Repository navigation
Architecture Abstraction Tree
This document is the repository contract for reusable abstractions. The
dependency contract in layers.md decides which imports are forbidden
between subjects. This document decides the finer question: when
two constructions share mathematical structure, where does that structure
live and how do the specializations relate to it?
Every reusable concept has one canonical root in its natural mathematical owner, identified by the placement procedure. Specialized geometry supplies instances, refinements, subobjects, quotients, or equivalences at the leaves. A leaf must not copy the carrier or fields of its root.
In Lean, "inheritance" means the narrowest of the following that expresses the mathematics:
- reuse an existing typeclass or structure directly;
- add a small capability typeclass whose parent is already an instance;
- use
extendswhen there is a canonical forgetful projection; - use an
abbrevfor a genuine specialization; - use a proved equivalence or agreement theorem for two independently useful presentations.
It does not mean putting a theorem, algebraicity claim, preservation result, or other unfinished conclusion into a structure field. It also does not mean forcing unlike formulas into one record merely because their carriers look similar.
This is the ownership target. Existing reverse edges are tracked defects to
burn down, not exceptions that authorize roots importing their consumers.
The diagram records construction and comparison relationships, not a chain
of Lean extends declarations or a list of completed implementations.
Current-versus-target paths are tracked in the cutover ledger.
ChargeFamily P N existing Wall namespace; no category required
├─ reindex / pullback change of chart / of class map
├─ zeroLocus δ charge vanishing; generally two real equations
├─ smul / phaseRotate β alignmentLocus_smul; a unit modulus fixes alignmentValue
│ └─ = phaseTiltRotation proved comparison, never a copy
├─ Wall.Exp.ofMoments m one polynomial, indexed by truncation m
│ ├─ scalar moments compressed H-degrees of a polarised n-fold
│ │ ├─ m = 1, m = 2 SlopeData.charge / stChargeFamily
│ │ ├─ m = 3 Threefold.chargeFamily
│ │ ├─ m = 4 free; no new polynomial
│ │ └─ (n,m) = (3,2) the tilt charge; Ku(cubic threefold)
│ │ rotated by 1/i ν is α · its unrotated chargeSlope
│ └─ form moments arbitrary divisor rank; no scalar compression
│ ├─ centralCharge Divisorial; proved instance of the kernel
│ ├─ StabilityParameters the (B,ω) chart; w = β + iα fixed here
│ ├─ mukaiCharge / κ κ is a pullback, never a coefficient
│ │ SqrtTodd the image of κ, not a truncation of it
│ └─ quadric, ℙ², blow-up rank-one slice reaches the scalar branch
├─ Wall.Exp.twist / discr the e^{-βH} action and Δ_H, stated once
├─ alignmentLocus v w determinant zero; no nonzero or sign claim
│ └─ positiveRayLocus v w nonzero charges plus a positive real scalar
└─ Spherical.nonpositiveRayLocus
a one-class signed-ray condition; never an
actual destabilization wall by terminology
positiveFrames Q ordered bases of positive planes
└─ forgetPositiveFrame explicit map to positivePlanes Q
└─ orthogonalityLocus δ planes orthogonal to δ; arrangement and
finiteness stay in neutral linear algebra
StabilityCondition.ChargeZero
└─ stabilityChargeFamily explicit comparison to ChargeFamily.zeroLocus
Bicategory Mathlib higher-categorical root
├─ Adjunction of 1-morphisms
│ ├─ adjoint equivalences and mates
│ └─ Cat specialization ≃ ordinary functor adjunction
└─ Pseudofunctor
├─ equivalence transport of Cat presentations
└─ fiberwise object properties / full subpseudofunctors
Category
├─ Limits and colimits
│ ├─ preservation through composition
│ └─ reflective transport using an ordinary adjunction
├─ Preadditive
│ ├─ Triangulated category Mathlib root
│ │ ├─ signed integral-shift exactness explicit CommShift + IsTriangulated
│ │ ├─ ObjectProperty.OnTriangle objectwise subcategory witness
│ │ │ └─ liftTriangle / liftTriangleMap
│ │ │ full-subcategory lift, comparison iso, functorial laws
│ │ ├─ Semiorthogonal decomposition
│ │ │ ├─ ExceptionalCollection
│ │ │ ├─ OrthogonalExceptionalBlocks positive blocks and residual
│ │ │ ├─ RightProjectionData chosen adjoint and universal Hom
│ │ │ └─ mutation / projection chains objectwise cones and iteration
│ │ └─ Stability conditions consume the independent charge core above
│ └─ Linear k C Mathlib root
│ ├─ linear Yoneda representability preimageIso / map_injective, no Serre datum
│ └─ SerreFunctorData duality on Hom spaces
│ · CategoryTheory/Linear/SerreFunctor/; the
│ · five entries below it are the shift-dependent
│ · refinements, under Triangulated/SerreFunctor/
│ ├─ SerreCategoryData chosen Serre autoequivalence
│ │ └─ EnriquesCategoryHypotheses square-to-shift refinement
│ ├─ IsSphericalObject two-degree self-Hom profile
│ ├─ IsPseudoprojectiveObject interval self-Hom profile
│ ├─ Ext profiles / transport Serre-compatible equivalences
│ └─ classification data supplied paper conclusions
├─ Abelian Mathlib typeclass
│ ├─ CochainComplex.homologyModel generic zero-differential model
│ │ └─ ModuleCat/division-ring formality
│ │ noncanonical and unbounded; no finiteness/naturality
│ ├─ CochainComplex.finiteCohomologyModel
│ │ ├─ finiteCohomologyModelHomologyIso homology preserves the finite biproduct
│ │ └─ finiteCohomologyModelIsoShifted degree i becomes degree zero shifted by -i
│ ├─ CochainComplex.FiniteCohomologyPresentation
