Skip to content

Roadmap Dg Enhancements

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

Dg Enhancements

schema: roadmap/1
slug: dg-enhancements
project: derived-alg-geo-lean
title: "DG enhancements — the H⁰ seam between dg categories and classical derived categories"
status: active
phase: sequenced
brief: |
  Open a second track alongside `stability-families`. Where that track climbs
  from an absolute Bridgeland foundation toward relative moduli, this one
  builds the *other* axis the repository's name promises: the passage between
  a triangulated category and a dg category enhancing it.

  The programme this track reads from is Canonaco–Stellari and
  Canonaco–Neeman–Stellari, fifteen years of work on existence and uniqueness
  of dg enhancements. Its motivating claim — that dg categories are the
  correct ambient objects because naturality of functors falls out of the
  definitions rather than being repaired afterwards — is exactly the kind of
  claim a formalization can test rather than repeat. This roadmap therefore
  treats "the dg side is easier" as a **hypothesis with a falsifiable check**
  (see dg-enhancements-m4 acceptance), not as a premise.

  What the pinned environment already supplies is unusually favourable and is
  the reason this track is filed now rather than parked. One item on the list
  below did not survive measurement: the monoidal structure on complexes does
  NOT reach `CochainComplex (ModuleCat k) ℤ` at the pin, so the enriched
  encoding is unavailable and ADR-0010 decided for a bespoke structure. See the
  `assumptions` block and https://github.com/chris-dare-dev/derived-alg-geo-lean/blob/965d62fc66afbbe998482ef0912cacc265a01c5a/.claude/notes/2026-08-13-dg-surface-reconnaissance.md.
  At `mathlib_rev` in `pins.json`, Mathlib has: the Hom cochain complex between cochain complexes
  (`Algebra/Homology/HomotopyCategory/HomComplex`, with its shift and
  cohomology files) — which is the dg structure on complexes, already built;
  the monoidal structure on homological complexes (`Algebra/Homology/Monoidal`);
  enriched and enriched-ordinary categories (`CategoryTheory/Enriched`);
  the homotopy category with its pretriangulated and triangulated instances;
  K-projective/K-injective objects; and `DerivedCategory` with a t-structure,
  `Ext`, and fractions. What Mathlib has **none** of is a dg category, a dg
  functor, a quasi-equivalence, or an enhancement. That gap is this track's
  subject, and DG0 measures it declaration by declaration before anything is
  built on the estimate above.

  The track deliberately stops short of the theorems the source papers are
  famous for. Uniqueness of enhancements, intrinsicness of subcategories of
  weakly approximable triangulated categories, and internal Homs in Hqe all
  need machinery — Bousfield localization of dgCat, model structures, Brown
  representability at scale — that does not exist at the pin. Those enter as
  *statements over the formalized definitions* with no inhabitant, on the same
  zero-claim discipline `registry/coverage-*.json` already enforces.

