Skip to content

Roadmap Charge Abstraction

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

Charge Abstraction

# Structural companion: mathematical-ownership.yaml (MO1, milestone 52).
# CA1–CA3 keep mathematical ownership; MO1 owns placement and extraction.
# See https://github.com/chris-dare-dev/derived-alg-geo-lean/blob/965d62fc66afbbe998482ef0912cacc265a01c5a/docs/reviews/2026-09-13-mathematical-ownership-plan.md for exact issue links.
schema: roadmap/1
slug: charge-abstraction
project: derived-alg-geo-lean
title: "One charge, indexed by truncation degree — and the leaves that reach it"
status: active
phase: sequenced          # init -> refined -> decomposed -> sequenced -> complete
brief: |
  This track exists because an adversarial abstraction audit on 2026-09-12 found
  the same defect in the charge lane that `numerical-k-theory` found in the
  Grothendieck-group lane: one mathematical object written several times because
  the literature spells it differently per example.

  The surface charge on three real coordinates
  (`CentralCharge/Numerical/SurfaceFamily.lean:26`) and the Bayer--Macrì--Toda threefold
  charge on four (`CentralCharge/Numerical/Threefold.lean:110`) are the same generating
  formula truncated at two codimensions. The threefold file's own docstring says
  so. They are nevertheless two unrelated roots with no shared parent, a fourfold
  charge does not exist, and the divisorial charge
  (`CentralCharge/Divisorial/Charge.lean`) — the one that works in arbitrary Picard
  rank — sits beside both.

  THE INDEX IS THE POINT, and it is where `numerical-k-theory` stopped. That
  track's fourth assumption reads "the fourfold numerical formulas admit the same
  uniform shape as the surface and threefold ones, so one indexed type covers all
  three", and it is marked FALSIFIED 2026-08-27, parking `numerical-k-theory-e5`.
  That falsification is correct and this track does not overturn it. What the
  audit adds is the reason: dimension was the wrong index. The uniform object is
  indexed by TRUNCATION DEGREE `m`, and the geometric transport carries the
  dimension `n` separately. The proof that these must be separate indices is that
  the cubic-threefold Kuznetsov charge is `(n, m) = (3, 2)`: a family indexed by
  dimension alone cannot state it at all. So this track unparks e5 by changing
  its index rather than by re-opening its assumption.

  What makes this track unusual is that its own first design was wrong in ways
  its own review caught. A completeness critic found that the design left the
  correction slot uninhabited, shipped a second Mukai pairing with no bridge to
  the existing one, justified the cubic-fourfold node with a statement about the
  wrong lattice, and dropped a rotation that a verifier had flagged. A
  remediation pass then proved five repairs in Lean, refuted three claims of the
  design by counterexample, and two adversarial checkers caught seven further
  proposals that would themselves have shipped a second spelling of an existing
  declaration. Every one of those is a recorded negative result rather than a
  deletion, and the epics below carry them.