│ │ ├─ pullback transport along an explicit HomotopyEquiv
│ │ ├─ isZero_homology_of_not_mem finite homology support consequence
│ │ ├─ homologyModelIsoFiniteCohomologyModel
│ │ │ generic finite-support comparison
│ │ ├─ ofFiniteSupport ModuleCat/division-ring constructor
│ │ └─ shiftedHomotopyEquiv presentation in the shifted normal form
│ ├─ weak-Serre exactness repository generic extension
│ └─ DerivedCategory C generic construction
│ ├─ ShortExact.singleTriangle Mathlib triangle construction
│ │ └─ map identity/composition neutral functoriality laws
│ └─ OppositeComparison C explicit derived/opposite bridge
│ └─ exact ModuleCat linear dual categorical specialization
├─ FiniteExactTower
│ └─ FiniteFiltration zero-to-object endpoint refinement
│ └─ almost-disconnected witness scheme-geometric leaf
├─ HasShift C A Mathlib root
│ └─ HasShift (X ⥤ Y) A pointwise, from the target's shift
│ └─ evaluation commutes strictly identity comparison, both laws free
├─ DGCategory C
│ ├─ IsShiftBy X n Y representable dg shift witness
│ │ └─ homIso dgHom W X ≅ (dgHom W Y)⟦-n⟧
│ ├─ DGLinear k C scalar refinement
│ │ ├─ homComplex existing Hom-complex repackaged in ModuleCat k
│ │ ├─ IsShiftBy.linearHomIso target shift as a shifted Hom-complex iso
│ │ │ ├─ H⁰ homComplex ≃ₗ Hom in H⁰ intrinsic quotient comparison
│ │ │ └─ Hⁿ homComplex ≃ₗ Hom in H⁰ target is the selected same-sign shift Y⟦n⟧
│ │ ├─ homComplexFiniteCohomologyPresentation
│ │ │ HomFiniteBounded support + field formality
│ │ ├─ postcompCochain fixed-source right composition, linear chain-map law
│ │ ├─ homFunctor k E Hom(E,-) as a k-linear dg functor to Cdg(ModuleCat k)
│ │ ├─ Cdg (ModuleCat k) standard model inherits Mathlib's k-linear structure
│ │ │ └─ linearTensorObj K X Mathlib total tensor product, same universe
│ │ │ ├─ tensorCochainLinearEquiv degreewise tensor--Hom adjunction
│ │ │ └─ HasLinearCopowers concrete scalar-linear copower instance
│ │ └─ IsLinearCopowerOf k K X Z represents k-linear cochains only
│ │ ├─ homComplexIso representing equivalence, compatible with differentials
│ │ ├─ coefficientMap homogeneous coefficient action, strict in composition
│ │ ├─ lift / lift_unique linear cochains are morphisms out of Z
│ │ ├─ compare / compareIso strict comparison, packaged as a Z⁰ isomorphism
│ │ ├─ scalar-unit witness single⁰(k) ⊗ X represents X itself
│ │ ├─ HasLinearCopower(s) Mathlib-style mere-existence capabilities
│ │ │ └─ linearCopowerFunctor Cdg(ModuleCat k) ⟶ C, k-linear dg functor
│ │ │ ├─ linearCopowerAdjunction (- ⊗ E) ⊣ Hom(E,-), strict at the dg level
│ │ │ │ ├─ unit selected universal copower chain map
│ │ │ │ └─ counit exactly selected scalar-linear evaluation
│ │ │ ├─ selected H⁰ invariance is a HomotopyCategory/DGEnhancement leaf
│ │ │ │ └─ LinearCopowerFiniteFree
│ │ │ │ ├─ basis expansion of a degree-zero copower
│ │ │ │ └─ finite-free finrank specialization
│ │ │ └─ finite presentation transport
│ │ │ ├─ finite biproduct of shifted degree-zero copowers in H⁰ C
│ │ │ └─ nested finite biproduct of shifts of X with finrank multiplicity
│ │ │ └─ K₀ class = homologyEulerChar • [X] for supplied finite-free data
│ │ └─ LinearEvaluationData k E scalar-linear Hom(E,-) ⊗ E assembly
│ │ ├─ functor k-linear dg functor
│ │ ├─ evaluation closed degree-zero map to the identity
│ │ ├─ compareIso coherent choice independence, strict over evaluation
│ │ ├─ IsEulerCopower generic H⁰ rank-one K₀ predicate
│ │ │ ├─ realization from supplied finite presentations
│ │ │ └─ automatic realization from HomFiniteBounded
│ │ └─ TwistConeData direct cone of scalar-linear evaluation
│ │ ├─ compareIso / exactness delegated to generic strict-square and cone APIs
│ │ ├─ twistTriangleFunctor H⁰ C ⥤ Triangle (H⁰ C), all distinguished
│ │ ├─ adjunctionTwistIso canonical Z⁰ identification with the copower--Hom counit cone
│ │ ├─ adjunctionTwistTriangleIso
│ │ │ coherent identification of the full H⁰ triangle functors
│ │ └─ K₀ action identity minus evaluation; HomFiniteBounded numerical twist
│ ├─ DGFunctor C D
│ │ ├─ HomogeneousNatTrans all degrees, differential, dg-functor category
│ │ │ ├─ IsClosed shared cocycle predicate for transformations
│ │ │ ├─ sourceShiftEquiv additive regrading through IsShiftBy.precompEquiv
│ │ │ ├─ whiskerLeft / whiskerRight both sign-free, additive, degree-preserving
│ │ │ │ ├─ interchange Godement, with the Koszul sign (-1)^(m n)
│ │ │ │ └─ hcomp the product itself; graded Leibniz, strict assoc
│ │ │ ├─ h0 closed degree zero becomes an ordinary NatTrans
│ │ │ │ ├─ composition strict dg composition descends to ordinary composition
│ │ │ │ ├─ whiskering descends through the canonical h0CompIso compositors
│ │ │ │ └─ h0Comparison H⁰(DGFunctor C D) ⥤ (H⁰ C ⥤ H⁰ D); no fullness claimed
│ │ │ │ └─ h0Iso Z⁰ isomorphisms descend, coherently
│ │ │ └─ ConeData functorial objectwise cones
│ │ │ └─ K₀ action target endpoint minus source endpoint
│ │ ├─ h0 / h0IdIso / h0CompIso functor descent with compositor associativity and unit laws
│ │ ├─ IsQuasiEquivalence Hom-complex quasi-isos plus essential surjectivity
│ │ │ └─ h0Equivalence induces an equivalence H⁰ C ≌ H⁰ D
│ │ ├─ shiftedFunctor n objectwise shift with the sign (-1)^(n p)
│ │ │ ├─ shiftedFunctorAdd / Zero degree coherence, closed and invertible
│ │ │ ├─ HasShift Z⁰(DGFunctor C D) ℤ standard Mathlib packaging, all unit/assoc laws
│ │ │ ├─ shiftedFunctor_h0_obj/_map the dg shift computes the H⁰ shift
