the neutral boundedness root
only until #1692 replaces it by bounded families. A stability-family package
may require it and a geometric moduli package may realize it with finite-type
parameter data, but neither consumer owns the predicate.
Mathlib's Pseudofunctor.ObjectProperty is the canonical fiberwise locus for a
Cat-valued pseudofunctor, and the substack mechanism:
-
IsClosedUnderIsomorphismsmakes the locus replete; -
IsClosedUnderMapObjmakes it stable under restriction; -
fullsubcategoryandιconstruct the sub-pseudofunctor and its inclusion.
Repository extensions to that mechanism live under
CategoryTheory/Bicategory/Functor/Cat/ObjectProperty/, where Mathlib defines
it. Do not introduce a parallel Subprestack carrier in algebraic geometry. An
indexed collection of isomorphism-closed fiber predicates is not yet a
subprestack until restriction stability is supplied.
Concrete finite-type witnesses, atlases, scheme presentations, semistable loci,
and Harder--Narasimhan filtrations remain under AlgebraicGeometry/Moduli/
because their signatures intrinsically mention geometry. The relative-perfect
predicates do not: pseudo-coherence, local finite Tor amplitude and relative
perfection are properties of one complex over one morphism, needed by base
change and derived operations before a moduli functor exists, so their owner is
AlgebraicGeometry/DerivedCategory/Perfect/ and the moduli problem consumes
them. Stalkwise flatness of a module sheaf over a morphism is a further step
down, at AlgebraicGeometry/Modules/Flat.lean.
AlgebraicGeometry.RelativePerfectModuliSelector is the canonical name for
the weaker geometric input used by finite-type boundedness: it stores the
independent replete familyLocus and geometricLocus over each actual scheme
base change. It is not a pseudofunctor object property and provides no
restriction functors. The affine construction first places its locus on an
actual Cat-valued pseudofunctor, applies universallyStable, and only then
calls fullsubcategory; that is the root-to-consumer relationship required
for a genuine subprestack.
Every structural pull request must include:
- for each moved or new public root, the Tier 1 row or the Tier 2 precedent that places it;
- the root-to-consumer import direction;
- the specialization map or a statement that the consumer directly reuses the root;
- any adjacent misplaced declarations discovered while editing, recorded in
docs/architecture/cutover-ledger.mdif they are not part of the same slice.
Exported from docs/architecture/placement.md at the commit that removed it from the tree.
Architecture
- Canonical roots and specialization trees
- Structural cutover ledger
- Generalization backlog
- Dependency direction
- Loop-engineering friction log
- Mathematical ownership and abstraction boundarie
- Declaration placement
- When a proposition-valued class is justified
- Enhanced spherical twists: conventions and remai
- Verification history
CI and runners
- Bounded audit batching probe, 2026-09-27
- CI1.10: audit and exception record contention, 2
- Proposed hosted-only CI cutover for CI1.07
- Mac VM runners
- Owner CI lane: self-hosted runners in a private
- CI performance baseline, 2026-09-27
- CI1.07 exact-tree Ubuntu routing sample, 2026-09
- Ubuntu workstation runners (historical topology)
Decision records
- ADR-0001 — The seam is cold: files, not runtime (superseded)
- ADR-0002 — Identity is minted in the topic repo, (superseded)
- ADR-0003 — Elaboration and axiom evidence is pro (superseded)
- ADR-0004 — The statement registry lives in the t (superseded)
- ADR-0005 — Seven axes, four values, no aggregate (superseded)
- ADR-0006 — All contract issues are filed on `bri (superseded)
- ADR-0007 — Where the contract package lives (superseded)
-
ADR-0008 —
@[cites]ships as a shared Lake dep (superseded) - ADR-0009 — The contract package lives with the e (superseded)
- ADR-0010 — How a dg category is encoded, and whe
- ADR-0011 — Hom-complexes are complexes of abelia
- ADR-0012 — One Mathlib-style source root
- ADR-0013 — Own reusable moduli infrastructure in
- ADR-0014 — Stacks, algebraic spaces and moduli:
Roadmaps