  What this is NOT: new stability mathematics. No epic here proves a new theorem
  about a variety. Each one either removes a copy in favour of an existing
  definition, or replaces an enumeration by a parameter, or records why an
  apparent unification is false.
goal:
  hmw: "How might we make the charge lane carry one polynomial indexed by truncation degree, so that a new dimension or a noncommutative example supplies data and inherits the charge, without restating a line of its mathematics?"
  objective: "Collapse every exponential central-charge definition in the repository onto one kernel indexed by truncation degree, give the geometric transport one owner for every (n, m, κ), and leave the discriminant, the twist and the Mukai pairing with exactly one root apiece."
  key_results:
    - "Exactly one exponential charge polynomial exists; a grep for a second returns nothing. The surface, threefold and slope charges are theorems about it at m = 2, 3, 1."
    - "The divisorial charge in arbitrary Picard rank is a proved instance of the same kernel, not a sibling of it. If only the rank-one slice reaches the root, the unification is cosmetic and has failed."
    - "A fourfold wall family exists and costs one `Polarization` and zero new polynomials."
    - "The Bogomolov--Mukai discriminant has one owner, reached by all seven of its current spellings."
    - "The (3,2) tilt charge is the existing surface family pulled back along a truncation, carries its rotation, and exactly one spelling of it survives."
    - "No declaration added by this track duplicates an existing one without a proved comparison theorem landing in the same pull request."
  assumptions:
    - tier: must
      text: "The surface and threefold charges are the same generating formula at two truncation degrees."
      validation: "CONFIRMED 2026-09-12 in Lean: `stCharge` and `Threefold.charge` are both proved equal to `Exp.ofMoments` at m = 2 and m = 3, and both family-level comparisons elaborate."
    - tier: must
      text: "Truncation degree and ambient dimension must be separate indices."
      validation: "CONFIRMED: the cubic-threefold Kuznetsov charge is (n, m) = (3, 2). A dimension-indexed family cannot state it. This is also why `numerical-k-theory-e5` was parked."
    - tier: must
      text: "The multi-divisor divisorial charge is a child of the kernel, not a sibling."
      validation: "CONFIRMED at the triple level: `expCharge_eq_ofMoments` is proved on a bare `Mukai.RealExtension D`. Two of three independent design angles asserted the opposite and were wrong; the keystone must be stated on the triple, not on `ChernCharacter`, or it cannot reach `mukaiCharge`."
    - tier: must
      text: "The correction class κ is a pullback, not a coefficient, and κ = 1 and κ = √td are two pullbacks."
      validation: "CONFIRMED: they give genuinely different walls. The transport at √td is proved equal to the existing Mukai charge on the rank-one slice for any surface, no K3 hypothesis."
    - tier: should
      text: "One dimension-indexed self-pairing can parent every discriminant in the tree."
      validation: "FALSIFIED 2026-09-12. A dimension-indexed Mukai self-pairing vanishes identically in odd degree, so the threefold leaf cannot reach it. The root is the fixed-arity pairing generalised over its COEFFICIENT RING; the arity stays three. Recorded in `docs/architecture/abstraction-tree.md`."
    - tier: should
      text: "A new graded pairing is needed to state the parity theorem."
      validation: "FALSIFIED at n = 2: the new form is literally `Mukai.realPairing (LinearMap.mul ℝ ℝ)` — same carrier, no weight, no transport. It survives only for n ≠ 2, and only because two further unifications are refuted. n = 2 must land as a comparison theorem, never as a definition."
    - tier: should
      text: "Parity of the compressed pairing decides the cubic-fourfold Kuznetsov node."
      validation: "FALSIFIED. The compressed form lives on `Fin (n+1) → ℝ`, dimension 5 at n = 4; the Mukai lattice of the component has rank 24 and signature (4,20). Symmetry there is automatic for an unrelated reason, so the parity argument is vacuous even on the correct space. The node needs a signature-additivity transfer, which is proved, plus geometry the tree does not have."
  wont:
    - "No new stability mathematics. No epic proves a new theorem about a variety."
    - "No `KuznetsovChargeData` carrier. A noncommutative example reaches the tree by restriction along its inclusion; a dedicated carrier is forbidden and the rule is recorded in the spine."
    - "No dimension-indexed generalisation of the Mukai pairing. The arity is fixed at three and only the coefficient ring generalises."
    - "No fusion of the cubic fourfold's charge lattice with its period-domain lattice. They are orthogonal complements for a very general X, so no comparison theorem between them can exist."
    - "No `sorry`, including as a refactoring convenience while a copy is being removed."
generated_by: { agent: claude-opus-5, at: "2026-09-12" }
generations:
  - at: "2026-09-12"
    note: |
      Decomposed from the 2026-09-12 abstraction audit, its completeness critic,
      and the remediation pass that repaired the audit's own design. The three
      documents are in https://github.com/chris-dare-dev/derived-alg-geo-lean/tree/965d62fc66afbbe998482ef0912cacc265a01c5a/docs/reviews; the scope that produced this file is
      https://github.com/chris-dare-dev/derived-alg-geo-lean/blob/965d62fc66afbbe998482ef0912cacc265a01c5a/docs/reviews/2026-09-12-charge-abstraction-track-scope.md.