goal:
  hmw: "How might we make a dg enhancement a first-class object in this repository, and make the passage H⁰ between a dg category and a classical triangulated category a proved equivalence rather than an informal identification?"
  objective: "Formalize dg categories, dg functors, H⁰, and enhancements at the pin; prove that the h-injective dg category of complexes enhances Mathlib's DerivedCategory; and expose a transport API that carries t-structures, slicings, and stability conditions across that equivalence in both directions."
  key_results:
    - "A dg category and a dg functor are definable at the pin with k-linearity explicit, no new axiom, and no sorry."
    - "H⁰ is a functor, is additive, and satisfies H⁰(C^dg A) ≌ HomotopyCategory A — proved, not asserted."
    - "A pretriangulated dg category transports a triangulated structure to its H⁰, and on K^dg(A) the transported structure is proved to agree with Mathlib's existing instance rather than merely to exist."
    - "DerivedCategory A has a constructed, non-vacuous enhancement, compatible with shift, distinguished triangles, and Ext."
    - "A t-structure, a slicing, and a pre-stability condition each transport across an enhancement equivalence with round-trip identities proved, so a Bridgeland-side statement can be made on either side of the seam."
    - "Every uniqueness-of-enhancements statement is a Prop over the formalized definitions with no inhabitant, and every source coordinate stays `target` until reviewed evidence promotes it."
  assumptions:
    - tier: must
      text: "A dg category is expressible at the pin as a category enriched over cochain complexes of k-modules, using CategoryTheory/Enriched/Ordinary together with Algebra/Homology/Monoidal."
      status: refuted
      validation: "REFUTED 2026-08-13 against mathlib_rev. `HomologicalComplex.HasTensor` does not synthesize for `CochainComplex (ModuleCat k) ℤ`; the degree-n tensor is a coproduct over the infinite set {(i,j) : i+j=n} and only the finite-fibre instances exist. `ChainComplex (ModuleCat k) ℕ` IS monoidal, which localizes the gap. Evidence and probes: https://github.com/chris-dare-dev/derived-alg-geo-lean/blob/965d62fc66afbbe998482ef0912cacc265a01c5a/.claude/notes/2026-08-13-dg-surface-reconnaissance.md. ADR-0010 Question 1 therefore decided for Option B (bespoke DGCategory); the enriched route survives as Option A′, an upstreamable ForMathlib slice that blocks nothing in DG1."
    - tier: must
      text: "Mathlib's HomComplex API is complete enough to give C^dg(A) without re-deriving the Hom complex."
      status: confirmed
      validation: "CONFIRMED 2026-08-13, and stronger than stated: besides `Cochain`, `Cochain.comp`, `δ`, `Cocycle` and `CochainComplex.HomComplex`, the pin already proves `CohomologyClass.homAddEquiv : Hⁿ(Hom•(K,L)) ≅+ Hom_{K(C)}(K, L⟦n⟧)`. The seam theorem of e4 is packaging, not construction. Evidence: https://github.com/chris-dare-dev/derived-alg-geo-lean/blob/965d62fc66afbbe998482ef0912cacc265a01c5a/.claude/notes/2026-08-13-dg-surface-reconnaissance.md."
    - tier: should
      text: "The h-injective (respectively h-projective) model of the derived enhancement is reachable before the Drinfeld dg quotient is."
      validation: "Mathlib already builds DerivedCategory through KInjective/KProjective. If the seam theorem cannot be closed on that model, the Drinfeld quotient epic un-parks and the milestone slips rather than the claim weakening."
    - tier: should
      text: "Transport across an enhancement equivalence makes at least one functor-naturality obligation cheaper than its classical proof."
      validation: "dg-enhancements-e12 must exhibit one such lemma with both proofs, or record the negative result in the roadmap and in the epic. A negative result retires the m4 rationale; it does not get quietly dropped."
  wont:
    - "No model structure on dgCat, no Tabuada machinery, no Bousfield localization of dg categories."
    - "No ∞-categorical enhancement, no comparison between dg and stable-∞ enhancements."
    - "No claim that a triangulated category not constructed in this repository admits an enhancement, and no existence claim from a uniqueness statement."
    - "No axiom, sorry-backed instance, or vacuous structure to make a source definition typecheck — the repository rule stands unchanged for this track."
    - "No replacement of Mathlib's DerivedCategory. The seam theorem compares the two objects; it does not fork one of them."
    - "No modification to formalization.yaml's `source` and no new key minted outside registry/bridgeland2007.json."
  evidence:
    - "arXiv:1507.05509v5 — Canonaco–Stellari, 'Uniqueness of dg enhancements for the derived category of a Grothendieck category', JEMS 20 (2018) 2607–2641."
    - "arXiv:2101.04404v1 — Canonaco–Neeman–Stellari, 'Uniqueness of enhancements for derived and geometric categories'."
    - "arXiv:2402.04605v1 — Canonaco–Haesemeyer–Neeman–Stellari, 'The passage among the subcategories of weakly approximable triangulated categories'."
    - "arXiv:2407.05946v2 — Canonaco–Neeman–Stellari, survey, to appear in Boll. Unione Mat. Ital."
    - "arXiv:1312.5619v3 — Canonaco–Stellari, 'Internal Homs via extensions of dg functors', Adv. Math. 277 (2015) 100–123."
    - "Pinned Mathlib surface: Algebra/Homology/HomotopyCategory/HomComplex{,Shift,Cohomology,Single,Induction}, Algebra/Homology/Monoidal, CategoryTheory/Enriched/{Basic,EnrichedCat,Ordinary/Basic}, Algebra/Homology/DerivedCategory/{Basic,TStructure,KInjective,KProjective,Ext/*}, CategoryTheory/Localization/*, CategoryTheory/Triangulated/*."
    - "Absence measured, not assumed: no declaration matching DifferentialGraded/DGCategory/quasi-equivalence exists at the pin. DG0 re-runs this search against `mathlib_rev` exactly and records the command and its output."
    - "ADR-0012 supersedes the temporary DGLean root: dg-category work now lives under DerivedAlgGeo/CategoryTheory/DGCategory and uses the CategoryTheory.DGCategory namespace."