│ │ │ ├─ shiftedFunctorH0Iso functor-level comparison; direct/transported CommShift-compatible
│ │ │ ├─ shiftedFunctorCompIsoIdH0 Mathlib opposite-shift cancellation descended to H⁰
│ │ │ └─ shiftedFunctorH0CommShift / IsTriangulated
│ │ │ transports exactness with the signed shift package
│ │ ├─ PreservesShifts free: every dg functor preserves shifts
│ │ └─ PreservesChosenCones free: mapped fst/snd split every image cone
│ │ ├─ ofIso invariant under Z⁰ dg-functor isomorphism
│ │ └─ h0CommShift / h0IsTriangulated automatic non-instance H⁰ exactness
│ │ └─ transportedH0 ordinary equivalence conjugate
│ │ ├─ CommShift / IsTriangulated composed from canonical packages
│ │ └─ Equivalence under explicit equivalence of H⁰ F
│ ├─ IsCopowerOf K X Z `Z = K ⊗ X`, by its universal property
│ │ ├─ lift / lift_unique cochains out of `K` are morphisms out of `Z`
│ │ ├─ compare closed canonical comparison, strict composition
│ │ ├─ HasCopower / HasCopowers Mathlib-style mere-existence capabilities
│ │ │ └─ copowerData a noncomputably selected CopowerData witness
│ │ ├─ HasEvaluationData E the narrower existence capability object twists need
│ │ │ └─ chosenEvaluationData a noncomputably selected EvaluationData E
│ │ └─ EvaluationData E selected directly or through HasEvaluationData
│ │ ├─ functor the dg functor `RHom(E,-) ⊗ E`
│ │ ├─ evaluation its degree-zero map to the identity
│ │ ├─ compareIso canonical Z⁰ isomorphism between any two choices
│ │ │ └─ compare_comp_evaluation strict compatibility with evaluation
│ │ └─ TwistConeData the object twist Cone(evaluation)
│ │ ├─ compareIso coherent Z⁰ iso across evaluation/cone choices
│ │ │ └─ inclusion compatibility is strict
│ │ ├─ twistTriangleFunctor H⁰ C ⥤ Triangle (H⁰ C), all distinguished
│ │ ├─ twistTriangleIso original same-evaluation cone-choice comparison
│ │ ├─ twistTriangleIsoOfEvaluation coherent across evaluation and cone choices
│ │ ├─ twistH0IsTriangulated automatic exactness; no caller-supplied cone witness
│ │ ├─ K₀ action generic cone formula: identity minus evaluation
│ │ └─ IsEulerCopower choice-invariant realization input for numerical twistK₀
│ ├─ DGAdjunction L R closed unit/counit plus triangle identities
│ │ ├─ CounitConeData counit-cone twist candidate
│ │ │ └─ twistAdjointComparison canonical closed `L T[-1] ⟶ R`, and its H⁰ map
│ │ │ └─ H⁰ inverse-rotation factorization
│ │ │ generic shiftedFstH0, right whiskering, and ordinary h0Counit
│ │ ├─ UnitConeData unshifted unit-cone cotwist candidate
│ │ │ ├─ cotwistAdjointComparison canonical closed `R ⟶ C L`, and its H⁰ map
│ │ │ │ └─ H⁰ unit/inclusion factorization generic h0Unit, right whiskering, and descended inr
│ │ │ └─ cotwistAdjointComparisonShiftedH0
│ │ │ canonical `R ⟶ (C[-1] L)[1]` normalization
│ │ └─ DGAdjunction.h0 an ordinary adjunction between the H⁰ functors
│ │ ├─ h0_whiskerLeft_counit generic left-whiskered counit normalization
│ │ ├─ h0_whiskerRight_unit generic right-whiskered unit normalization
│ │ └─ H0Presentation supplied endpoint isos after equivalence transport
│ │ ├─ presentedCounitTriangle generic normalization of the transported dg counit cone
│ │ │ └─ transportedTwist unchanged third vertex; pointwise distinguished
│ │ ├─ presentedUnitTriangle generic normalization of the transported dg unit cone
│ │ │ └─ transportedUnitCone unshifted third vertex; pointwise distinguished
│ │ ├─ presentedCotwistTriangle Mathlib inverse rotation of the presented unit triangle
│ │ │ ├─ transportedCotwist pointwise `[-1]` shift of the transported unit cone
│ │ │ └─ transportedCotwistH0Iso transport of the actual shifted dg cone agrees up to iso
│ │ │ ├─ CommShift compatibility canonical sign-correct packages agree across the iso
│ │ │ └─ exact equivalence exact for every cone under supplied triangulated transport;
│ │ │ equivalence from explicit H⁰ input
│ │ └─ Fourier--Mukai adapters existing left/right adjoint-kernel data
│ ├─ HomogeneousSquare arbitrary-degree vertical maps and homotopy
│ │ ├─ HomotopySquare degree zero with closed vertical maps
│ │ │ └─ strict a commuting square with zero homotopy
│ │ └─ IsConeOf.homogeneousLift all-degree cone maps, d/id/add/comp laws
│ │ ├─ IsConeOf.lift the degree-zero case, not a second owner
│ │ │ └─ IsConeOf.Morphism cone map with strict `inr` and `fst` squares
│ │ │ └─ ConePresentation category composes without a shift witness
│ │ ├─ IsConeOf.isoOfStrictSquare endpoint isos lift canonically to cone isos
│ │ └─ HomogeneousNatTrans.ConeData objectwise cones assemble to a dg functor
│ │ ├─ fst / snd cone projections, graded-natural
│ │ ├─ isConeOf a cone in the dg category of dg functors
│ │ ├─ isoOfStrictSquare the generic lift in Z⁰(DGFunctor C D)
│ │ ├─ preservesShifts automatic for the assembled dg functor
│ │ ├─ preservesChosenCones retained endpoint-based 3-by-3 witness; existence automatic
│ │ ├─ triangleFunctor H⁰ C ⥤ Triangle (H⁰ D), values distinguished
│ │ │ ├─ triangleNatTrans natural in a STRICT square of transformations
│ │ │ ├─ triangleIsoOfStrictSquare endpoint isos give a NatIso
│ │ │ └─ compareIso the cone choices do not matter, canonically
│ │ ├─ inverseRotateFirstH0 first maps of inverse-rotated cone triangles
│ │ ├─ shiftedFstH0 H⁰ of the source-regraded degree-one projection
│ │ ├─ shiftedFstH0_eq equality through shiftedFunctorH0Iso and packaged
│ │ │ `[1][-1]` cancellation
│ │ └─ DGAdjunction.CounitConeData counit-cone twist candidate
│ │ └─ twistTriangleFunctor H⁰ D ⥤ Triangle (H⁰ D), all distinguished
│ ├─ H0 C