      CA4 (noncommutative varieties) is deliberately NOT filed as GitHub issues.
      All three of its epics are blocked on obligations outside the tree, and an
      open unblockable issue trains people to ignore the tracker. The entries
      carry `gh_issue: null` until their blockers are discharged.
items:
  - id: charge-abstraction-m1
    gh_milestone: 49
    kind: milestone
    title: "CA1 — One charge, indexed by truncation degree"
    status: planned
    lane: now
    acceptance:
      - "`Wall.Exp.ofMoments` exists once, and the surface, threefold and slope charges are proved equal to it at m = 2, 3 and 1."
      - "The divisorial charge is a proved instance of the same kernel at the intersection-form moments, stated on the triple so that it reaches `mukaiCharge`."
      - "The κ slot has a real inhabitant: the transport at √td is proved equal to the existing Mukai charge on the rank-one slice, for any surface and with no K3 hypothesis."
      - "The Bogomolov--Mukai discriminant has exactly one owner; a grep for a second definition on H-degrees returns nothing."
      - "No declaration added by this milestone duplicates an existing one without a proved comparison theorem in the same pull request."
      - "Zero sorries; all gates green, including `check_roadmap.py --require-api`."
  - id: charge-abstraction-e1
    gh_issue: 1219
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m1
    title: "The exponential charge kernel, indexed by truncation degree"
    status: done
    summary: "`CentralCharge/Exponential/Kernel.lean`: `ofMoments m μ = -∑_{j≤m} (-1)^j/j! · μ j`, `HDeg`, `moments`, `charge`, `chargeFamily`, and the two charts; `CentralCharge/Exponential/Comparison.lean` proves `stCharge` (`CentralCharge/Numerical/SurfaceFamily.lean:26`) and `Threefold.charge` (`CentralCharge/Numerical/Threefold.lean:110`) are this kernel at m = 2 and m = 3. The kernel imports only `CentralCharge/Family.lean` and Mathlib, so the root imports no leaf, and lands with two inhabitants. The `Fin (m+1)` indexing stands: the audit's claim that the root had to be ℕ-indexed to keep `degH` definitional was falsified on both halves."
    priority: must
    size: M
    tags: [lean, charges, root]
    labels: [area:lean, area:stability, type:feature, sev:high]
  - id: charge-abstraction-e2
    gh_issue: 1220
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m1
    title: "The divisorial keystone reaches the charge kernel"
    status: done
    summary: "`CentralCharge/Exponential/Divisorial.lean`: the intersection-form moments and `expCharge_eq_ofMoments`, stated on a bare `Mukai.RealExtension D` rather than on `ChernCharacter`. The triple level is load-bearing — `mukaiCharge` (`CentralCharge/Divisorial/Mukai.lean:144`) is the same `expChargeHom` on a different triple, so a `ChernCharacter`-level statement cannot reach it and the κ epic depends on that reach. This is the slice to review hardest: two of three independent design angles declared the multi-divisor charge a non-child and were wrong."
    priority: must
    size: M
    tags: [lean, charges, keystone]
    labels: [area:lean, area:stability, type:feature, sev:high]
    blocked_by: [1219]
  - id: charge-abstraction-e3
    gh_issue: 1221
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m1
    title: "The polarised transport: one map for every (n, m, kappa)"
    status: done
    summary: "`AlgebraicGeometry/Numerical/Stability/PolarisedWallTransport.lean`: `corrComp`, `hDegrees`, `hDegreesHom`, `wallChargeFamily`, and the two dimension comparisons `surface_toNumClass_eq` and `threefold_toNumClass_eq`, both proved with zero sorries. The n = 3 twin did not exist in any form before remediation; it was a code comment. `corrComp` must land with `corrComp_sqrtToddComp` and `corrComp_unitCorr` or it is a third spelling of an existing convolution. Two open obligations ride along: the name collision with `Surface.wallChargeFamily` (`WallTransport.lean:156`) and the tilt-spelling comparison, the latter owned by e10."