generated_by: { agent: claude, at: "2026-08-13" }
generations:
  - { rev: 1, at: "2026-08-13", note: "Track proposed. Nothing materialized: every gh_milestone and gh_issue is null, and no label, milestone, issue, or project view exists yet. Materialization is dg-enhancements-e0." }
  - { rev: 2, at: "2026-08-13", note: "Partially materialized: eight labels created, milestones #19-#24 created, and the two DG0 epics filed as #328 and #329. DG1-DG5 epics remain unfiled by design. Project views and fields are still pending the project scope on the gh token." }
  - { rev: 3, at: "2026-08-13", note: "Source pins landed: five zero-claim coverage maps (1312.5619v3, 1507.05509v5, 2101.04404v1, 2402.04605v1, 2407.05946v2), each hashed against the PDF as served on 2026-08-13 and with section titles read from that PDF. registry/README.md documents all five. formalization.yaml and registry/bridgeland2007.json untouched." }
  - { rev: 4, at: "2026-08-13", note: "ADR-0010 Question 1 decided: Option B, a bespoke DGCategory structure, after the pinned-Mathlib measurement refuted the enriched route's enriching category. Assumption 1 marked refuted, assumption 2 confirmed and strengthened, e2 resized up and e4 down. DG1 is unblocked." }
  - { rev: 5, at: "2026-08-13", note: "DG1 epics filed as #335 (e2), #336 (e3), #337 (e4). e2 is agent-ready; implementation started on branch agent/dg-e2-dgcategory." }
  - { rev: 6, at: "2026-08-14", note: "Root trigger corrected to match what happened. #343 (e2) created DGLean/ with fifteen proved lemmas and full gate wiring, two epics ahead of the e4 trigger this file named. ADR-0010's Question 2 is amended rather than this file being the primary move: its trigger was a prediction, and its citation of ARCHITECTURE.md growth rule 1 was wrong twice over -- that rule is about module placement, and the 'first real theorem' sentence is CLAUDE.md's. ARCHITECTURE.md gains growth rule 6 and a DGLean row. m1 acceptance gains a namespace clause: e2 landed the directory but not the DGLean namespace, and every declaration is still at the root." }
  - { rev: 7, at: "2026-08-14", note: "DG0 and DG1 completed. Issues #328, #329, #335, #336, and #337 are closed; milestones #19 and #20 are closed. ADR-0012 supersedes the temporary DGLean root and the implementation now lives under DerivedAlgGeo/CategoryTheory/DGCategory with namespace CategoryTheory.DGCategory." }
  - { rev: 8, at: "2026-08-14", note: "Landing evidence recorded per epic, in the summaries, matching how stability-families.yaml records a closed epic: e0 and e1 in #330, e2 in #343, e3 in #354, e4 in #370. One correction to rev 7. It marked m1 done citing ADR-0012's namespace, but #372 moved the files without renaming the declarations, so every dg declaration was still at the root when that was written; the clause is satisfied by #374, which wraps the nine dg files in namespace CategoryTheory. ADR-0012 asks for established mathematical namespaces rather than a repository prefix, so the classes sit at CategoryTheory.DGCategory, CategoryTheory.DGFunctor and so on, with the DGCategory-specific lemmas nesting under CategoryTheory.DGCategory." }
  - { rev: 9, at: "2026-08-15", note: "DG2 epics filed as #376 (e5), #377 (e6), #378 (e7), each body written against a fresh measurement of mathlib_rev 520045ab rather than against the plan. Three results. (1) e2's summary claimed a shift on Hom-complexes that #343 never landed; the claim is corrected here and the work moves to e5. (2) Pretriangulated requires HasShift C ℤ, HasZeroObject and additivity of every shiftFunctor, and H0 C has only Preadditive -- that, not the six class fields, is e6's real cost, and it was invisible when e6 was sized XL. (3) m2's acceptance asks for agreement 'stated as an isomorphism of triangulated structures', a notion the pin does not have; since Pretriangulated.distinguishedTriangles is a field, agreement is an equality of Set (Triangle _) plus Functor.CommShift. That clause is left standing and #378 carries the obligation to amend it in the PR that proves the theorem, so the wording changes with the evidence rather than ahead of it. Also measured as already present at the pin: CochainComplex.mappingCone, HomotopyCategory.triangle and distinguishedTriangles, Pretriangulated and IsTriangulated on HomotopyCategory, Cochain.shift/rightShift/leftShift/rightUnshift/leftUnshift, and Localization.pretriangulated as candidate transport machinery. m2 moves to lane now." }
  - { rev: 10, at: "2026-08-15", note: "e5 landed as #382 and e6 opened. Two departures from e5 as filed, both in the implementation rather than the plan. (1) Shift and cone are stated as representability -- right composition with one fixed closed element is bijective in every degree from every object -- rather than as twisted-complex constructions. That is what makes naturality fall out of dgComp_assoc instead of becoming an axiom every witness must discharge. (2) IsPretriangulated carries a dg-level zero object, exists Z with dgId Z = 0, which is stronger than the literature: Cone (id X) is contractible so the cone clause alone would suffice, but contractibility is a statement about H0 and belongs to e6, so assuming it in e5 would invert the dependency. Every model here has a strict dg zero object, so it costs nothing. rev 9's measurement of e6 is confirmed independently: HasZeroObject (H0 C) is immediate from the zero clause and is done, and HasShift (H0 C) Z is the whole remaining cost. exists_shift gives an existential, not a choice, and HasShift needs a functor with shiftFunctorZero and shiftFunctorAdd coherence. e6 s first slice therefore lands the Enhancement structure, Cdg.enhancement as its first inhabitant, H0.hasZeroObject, and the uniqueness of shifts up to canonical isomorphism that any such functor will need; the functor itself is still open." }
  - { rev: 11, at: "2026-08-18", note: "e6 closed. H⁰ of a pretriangulated dg category is pretriangulated, with axioms [propext, Classical.choice, Quot.sound] and no sorry. The finding worth carrying forward is about the axiom system rather than about dg categories: Pretriangulated states TR2 as an iff, only the forward half is ever proved by a construction, and the reverse half follows from the other four axioms. It is proved once, generically, over Set (Triangle D), which also keeps it clear of the H0 C := C transparency hazard that shaped every proof in this track -- the proof never meets H0 C. Pretriangulated.Axioms and Axioms.pretriangulated are the reusable form: five obligations in, the class out. Also recorded against e6 acceptance criterion 4: Localization.pretriangulated is inapplicable here rather than merely expensive, because it transports from a category that already carries the structure. e7 is unblocked." }
  - { rev: 12, at: "2026-08-28", note: "DG2 is complete: #378 closed through #771 and main contains Model/Agreement.lean. Milestone #21 is closed. DG3 is now materialized as #853 (the reusable K-injective model), #854 (the derived seam), and #855 (linear/exact H0 functors); #853 and #855 are independent ready lanes." }
  - { rev: 13, at: "2026-08-28", note: "Implementation started on #855 in a fresh worktree. #853 remains the other independent ready lane; #854 stays blocked on the derived model rather than being represented by an empty seam structure." }
  - { rev: 14, at: "2026-09-30", note: "e10 (#855) closed out. An audit against main found all four scope items already landed across #861, #1239, #1276, #1367 and 21b8db6c while the issue and this entry still read in_progress; the one acceptance row with no named theorem, that the induced equivalence agrees with DGFunctor.h0, is now h0Equivalence_functor, and isTriangulated_h0Equivalence (Mathlib's Equivalence.IsTriangulated) joins the two halves. Three stale docstrings that still said exactness was unpackaged, or that no theorem related a quasi-equivalence to H⁰, are corrected." }
retired: []