│ │ └─ coneTriangleFunctor functor to distinguished triangles
│ └─ IsPretriangulated C
│ └─ Enhancement T comparison data, not a class
│ ├─ liftedCocycle noncanonical closed representative of an ordinary map
│ ├─ conePresentation representative plus a noncanonical dg cone
│ └─ coneTriangleFunctor dg cones read in `T` through the equivalence
├─ Triangle.FirstMapNormalizationData raw endpoint isos plus a named first-map square
│ ├─ normalizedTriangle literal first two vertices and literal first map
│ ├─ rawIsoNormalized natural triangle comparison; third component identity
│ └─ ComparisonData third-vertex iso + remaining squares over identity endpoints
├─ Fourier--Mukai correspondence
│ ├─ kernelTransform functor from kernels to transforms
│ ├─ kernelEvaluation one source object's kernel-variable functor
│ ├─ ExactFamily extends Mathlib CommShift; pointwise exactness only
│ ├─ ExactBifunctor extends Mathlib CommShift₂Int; Koszul-compatible
│ │ ├─ first/secondFamilyCommShift explicit adapters, evaluation agreement
│ │ └─ firstFamily / secondFamily exact-family projections through those adapters
│ ├─ kernelConeTransformTriangleFunctor pointwise image of dg cones of an enhanced kernel category
│ ├─ KernelConeNormalizationData FM cone/shift data plus generic-normalization adapter
│ │ ├─ firstMapNormalizationData categorical endpoint/first-map normalization
│ │ ├─ coneKernel ordinary image of the selected enhanced cone
│ │ └─ shiftedConeKernel enhanced shift with transform/shift comparison
│ └─ KernelTransformationData ordinary kernel arrow realizing a named transformation
│ ├─ KernelTransformationConeData noncanonical enhanced representative and dg cone
│ │ └─ normalizationData reusable literal endpoint/first-map transport
│ ├─ CounitKernelData equivalent adjunction-counit specialization; stable API
│ │ └─ CounitKernelConeData enhanced counit specialization
│ │ ├─ twistKernel ordinary image of the selected dg cone
│ │ ├─ kernel-presented twist candidate exact image of its dg cone
│ │ ├─ counitTriangleInSource literal counit triangle, pointwise distinguished
│ │ ├─ presented dg comparison pointwise/natural/shift-compatible layers
│ │ └─ conditional KernelAutoequivalence from explicit H⁰ equivalence + natural comparison
│ ├─ AdjunctionUnitKernelData definitional right-adjunction-unit specialization
│ │ └─ AdjunctionUnitKernelConeData selected enhanced unit arrow and dg cone
│ │ ├─ unshifted cotwist-cone candidate exact image of its dg cone
│ │ ├─ unitTriangleInSource literal unit triangle, pointwise distinguished
│ │ ├─ cotwist = cotwistCone⟦-1⟧ pointwise functor-category shift
│ │ ├─ cotwistKernel shifted enhanced cone presenting cotwist
│ │ ├─ cotwistTriangleInSource inverse rotation, pointwise distinguished
│ │ ├─ presented dg comparison pointwise/natural/shift-compatible layers
│ │ └─ conditional KernelAutoequivalence from explicit H⁰ equivalence + natural comparison
│ ├─ DualTwistKernelData left-adjunction unit specialization after swapping
│ │ └─ DualTwistKernelConeData reuses the unit/cotwist cone and normalization
│ │ ├─ dualTwistKernel shifted enhanced cone presenting dual twist
│ │ ├─ dualTwistTriangleInTarget inverse rotation, pointwise distinguished
│ │ ├─ presented dg comparison cotwist comparison reused after swapping
│ │ └─ conditional KernelAutoequivalence explicit H⁰ equivalence + natural comparison
│ └─ DualCotwistKernelData left-adjunction counit specialization after swapping
│ └─ DualCotwistKernelConeData reuses the counit/twist cone and normalization
│ ├─ dualCotwistKernel selected enhanced cone presenting dual cotwist
│ ├─ dualCotwistTriangleInSource literal counit triangle, pointwise distinguished
│ ├─ presented dg comparison twist comparison reused after swapping
│ └─ conditional KernelAutoequivalence explicit H⁰ equivalence + natural comparison
├─ Enhanced spherical-functor lane
│ ├─ EnhancedAdjunctionCones four adjunction-map cone choices
│ │ ├─ dualTwistFunctor / cotwistFunctor conventional shifted cone functors
│ │ ├─ cotwistH0Equivalence spends the unshifted condition after `[-1]`
│ │ ├─ H⁰ exactness automatic for every dg functor; no endpoint cone hypotheses
│ │ ├─ triangulated equivalences canonical Mathlib package for twist and cotwist
│ │ └─ K₀ action identity minus the corresponding adjunction composite
│ ├─ AdjointComparisonConditions IsIso on the two canonical H⁰ comparison maps
│ │ └─ cotwistShiftedIso conventional `R ≅ (F L)[1]` accessor
│ └─ TwistCotwistEquivalenceConditions explicit sufficient-condition input only
│ ├─ FM twist/cotwist adapters conditional KernelAutoequivalence, no new record
│ └─ full sphericality pending Morita/two-of-four theorem
├─ Derived-category extensions
│ └─ Ext adjunction / dimension shift / resolution naturality
├─ filtered-complex spectral sequences
├─ Moduli
│ └─ BoundednessProblem neutral boundedness predicate, the root until #1692
├─ GrothendieckPresentation
│ ├─ K₀Ab short-exact relations
│ ├─ K₀ triangle relations
│ │ ├─ finite biproduct calculus [⨁ Xᵢ] = Σ [Xᵢ], constant family = n·[X]
│ │ ├─ rankOne / IsRankOne factorization through ℤ; objectwise before exactness
│ │ ├─ Realization := K₀ C →+ A additive target
│ │ │ └─ Descends commuting realization square
│ │ └─ EulerForm := K₀ C →+ K₀ C →+ ℤ
│ │ ├─ ofLinear canonical linear-category pairing
│ │ └─ Preserves exact-functor compatibility
│ └─ K₀dg := K₀ (H0 C) reuse, not a third presentation
└─ Sites / descent / stacks in groupoids