    priority: must
    size: L
    tags: [lean, charges, transport]
    labels: [area:lean, area:geometry, type:feature, sev:high]
    blocked_by: [1219]
  - id: charge-abstraction-e4
    gh_issue: 1222
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m1
    title: "Inhabit the correction slot at sqrt(td)"
    status: done
    summary: "`wallChargeFamily_sqrtToddComp_eq_mukaiCharge`, proved for any surface realization with no K3 hypothesis, plus the K3 corollary through `mukaiCharge_of_isK3` (`DivisorialMukai.lean:124`). Without it the κ slot has zero inhabitants and the spine's `κ = √td` line is unbacked. Must use `R.realizePolarization` (`DivisorialChargeNumerical.lean:106`); the remediation scratch file's local `polDiv` duplicates it. The audit's claim that the repository holds two unrelated √td data is half wrong — the bridge exists at `DivisorialMukai.lean:69` and the design cited neither; what is genuinely wrong is the audit's `√td = (1,0,1)`, which as an H-degree vector is `(∫H², 0, 1)`."
    priority: must
    size: S
    tags: [lean, charges, mukai]
    labels: [area:lean, area:geometry, type:feature, sev:medium]
    blocked_by: [1221]
  - id: charge-abstraction-e5
    gh_issue: 1223
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m1
    title: "One discriminant owner, and the inert beta dropped"
    status: done
    summary: "Generalise `Mukai.pairing` (`Lattice/Mukai/Basic.lean:56`) over its coefficient ring IN PLACE, with `Mukai.realPairing` an abbrev at R = ℝ; move `DivisorSpace.realDiscriminant` up to `RealForm.lean`, whose current home imports two of the leaves it would parent; land the eight leaf projections and five bridges; drop the inert β from `Threefold.discrHBeta` (`BMT.lean:103`). No new declaration name. The audit shipped two rival parents for these same seven leaves and adjudicated neither; the loser — a dimension-indexed Δ_H — is recorded as a negative result because a dimension-indexed self-pairing vanishes identically in odd degree."
    priority: must
    size: M
    tags: [lean, discriminant, refactor]
    labels: [area:lean, type:refactor, sev:high]
  - id: charge-abstraction-e6
    gh_issue: 1224
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m1
    title: "The e^-betaH twist on the kernel, and its two real gaps"
    status: done
    summary: "`CentralCharge/Exponential/Twist.lean`: `Exp.twist`, `twist_twist` for arbitrary real parameters, and `charge_twist` at every truncation degree — twisting the degrees by β moves the charge parameter from `w` to `w + β`. Both rest on one finite exponential convolution, proved by the binomial theorem and a triangular reindex; `twist_add_beta` was not reusable, as the epic predicted. All four prior spellings reach the root: the threefold group law inherits it through `betaTwist_eq_exp`, `ChargeCoordinates.twistByScalar_eq_exp` and `ChernCharacter.twist_rankOne_eq_exp` cover the surface scalar and the `B = βH` full character, and `Numerical/Stability/ExponentialTwist.lean` compares the retained rational `TwistedChern.twist` after weighted degrees for `m ≤ n`. One recorded negative result: an arbitrary divisor `B` does not descend to compressed `H`-degrees, and the rank-two counterexample is in the canonical-root policy. Landed in #1337 with no `sorry`; the nine new declarations passed the focused axiom audit on `propext`, `Classical.choice` and `Quot.sound` only."
    priority: should
    size: M
    tags: [lean, charges, twist]
    labels: [area:lean, type:feature, research, sev:medium]
    blocked_by: [1219]
  - id: charge-abstraction-e7
    gh_issue: 1225
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m1
    title: "The fourfold wall layer, at zero polynomial cost"
    status: done