# GitHub milestones are roadmap phases. Epic issues live inside them, and
# gh_sub_issues records native GitHub sub-issue relationships. IDs are
# write-once; retire an item instead of reusing its id.
#
# Materialized items carry their GitHub IDs; later phases remain null until
# their issue bodies are ready to file.
items:
  - id: dg-enhancements-m0
    gh_milestone: 19
    kind: milestone
    title: "DG0 — Boundary, pins, and the definition decision"
    status: done
    lane: now
    acceptance:
      - "A reconnaissance note records, declaration by declaration and against `mathlib_rev` in pins.json, what the enriched-category, HomComplex, monoidal, homotopy-category, K-injective, localization, and DerivedCategory surfaces actually provide — including the commands run and their output."
      - "ADR-0010 records the dg-category encoding decision (enriched over cochain complexes vs bespoke structure) with the elaboration evidence for the chosen route; if it stays `open`, it names the owner and what changes either way."
      - "ADR-0010 fixes the dg encoding; ADR-0012 supersedes its temporary root decision and places the implementation under DerivedAlgGeo/CategoryTheory/DGCategory with namespace CategoryTheory.DGCategory."
      - "Five zero-claim coverage maps exist under registry/ for 1312.5619v3, 1507.05509v5, 2101.04404v1, 2402.04605v1, and 2407.05946v2, each validating under scripts/check_coverage_map.py with every part at `target`."
      - "formalization.yaml's `source` is unchanged, registry/bridgeland2007.json is unchanged, and no coverage map contains a corpus-derived key."
      - "Labels, milestones, and project views for this track exist; every epic below carries at least one `source:` label and exactly one `area:` primary."
      - "No Lean declaration lands in DG0. If one is written to test elaboration, it lives in a throwaway branch named in the reconnaissance note and is not merged."