├─ StackInGroupoids the one stack carrier (ADR-0014 §1)
│ ├─ fibres in `Cat.{v', u'}` target (#1631); `Cat.{w, w}` today
│ ├─ discrete, representable stacks on subcanonical sites (STK2.1)
│ ├─ pointwise representability `FiberRepresentation`: one equivalence per
│ │ test object, no naturality; `Pointwise…` (#1626)
│ ├─ natural representability Tag 02Y4; target (#1636), with a proved
│ │ projection to its `Pointwise…` predecessor
│ └─ categories fibred in groupoids proved comparison only; target (#1638)
├─ over-site functoriality
│ └─ cocontinuity of `Over.post`
├─ `CoversTop` equivalence transport
├─ Sheaves
│ ├─ coverwise detection of local equivalences
│ ├─ constant pullback and cohomology pushforward
│ └─ module sheaves on a ringed site
│ ├─ restriction to an over site
│ ├─ exact forgetful functor
│ ├─ finite-presentation transport
│ │ ├─ isomorphism invariance / cover locality
│ │ └─ zero objects / short-exact extensions
│ └─ intrinsic IsInvertible
│ ├─ rank-one local trivializations
│ ├─ arbitrary-site tensor descent
│ └─ topological specialization
│ ├─ stalk/tensor comparison
│ └─ stalkwise arbitrary-factor strengthening
├─ site-theoretic Čech complexes and derived comparison
│ └─ compact-basis and finite-cover boundedness
└─ scheme-site realizations
LinearAlgebra
├─ finite free integral lattices Mathlib: [Module.Finite ℤ Λ] [Module.Free ℤ Λ]
│ └─ NumericalVarietyData.NumericalQuotient finite by Module.Finite.quotient; free, once
│ torsion-free, by Module.free_of_finite_type_torsion_free'
├─ weighted-basis graded pieces
│ ├─ internal direct-sum decomposition
│ └─ NumericalRingData.ofGradedBasis geometric numerical consumer
├─ paired complex functional
│ └─ PeriodDomain.centralCharge ComplexPairing.lean; ker Z = plane ⊥
│ ├─ stability wall adapter CentralCharge/Quadratic.lean
│ ├─ kernel negativity one half of quadratic support
│ └─ Mukai.expCharge existing exponential presentation
├─ lattice root (#1230) WITHDRAWN 2026-09-30, never landed
│ └─ Lattice.pairCharge expCharge already specialises centralCharge
├─ graded pairing (#1230) WITHDRAWN: no consumer; odd form refuted
│ └─ n = 2 must compare to realPairing never a second definition
└─ Mukai.pairing / selfPairing Lattice/Mukai/Basic.lean:56,143
ARITY IS FIXED AT THREE generalise over the coefficient ring only
root of the discriminant pairing 3 v v = 0 identically
Algebra
├─ ordinary ring and module theory
│ └─ module localization
│ └─ kernel maps consumed by coherent-sheaf geometry
├─ numerical polynomials on integer lattices
│ ├─ mixed finite differences and degree
│ └─ top multilinear coefficients
│ └─ Snapper polynomiality geometric Picard/coherent consumer
├─ multivariate polynomials
│ └─ division by monomials consumed by projective localization
├─ saturation of an additive subgroup
│ └─ saturated quotient and torsion-free universal property
└─ relative numerical algebra
├─ indexed sums and saturated family-relation quotients
└─ additive-map images and finite-index-overlattice predicates
AlgebraicGeometry
├─ SheafOfModules(X)
│ ├─ QuasicoherentSheaf(X)
│ │ └─ CoherentSheaf(X)
│ │ ├─ Abelian instance under geometric hypotheses
│ │ └─ DerivedCategory (Coh X) consumes the generic construction
│ ├─ sheafified tensor / MonoidalCategory X.Modules
│ │ ├─ intrinsic IsInvertible as ObjectProperty.IsMonoidal
│ │ │ ├─ InvertibleSheaf inherited full monoidal subcategory
│ │ │ └─ pullback preservation theorem inherited by every consumer
│ │ ├─ Picard classes consume intrinsic IsInvertible
│ │ └─ Functor.Monoidal pullback input standard tensor/unit comparison surface
│ └─ LineBundleData invertible sheaf plus chosen tensor inverse
│ ├─ determinant and Picard interpretations
│ ├─ monoidal pullback and projection formula
│ ├─ CartierDivisor.lineBundleData associated-sheaf/Picard agreement leaf
│ │ ├─ coherent, derived, bounded-derived canonical objects
│ │ └─ effective-divisor sequences twisted by arbitrary line bundles
│ └─ almost-disconnected graded pieces scheme-geometric leaf
├─ scheme-derived category `DerivedCategory/`
│ ├─ Dqc neutral geometric locus
│ │ ├─ canonical zero owned by Dqc, all schemes
│ │ └─ explicit comparison evidence representatives, not instances
│ ├─ bounded coherent locus consumes generic triangle lift
│ │ ├─ Cartier-divisor objects consume canonical line-bundle bridge
│ │ └─ short-exact triangles/maps includes arbitrary line-bundle twists
│ ├─ scheme pullback and geometric kernels
│ └─ absolute perfect locus thick envelope in `D(Coh X)`
│ └─ essential image in Dqc
│ └─ compact objects only with explicit evidence
├─ relative-perfect locus over `p : X ⟶ S` `DerivedCategory/Perfect/Relative`
│ ├─ pseudo-coherent + finite Tor amplitude not absolute perfection by definition
│ └─ stalkwise flat module sheaf over `p` `Modules/Flat`, needs no derived category
├─ two-term determinant presentation explicit finite-locally-free resolution
│ └─ absolute perfect degree-zero object proved comparison adapter
├─ numerical K-theory
│ ├─ Euler quotient
│ │ └─ future scheme-specific relation generators consume Algebra root
│ ├─ Riemann--Roch and Mukai transfer
│ │ └─ consume categorical K₀ realizations and Euler forms
│ └─ Polarised.wallChargeFamily one transport for every (n, m, κ)
│ ├─ n = 2, 3 exist; n = 4 free K3, ℙ², quadric, ℙ³, quintic, ℙ⁴, sextic
│ ├─ κ = √td is inhabited = mukaiCharge at sqrtTodd on rankOne
│ │ any surface no K3 hypothesis is needed
│ └─ κ = 1 and κ = √td two pullbacks; different walls, not one