  - id: dg-enhancements-e0
    gh_issue: 328
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m0
    title: "Materialize the DG track: labels, milestones, views, and the source pins"
    status: done
    summary: "Create the eight new labels, the six milestones, the project views and fields, and the five zero-claim coverage maps. Mechanical, but it is the gate that makes every later epic filable with correct provenance. Landed in #330: labels, milestones #19-#24, project 8, registry/README.md, and the five maps. Issue #328 is closed."
    priority: must
    size: S
    tags: [governance, tracker, source-pin]
    labels: [area:governance, type:infra]

  - id: dg-enhancements-e1
    gh_issue: 329
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m0
    title: "Measure the pinned Mathlib dg surface and decide the encoding"
    status: done
    summary: "Reconnaissance plus ADR-0010. Establish whether EnrichedOrdinaryCategory over CochainComplex (ModuleCat k) ℤ elaborates at the pin with the monoidal instance Mathlib actually ships, and choose between that and a bespoke DGCategory structure. Output is a note and an ADR, not a library. Landed in #330: the reconnaissance note and ADR-0010, which chose Option B because HomologicalComplex.HasTensor does not synthesize for the ℤ-indexed shape at the pin. Issue #329 is closed."
    priority: must
    size: M
    tags: [research, adr, mathlib-surface]
    labels: [area:dg, area:governance, type:spike, research]

  - id: dg-enhancements-m1
    gh_milestone: 20
    kind: milestone
    title: "DG1 — The dg category, its H⁰, and the complexes model"
    status: done
    lane: now
    acceptance:
      - "A dg category and a dg functor are defined at the pin, k-linearity carried explicitly rather than as a global instance, with no new axiom and no sorry."
      - "Z⁰ and H⁰ are defined; H⁰ is a functor on dg categories and is proved additive."
      - "Quasi-equivalence is defined (fully faithful on H⁰ up to quasi-isomorphism of Hom-complexes, essentially surjective on H⁰). It is a definition in this milestone; no comparison theorem is claimed."
      - "C^dg(A) is constructed from Mathlib's HomComplex for an abelian (or preadditive, if the reconnaissance permits) category A, without re-deriving the Hom complex."
      - "H⁰(C^dg A) ≌ HomotopyCategory A (ComplexShape.up ℤ) is proved and exported from `DerivedAlgGeo.CategoryTheory.DGCategory`."
      - "The subject root builds, is source-independent, and exports through its own umbrella and `DerivedAlgGeo.CategoryTheory`."
      - "Every dg-category declaration lives in the `CategoryTheory.DGCategory` namespace or an established Mathlib namespace when extending its API."
      - "Zero sorries; lake build, both audits, both linters, lint-style, check_pin, check_anchor_free, check_coverage_map, and the emitter gate are green."

  - id: dg-enhancements-e2
    gh_issue: 335
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m1
    title: "DG categories, dg functors, and the Hom-complex shift"
    status: done
    summary: "The definitional core: a bespoke `DGCategory` structure over a commutative ring k (ADR-0010 Question 1, Option B), dg functors, the opposite and product dg categories, and the shift on Hom-complexes with its sign conventions fixed once and documented where a reader will find them. Composition is written directly on `Cochain.comp`; no monoidal structure is required or assumed. NOT DELIVERED, and moved to e5 (#376): the shift on Hom-complexes named in this summary is absent from the tree -- grep over DerivedAlgGeo/CategoryTheory/DGCategory finds two doc-comment mentions and one shiftFunctorZero use in Model/Seam.lean, and no shift construction. The rest landed in #343, with ADR-0011 settling the Hom-complexes over AddCommGrpCat and k-linearity as the refinement `DGLinear`. Issue #335 is closed."
    priority: must
    size: XL
    tags: [lean, definitions, dg-core]
    labels: [area:dg, type:feature, source:1507.05509]

  - id: dg-enhancements-e3
    gh_issue: 336
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m1
    title: "Z⁰, H⁰, and the definition of a quasi-equivalence"
    status: done
    summary: "The passage functors. Z⁰ and H⁰ from a dg category to an ordinary category, additivity of H⁰, functoriality on dg functors, and quasi-equivalence as a definition with no comparison theorem attached to it yet. Landed in #354. Issue #336 is closed."
    priority: must
    size: M
    tags: [lean, h0, seam]
    labels: [area:dg, bridge:h0-seam, type:feature, source:1507.05509]

  - id: dg-enhancements-e4
    gh_issue: 337
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m1
    title: "C^dg(A) from HomComplex, and H⁰(C^dg A) ≌ HomotopyCategory A"
    status: done
    summary: "The first bridge to an object the repository already uses. Build the dg category of cochain complexes on Mathlib's HomComplex, then prove its H⁰ is the homotopy category and export it through `DerivedAlgGeo.CategoryTheory.DGCategory`. Landed in #370 as `Cdg.seam : H0 (Cdg A) ≌ HomotopyCategory A (ComplexShape.up ℤ)`, the track's first theorem, on propext, Classical.choice and Quot.sound alone. Issue #337 is closed."
    priority: must
    tags: [lean, model, seam, first-theorem]
    size: M
    labels: [area:dg, area:derived, bridge:h0-seam, type:feature, source:1507.05509]

  - id: dg-enhancements-m2
    gh_milestone: 21
    kind: milestone
    title: "DG2 — Pretriangulated dg categories and the enhancement structure"
    status: done
    lane: now
    acceptance:
      - "Shift and cone are available at the dg level, and `IsPretriangulated` is a predicate on a dg category rather than an assumption carried in prose."
      - "`Enhancement T` is a structure bundling a dg category with an equivalence H⁰ A ≌ T, and it is inhabited by a constructed example before any theorem quantifies over it."
      - "A pretriangulated dg category transports a Pretriangulated (and, where the octahedral axiom is available, Triangulated) instance to its H⁰."
      - "On C^dg(A) the transported structure is proved to agree with Mathlib's existing instance on HomotopyCategory A — an explicit comparison, stated as an equality of the two `Set (Triangle _)` of distinguished triangles (`Cdg.seam_distinguishedTriangles_eq`, and `Cdg.distinguishedTriangles_eq` read on H⁰) with the shift compatibility carried by a `Functor.CommShift ℤ` on the seam (`Cdg.h0FunctorCommShift`), not left as a definitional coincidence. Amended in the PR closing #378: the original wording asked for 'an isomorphism of triangulated structures', a notion mathlib_rev 520045ab does not have — `Pretriangulated.distinguishedTriangles` is a field, so agreement is a set equality plus `CommShift`, and that is what was proved."
      - "Nonvacuity: K^dg(A) inhabits Enhancement (HomotopyCategory A) with the agreement theorem as its evidence."
      - "Zero sorries; all gates green; nothing anchor-free depends on the Anchor or Compatibility roots."

  - id: dg-enhancements-e5
    gh_issue: 376
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m2
    title: "Dg-level shift and cone, and the pretriangulated predicate"
    status: done
    summary: "Twisted-complex-style shift and cone objects inside a dg category, and IsPretriangulated as a predicate. Where Mathlib's MappingCone already supplies the model case, consume it rather than re-proving it. Scope grew on measurement: this epic also owns the shift on Hom-complexes that e2's summary claimed and #343 did not land."
    priority: must
    size: L
    tags: [lean, pretriangulated, cone]
    labels: [area:dg, type:feature, source:1507.05509]

  - id: dg-enhancements-e6
    gh_issue: 377
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m2
    title: "The Enhancement structure and transport of the triangulated structure to H⁰"
    status: done
    summary: "Enhancement T as a structure, and the transport theorem: H⁰ of a pretriangulated dg category carries a pretriangulated structure, proved rather than instantiated by fiat. Closed by the instance in DGCategory/H0Triangle.lean. The measured obstacle -- HasShift (H0 C) ℤ and HasZeroObject (H0 C) -- was the larger half and landed first; the six class fields followed as cone projections (#587), the distinguished triangles and three axioms (#596), the rotation comparison (#609), the forward rotation axiom (#615), and the lifting axiom plus the instance here. Two departures from the plan, both recorded in the PRs. (1) The reverse direction of the rotation axiom is not a dg construction: it is a theorem about the other four axioms, and it is proved generically in CategoryTheory/Triangulated/PretriangulatedAxioms.lean as Pretriangulated.Axioms.mem_of_rotate_mem, so a category that supplies the forward half supplies the class. That file is upstream-candidate. (2) Localization.pretriangulated was NOT used, and not because it was dearer: it transports a structure from a category that already has one, and H⁰ C is where this structure originates, so there is nothing to localize at. The equivalence variant would need an Enhancement with a pretriangulated target, which is e7 and would invert the dependency."
    priority: must
    size: XL
    tags: [lean, enhancement, transport, seam]
    labels: [area:dg, area:derived, bridge:h0-seam, type:feature, source:2101.04404]

  - id: dg-enhancements-e7
    gh_issue: 378
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m2
    title: "K^dg(A) enhances the homotopy category, and the structures agree"
    status: done
    summary: "The first inhabited enhancement, plus the agreement theorem against Mathlib's own triangulated instance. Without the agreement clause this epic would produce a second, incomparable triangulated structure on the same category. Mathlib already has Pretriangulated and IsTriangulated on HomotopyCategory at the pin, so the target structure is free and the cost moves to Functor.CommShift for Cdg.seam. Delivered by #771, which closed #378; DGCategory/Model/Agreement.lean carries the agreement theorem in main."
    priority: must
    size: L
    tags: [lean, nonvacuity, agreement]
    labels: [area:dg, area:derived, bridge:h0-seam, type:feature, source:1507.05509]