├─ moduli prestacks in groupoids; carrier `StackInGroupoids`
│ ├─ fiberwise replete locus selector not a subprestack
│ ├─ finite-type boundedness witness consumes selector + generic predicate;
│ │ a pointwise cover, it does not bound (#1626)
│ ├─ affine stable subprestack consumes pseudofunctor object property
│ ├─ stack presentation pointwise (Zariski, fppf); not an algebraic stack
│ └─ perfect-complex specialization pointwise presentations only
└─ stability-consuming children `DerivedCategory/Stability/`, `Moduli/`,
`Numerical/`, `Stability/` — four, not one
The arrows implied by this tree point downwards from consumers to roots. In particular:
- do not introduce a second adjunction hierarchy: reuse Mathlib's
Bicategory.Adjunction, and recover ordinary adjoint functors through the bicategoryCat; place results that merely assume an adjunction with their actual conclusion, such as limit preservation or derivedExt; - do not add
KLinearCategory; use Mathlib'sPreadditiveandLinearand bridgeDGLinearto them; - do not add a coherent, derived, dg, projective-space, or relative sibling of the Grothendieck group presentation; specialize the canonical presentation and prove comparison maps;
- represent additive K₀ targets and categorical Euler pairings by the canonical
K₀.RealizationandK₀.EulerFormaliases. Their descent and preservation laws stay with the triangulated Grothendieck-group root; geometricIsRiemannRoch, Euler-transfer, and Mukai-pairing theorems consume them; - keep indexed additive-group sums, additive-subgroup saturation,
family-relation systems, additive-map images, and finite-index-overlattice
predicates in
Algebra/RelativeNumerical; a geometric consumer must introduce actual schemes, connectivity, relative perfection, or other geometric data and import this root directly; - keep integer-lattice numerical functions, mixed forward differences,
finite-difference degree, Newton coefficients, and top multilinear
coefficients in
Algebra/NumericalPolynomial; Snapper's theorem imports that root and adds the Picard, coherent-sheaf, and Euler-characteristic data; - do not add a class for finite free abelian groups: Mathlib's
Module.Finite ℤandModule.Free ℤinstances are the interface, and freeness of a finitely generated torsion-free group is Mathlib's instanceModule.free_of_finite_type_torsion_free'; the numerical Euler-radical quotient supplies those hypotheses and nothing else; - keep ordinary module theory under
Algebra, generic sheaves underCategoryTheory/Sites/Sheaves, module sheaves on ringed sites underAlgebra/Category/ModuleCat/Sheafwhere Mathlib definesSheafOfModules, and only the scheme-indexedSheafOfModules(X) -> QCoh(X) -> Coh(X)refinements under algebraic geometry; in particular, coherent-sheaf kernels directly reuse the module-localization kernel maps rather than owning them; - keep detection of additive-presheaf local equivalences on a
CoversTopfamily under the generic site/sheaf root; scheme charts and sheafified tensor constructions only consume that theorem chain; - keep cocontinuity of
Over.post, module-sheaf restriction to over sites, and equivalence transport ofCoversTopfamilies under the generic site and sheaf roots; open-immersion geometry consumes them only after introducing schemes and their open-set sites; - keep restriction of presentations, generating sections, and quasicoherent
presentation data under the generic ringed-site presentation root; affine
geometry consumes that API only when it introduces
Spec Rand basic opens; - keep isomorphism invariance, over-site preservation, and covering-family
descent of finite presentation under the same generic presentation root;
coherent-sheaf geometry consumes them only for
coherent Xand affine-open criteria; - keep stalk/tensor comparisons for module presheaves under
Topology/Sheaves/ModuleTensor/: their use of open neighbourhoods, germs, and stalk functors makes them topological specializations of arbitrary-site sheaf theory, not ordinary module algebra; - keep the braided transfer between the two sheafification whiskering
comparisons under
Algebra/Category/ModuleCat/Sheaf/Tensor.lean: conjugatingM ◁ toSheafifyintotoSheafify ▷ Muses only symmetry of the objectwise presheaf tensor, so it holds on an arbitrary site and must not be restated inside the scheme consumer that supplies its stalkwise hypothesis; - put the sheafified tensor and its coherence on all scheme-module sheaves;
invertible sheaves inherit that root through
ObjectProperty.IsMonoidal, and pullback monoidality uses Mathlib'sFunctor.Monoidalrather than a parallel capability class; - keep basiswise detection of topological sheaf isomorphisms under
Topology/Sheaves/Basis.lean; affine comparison consumes it only after introducingSpec R, distinguished opens, and module localization; - keep identities involving the lattice and categorical products of
prime-spectrum opens under
RingTheory/Spectrum/Prime/, where Mathlib definesbasicOpen; neither the ring input nor the topological output moves them; - extend Mathlib's abelian-category hierarchy under
CategoryTheory/Abelianand make the geometric proof thatCoh(X)is abelian an input to the generic derived-category construction; - generic stacks, replete subprestacks, and boundedness predicates do not import stability-condition or geometric modules; scheme presentations and finite-type witnesses remain geometric consumers;- a paper-specific module may instantiate these roots but never becomes a root imported by them.