  - id: dg-enhancements-m3
    gh_milestone: 22
    kind: milestone
    title: "DG3 — The derived enhancement and the classical seam"
    status: in_progress
    lane: now
    acceptance:
      - "D^dg(A) is constructed on the h-injective (or h-projective) model, reusing Mathlib's KInjective/KProjective rather than a new resolution theory."
      - "The seam theorem H⁰(D^dg A) ≌ DerivedCategory A is proved as an equivalence of triangulated categories under Mathlib's HasDerivedCategory hypotheses, with the universe conditions stated where a reader meets them."
      - "The equivalence is proved compatible with shift, with distinguished triangles, and with Mathlib's Ext through the enhancement."
      - "A dg functor induces an exact functor on H⁰, and H⁰ sends a quasi-equivalence to an equivalence."
      - "Nothing in this milestone weakens or duplicates Mathlib's DerivedCategory; the comparison direction is explicit in both the statement and the module docstring."
      - "Zero sorries; all gates green."

  - id: dg-enhancements-e8
    gh_issue: 853
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m3
    title: "D^dg(A) on the h-injective model"
    status: done
    summary: "The derived enhancement as the full dg subcategory of h-injective (dually h-projective) complexes, built on Mathlib's KInjective/KProjective so the resolution theory is consumed rather than rebuilt."
    priority: must
    size: XL
    tags: [lean, derived, h-injective]
    labels: [area:dg, area:derived, type:feature, source:1507.05509]

  - id: dg-enhancements-e9
    gh_issue: 854
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m3
    title: "The seam theorem: H⁰(D^dg A) ≌ DerivedCategory A"
    status: in_progress
    summary: "The theorem the track exists for. An equivalence of triangulated categories between the H⁰ of the derived enhancement and Mathlib's DerivedCategory, compatible with shift, triangles, and Ext. Unblocked when e8 landed the h-injective model. #1392 landed the comparison functor and the first of the three mathematical obligations #854 records: derivedSeamFunctor is full and faithful UNCONDITIONALLY, and derivedSeam is an equivalence of CATEGORIES under HasKInjectiveResolutions -- Spaltenstein's theorem, which Mathlib does not have at mathlib_rev. That class states the mathematics upstream of the theorem rather than the conclusion, and is spent in derivedSeamFunctor_essSurj alone, so it becomes an instance the day the construction is formalized and no consumer changes. #1392 does NOT close #854 and says so: the other two obligations -- transported shifts and distinguished triangles agreeing with the target instances, and the Ext comparison commuting with the seam -- are not proved yet, and closing the issue on the headline alone is the #554 defect RM-07 exists to catch."
    priority: must
    size: XL
    tags: [lean, seam, headline]
    labels: [area:dg, area:derived, bridge:h0-seam, type:feature, source:2101.04404]

  - id: dg-enhancements-e10
    gh_issue: 855
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m3
    title: "Dg functors induce exact functors, and quasi-equivalences induce equivalences"
    status: done
    summary: "The functor half of the passage. This is where the programme's naturality claim first becomes testable: an induced functor should arrive with its compatibilities rather than acquire them by separate argument. Landed in four slices: #861 (Linear k (H0 C) and F.h0 linear), #1239/#1276/#1367 (every dg functor preserves shifts and chosen cones, and h0IsTriangulated derives the CommShift and IsTriangulated structure from them), 21b8db6c (a quasi-equivalence induces an equivalence on H⁰), and the close-out PR (the induced equivalence has functor H⁰ F, and is a triangulated equivalence between pretriangulated dg categories)."
    priority: must
    size: L
    tags: [lean, functors, naturality]
    labels: [area:dg, area:derived, bridge:h0-seam, type:feature, source:1312.5619]

  - id: dg-enhancements-e11
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m3
    title: "The Drinfeld dg quotient"
    status: parked
    summary: "Parked. The explicit Drinfeld quotient needs no model structure, but relating H⁰(A/N) to the Verdier quotient H⁰(A)/H⁰(N) is a research-scale obligation. Un-park trigger: the seam theorem in e9 is closed on the h-injective model AND a downstream epic needs a quotient the h-injective model cannot supply."
    priority: could
    size: XL
    tags: [parked, quotient, research]
    labels: [area:dg, parked, research, source:1507.05509]

  - id: dg-enhancements-m4
    gh_milestone: 23
    kind: milestone
    title: "DG4 — Transport: t-structures, slicings, and stability across the seam"
    status: planned
    lane: later
    acceptance:
      - "A t-structure transports across an enhancement equivalence in both directions, with the round trip proved to be the identity."
      - "A slicing and a pre-stability condition from `CategoryTheory.Triangulated` transport the same way, with heart, phase, and Harder–Narasimhan data proved to correspond."
      - "The transported slicing is proved natural with respect to H⁰ of a dg functor, and that statement is connected to the pullback/pushforward interfaces of the stability-families track rather than restated independently."
      - "The programme's central claim is tested, not assumed: at least one naturality obligation is proved on both sides of the seam and the two proofs are compared in the epic. If the dg proof is not cheaper, the negative result is recorded in this roadmap and in formalization.yaml prose, and the m4 rationale is revised."
      - "No stability-condition result acquires a geometric hypothesis it did not have before; realization hypotheses stay explicit, per CLAUDE.md."
      - "Zero sorries; all gates green."

  - id: dg-enhancements-e12
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m4
    title: "Transport a t-structure and a slicing across an enhancement equivalence"
    status: planned
    summary: "The transitional API. Transport in both directions with round-trip identities, so a statement may be made on whichever side is cheaper and read on the other."
    priority: must
    size: L
    tags: [lean, transport, t-structure, slicing]
    labels: [area:dg, area:stability, bridge:h0-seam, type:feature, source:2101.04404]

  - id: dg-enhancements-e13
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m4
    title: "Transport stability conditions, and test the naturality claim"
    status: planned
    summary: "Carry a pre-stability condition across the seam, prove the transported slicing natural under H⁰ of a dg functor, and settle whether the dg side is genuinely cheaper by proving one obligation twice. The negative result is a permitted and publishable outcome of this epic."
    priority: should
    size: XL
    tags: [lean, stability, naturality, falsifiable]
    labels: [area:dg, area:stability, bridge:h0-seam, type:feature, source:2101.04404]
    links:
      roadmaps: [stability-families]

  - id: dg-enhancements-e14
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m4
    title: "Dg-native statements of the relative and family interfaces"
    status: blocked
    summary: "State the §§20–23 family conditions dg-natively once the seam and the uniform-support layer both exist. Blocked on stability-families SF2; filing it now only fixes the dependency, and no interface may be introduced without an inhabitant."
    priority: could
    size: XL
    tags: [blocked, families, cross-track]
    labels: [area:dg, area:families, blocked, source:2101.04404]
    links:
      roadmaps: [stability-families]

  - id: dg-enhancements-m5
    gh_milestone: 24
    kind: milestone
    title: "DG5 — Uniqueness of enhancements: the statement layer"
    status: planned
    lane: later
    acceptance:
      - "Uniqueness of enhancements is stated as a Prop over this repository's own definitions, for each decoration the source papers treat, with no inhabitant, no axiom, and no sorry."
      - "Every source coordinate in the five coverage maps remains `target` unless a landing PR carries the evidence object the validator demands; nothing here promotes a status."
      - "No prose in README.md, CONTRIBUTING.md, or formalization.yaml claims coverage of any uniqueness theorem."
      - "The weakly-approximable vocabulary is recorded as source coordinates only; no definition is imported into the library without a theorem that needs it."
      - "If a statement cannot be written without machinery the pin lacks, that is recorded as a named `mathlib-gap` kind_label in the appropriate map, and the corresponding command in registry/README.md is updated in the same commit."

  - id: dg-enhancements-e15
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m5
    title: "State uniqueness of enhancements for the derived category, zero-claim"
    status: planned
    summary: "Write the statements of Canonaco–Neeman–Stellari for the unbounded, bounded, bounded-above and bounded-below decorations as Props over the formalized definitions. No proof is attempted and none is implied."
    priority: should
    size: M
    tags: [statement-layer, zero-claim]
    labels: [area:dg, type:docs, source:2101.04404, source:1507.05509]

  - id: dg-enhancements-e16
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m5
    title: "Record the weakly-approximable and intrinsic-subcategory vocabulary"
    status: planned
    summary: "Map the 2402.04605 and 2407.05946 vocabulary onto coverage coordinates without importing any of it into the library. A definition enters the library only when a theorem in this repository needs it."
    priority: could
    size: S
    tags: [statement-layer, vocabulary]
    labels: [area:dg, type:docs, source:2402.04605, source:2407.05946]

  - id: dg-enhancements-e17
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: dg-enhancements-m5
    title: "Internal Homs in Hqe and derived Morita theory"
    status: parked
    summary: "Parked with an un-park trigger. Toën's theorem, reproved by Canonaco–Stellari via extensions of dg functors, needs the localization of dgCat at quasi-equivalences — a homotopy theory the pin does not have. Un-park trigger: Mathlib gains a model structure or a Bousfield localization applicable to dgCat, or the owner accepts a bespoke localization as a track of its own."
    priority: wont
    size: XL
    tags: [parked, hqe, morita]
    labels: [area:dg, parked, source:1312.5619]

Exported from .claude/roadmap/dg-enhancements.yaml at the commit that removed it from the tree.

Home

Architecture

CI and runners

Decision records

Roadmaps

Clone this wiki locally