- keep exceptional blocks, residual right orthogonals, chosen right-adjoint
projections, objectwise mutation and projection-chain arguments, Serre
duality, Enriques-category relations, and spherical or pseudoprojective
self-Hom profiles and transport under the generic triangulated-category
root. Kernel-presented exceptional-block extension remains under the
Fourier--Mukai root. A surface-specific residual category reuses those
structures and binds the candidates to projections of its actual line
bundles; the numerical curve chain is geometric, its derived triangles and
successive block maps consume the reusable Cartier-sequence roots, and only
quotient orthogonality remains an explicit geometric input. Functorial dg
cones and their pointwise transform triangles now have generic roots, and
the kernel category enters through an
Enhancementrather than as anH⁰on the nose; the enhancement of the geometric kernel category with its exact comparison, the actual kernel morphism, and exactness of kernel evaluation remain explicit realization inputs. - keep abstract charges and their geometric constructors distinct. An abstract
charge is an additive map
Λ →+ ℂ; a complex parameter, Chern character and Todd correction are inputs to particular constructors, not fields required of every charge. Reuse the existing exponential kernel for the supported surface, threefold and slope families at their respective truncation degrees. Truncation degreemis the index, never the ambient dimensionn: the cubic-threefold tilt charge is(n, m) = (3, 2), so a root indexed byncannot state it. A new dimension supplies aPolarizationand inherits the polynomial; if a lane finds itself writing a fourth charge polynomial, the placement is wrong. - a correction class
κis a pullback, never a coefficient.κ = 1andκ = √tdgive genuinely different walls, so they are two pullbacks of one transport and must not be fused into a single family.SqrtToddis the image ofκunder a numerical realization, not a truncation of it. - the Mukai pairing has fixed arity three. Generalise it over the coefficient
ring, never over dimension: a dimension-indexed self-pairing vanishes
identically in odd degree, so it cannot be the root of the Bogomolov
discriminant. The graded pairing is a separate object that declares nothing new
at
n = 2, where it must compare withMukai.realPairing, using the actual coordinate equivalence and normalization. Fin-indexed functions and nested products must not be assumed definitionally equal. - do not add a
KuznetsovChargeDatacarrier. Reuse the abstract charge or pairing root. An ambiently induced charge reaches the component by restriction along its inclusion; a charge constructed on its own numerical lattice must name that realization instead. For the cubic threefold the charge is induced from the rotated tilt charge. Its odd-degree compressed pairing is alternating and cannot supply a positive-plane quadratic domain; this says nothing about other period-domain constructions for the variety or component. For the cubic fourfold the charge lives on the algebraic Mukai lattice while the period domain lives on the orthogonal complement of anA₂sublattice, and those are two different lattices that must not be fused.
Bicategories are the first implemented higher-categorical stage. A future
general n-category or (∞,1)-category layer must name its formal model and
its comparison with this spine; an empty directory does not establish an
abstraction relationship.
Clause 6 below requires a falsified generalization to be recorded rather than forgotten. These were each established against the tree, and each one is a unification that looks right in the literature and is false here.
-
A dimension-indexed Mukai self-pairing is not the root of the discriminant.
It vanishes identically in odd degree, so the threefold leaf cannot reach it.
The root is the fixed-arity pairing generalized over its coefficient ring.
The winner landed with
#1223:Mukai.pairingwas generalized in place over[CommRing R] [Module R M], keeping arity three and its public name, and all seven discriminants — two on wall classes, two divisorial, oneA-valued on a numerical variety, and the surface and threefoldH-compressions — reach it by a*_eq_selfPairingprojection. Three of those seven areℚ- orA-valued, which is the whole reason the ring had to move and the dimension did not. -
The multi-divisor charge does not factor through the compressed H-degrees.
The compression is not injective once the Picard rank exceeds one, so the
intersection-form branch is a genuine sibling of the scalar branch and not a
specialization of it.
The same obstruction applies to twisting by an arbitrary divisor
B: on the plane with formdiag(1,-1), takeH=(1,0)andB=(0,1). The triples(0,(0,0),0)and(0,(0,1),0)have identical compressedH-degrees, but theirB-twisted top coordinates are respectively0and1.ChernCharacter.twist_rankOne_eq_expcompares the validB=βHslice;ChargeCoordinates.twistByScalar_eq_expincludes the weightH²on rank. -
A nonsymmetric lattice charge root is not worth a name (#1230, withdrawn).
Mukai.expChargeis by definitionPeriodDomain.centralCharge (realForm b)at the exponential pair, so the two charges were already one root. Neither leaf's API uses symmetry ofb, which means dropping it generalizes nothing they need, and for nonsymmetricbthe identityexpCharge = pairCharge (realPairing b)is false:polar (realForm b)is the symmetrization. Under clause 2a the remaining content was a thin Lift (add, smul, zero, kernel) with no consumer asking for it. The one hand-rolled copy of the form,cPair, is symmetric and may be re-expressible throughcentralCharge(#1810). Recorded in the generalization backlog. -
The graded pairing needs no second root at
n = 2. Its comparison withMukai.realPairingmust exhibit the actual coordinate identification and normalization. A function space and a product carrier are not thereby definitionally equal. Prove the bridge rather than add a duplicate form. -
Parity of the compressed pairing says nothing about a Kuznetsov component.
The compressed form lives on a space of dimension
n + 1; the Mukai lattice of the component does not. Symmetry there is automatic for an unrelated reason, so the parity argument is vacuous even on the correct space. - The rotation that induces the cubic-threefold charge is not cosmetic. It leaves every wall fixed but changes phases, so a construction that drops it is wrong about semistability even where it is right about walls.
- The tilt charge is not a chart change of the threefold charge. Their wall loci genuinely differ; the tilt family is the surface family pulled back along a truncation, which is a different operation.
- A geometric Serre functor is not a new structure. The categorical root already exists and the geometric side owes it a bridge; a parallel structure fails the adoption clause below and would add a third duplicate twist.
- The Fourier--Mukai convolution classes are not pseudofunctor data. They resemble compositor and unitor data without being an instance of it, so that lane cannot own them and the resemblance must not be used to place them.
Every issue or pull request that introduces a public structure, class,
quotient carrier, or category must answer these questions before implementation.
-
Canonical owner. What existing root is closest? Give its declaration and module. If none exists, name the proposed neutral module.
-
Adoption. Name two independent consumers, or say explicitly that this is statement-layer data whose purpose is to compare multiple inhabitants. When the structure is being extracted from an existing module, that module does not count as one of the two: an extraction that only its own former home consumes is a rename, and belongs in that home as a theorem or an
abbrev. This is what keeps an ownership repair from producing an empty hierarchy.This clause governs carriers only. It does not govern a theorem or definition restated at weaker hypotheses. A proved weakening is its own adoption argument and needs no second consumer before implementation: for those, apply the proof-witness test below. 2a. Proof-witness test. Before applying clause 2, decide which of three things is being proposed. The compiler decides, not taste.
-
Lift. The general statement compiles at the ancestor module with a
binder/typeclass set strictly weaker than the leaf's, and the leaf is
re-derived from it in one line (
instance,abbrev, or:= general …). Both artifacts appear in the same commit. → Accept. Consumer count is not asked. The former home counts, because a proved weakening is not a rename. -
Carrier with content. A new carrier, and at least one non-
sorrytheorem in the same commit whose statement mentions that carrier and is not expressible without it. → Judge by clause 2 and questions 3, 4, 5. A theorem compiling at the new altitude counts as one of the two consumers. -
Empty carrier. A new carrier with zero such theorems in the commit.
→ Blocker. This is the only kind of abstraction this repository
refuses on sight, and it is exactly the
wontitem recorded in.claude/roadmap/mathematical-ownership.yaml.
Generality must arrive carrying a proof, not a promise. Equally, a statement that assumes a hypothesis its own proof never uses is a defect, not a matter of taste — see question 7.
-
Lift. The general statement compiles at the ancestor module with a
binder/typeclass set strictly weaker than the leaf's, and the leaf is
re-derived from it in one line (
-
Projection. How does a specialization forget to, refine, or compare with the root? The answer must be an existing instance, a projection, an
abbrev, or a theorem—not prose. -
Diamond. If both the root and leaf synthesize inherited instances, add a compile-time or equality test showing that the paths agree.
-
Dependency direction. Confirm that the root imports no specialization, realization or paper-specific consumer, including through umbrellas. Keep agreement theorems downstream of both presentations. State the inspected import closure, not just the absence of one direct import.
-
Negative result. If the apparent generalization is false, record the counterexample and keep the leaves separate. A falsified unification is a successful architecture result. Record it in the generalization backlog as
FALSIFIEDwith its counterexample rather than deleting the question; the next milestone should not have to rediscover it. -
Altitude of the parent. State the hypotheses the proposed root actually needs — read the proof, not the signature. If the existing parent assumes more than its own subtree requires, record that as an altitude finding against the parent rather than hanging a new leaf beneath it. This tree is a target, not an oracle: a root sitting one layer too low is a defect like any other, and the ceiling is allowed to rise.
Use the ownership decision record
to connect these answers to the issue or PR. Moving declarations is not
complete until imports, umbrellas, audit routing, registry/source-owner
bindings, documentation and relevant checks are updated together. Preserve
required historical names through the existing executable-only mechanism;
retired-path import shims are forbidden. See CONTRIBUTING.md.
For the 2026-09-13 ownership review the answers to questions 1, 3 and 5 are
already recorded per finding in the owner map in
docs/architecture/cutover-ledger.md; a pull request implementing one of those
rows cites the row instead of re-deriving it, and answers question 2 there or
drops the proposed carrier.
If the target already carries the structure being constructed, agreement is an acceptance criterion, not follow-up cleanup. The dg enhancement of the homotopy category is the model: its transported shift and distinguished triangles are compared with Mathlib's existing instances. The same rule applies to exact versus triangulated K-groups, ordinary versus dg linearity, and classical versus derived moduli truncations.
The policy is partly mechanical and partly a review obligation:
-
scripts/check_layering.pyenforces the policy edges inlayers.md; -
scripts/check_neutral_stability_imports.pychecks the neutral derived-category umbrella's full internal import closure against the two Stability trees; -
scripts/check_umbrella_coverage.pykeeps every specialization in the public tree; -
scripts/check_single_instantiation.pychecks its configured paths for thin abstractions; the clause 2 adoption obligation for carriers still applies outside its scan; -
scripts/check_roadmap.pykeeps materialized lanes synchronized with their tracker issues; - the pull-request template and the ownership decision record capture the non-mechanical root, projection and agreement decisions. Component import boundaries not yet encoded by a gate remain review obligations, with checks added in their source cutovers.
The projective-families roadmap carries the first finer-grained burn-down: generic stack roots move out of scheme geometry, moduli-to-stability reverse imports are removed, and geometric relative numerical K-theory consumes the indexed quotient machinery rooted in ordinary algebra.
Exported from docs/architecture/abstraction-tree.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