    summary: "`Examples/Fourfold/ProjectiveSpaceWalls.lean` and `CalabiYauWalls.lean`. The fourfold numerical models already exist with proved HRR; no fourfold charge or wall family exists anywhere in the tree. The entire content of this epic is one `Polarization` per model, and it is the slice to point at when asked what the root bought. One gap to record rather than hide: the n = 4 graded pairing has no repository counterpart, no comparison and no consumer."
    priority: should
    size: S
    tags: [lean, charges, examples]
    labels: [area:lean, area:geometry, type:feature, sev:medium]
    blocked_by: [1221]
  - id: charge-abstraction-m2
    gh_milestone: 50
    kind: milestone
    title: "CA2 — The tilt node, and the rotation that was missing"
    status: planned
    lane: next
    acceptance:
      - "`threefoldTruncate` exists and the (3,2) tilt family is the existing surface family pulled back along it, with no new polynomial."
      - "The rotation is proved to be the existing `phaseTiltRotation` rather than a copy; walls are proved invariant under it and phases are proved not to be."
      - "Exactly one of the three candidate spellings of the tilt charge survives, with comparisons proved to the others or the others deleted."
      - "`Threefold.nu` is tied to the charge slope, and the `Threefold.nu` versus `BMT.nu` question is settled in writing."
      - "Zero sorries; all gates green, including `check_roadmap.py --require-api`."
  - id: charge-abstraction-e8
    gh_issue: 1226
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m2
    title: "The charge-family rotation, as the existing tilt rotation"
    status: done
    summary: "`Walls/Rotation.lean`: `ChargeFamily.phaseRotate β := smul (exp(-πβi))`, with the two bridges that stop it duplicating what exists — it IS `phaseTiltRotation` (`TiltGeometry.lean:42`, not `:43` as the audit cited) and it agrees with `phaseTiltCharge` (`:53`) on the constant family. Must NOT be merged into `Walls/ChargeFamily.lean`, which imports only Mathlib today and must stay stability-neutral. The rotation was flagged by a verifier during the audit and never reached the audit's deliverable; it is free on walls and provably not free on phases. LANDED: the charge-family root is `CentralCharge/Family.lean` and the wall lemmas are `Walls/Alignment.lean` since MO1.02/MO1.03, so `Walls/Rotation.lean` imports those two plus `TiltGeometry.lean`; `wall`/`wallValue` are `alignmentLocus`/`alignmentValue`. Phase non-invariance is recorded as `exists_chargeSlope_phaseRotate_ne`."
    priority: must
    size: S
    tags: [lean, charges, rotation]
    labels: [area:lean, area:stability, type:feature, sev:medium]
  - id: charge-abstraction-e9
    gh_issue: 1227
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m2
    title: "The tilt family by truncation, and its slope"
    status: done
    progress: "Landed in #1377. `threefoldTruncate` is the only new construction; `tiltFamily` is `stChargeFamily` reindexed by `Prod.swap` and pulled back along it, with the chart transposition in exactly one `reindex`. `nu_eq_tilt_slope` and `chargeSlope_tilt` prove the tilt slope is the charge slope rescaled by alpha, and `rotatedTiltFamily_alignmentLocus_eq_preimage` makes every rotated tilt wall a `Prod.swap` preimage of a surface one. Both negative results landed with computed witnesses at (alpha, beta) = (1, 0). #1228 still owns adjudicating the three tilt spellings and the two `nu`."
    summary: "`Walls/Threefold/Tilt.lean`: `threefoldTruncate : Threefold.NumClass →+ NumClass` (which does not exist today, and without which the nested-semicircle theorems cannot apply to ν-walls at all), `tiltFamily := (stChargeFamily.reindex Prod.swap).pullback threefoldTruncate`, `rotatedTiltFamily`, and `ν = α · chargeSlope` tying it to `Threefold.nu` (`Threefold/Basic.lean:208`). THE STRONGEST RESULT OF THE REMEDIATION: the (3,2) charge needs no new polynomial and no kernel, so this epic no longer depends on e1--e6 as originally scoped. Two negative results land with it: the tilt family is not a chart change of `Threefold.chargeFamily` (their wall loci differ, with a computed witness) and the rotation is not cosmetic."
    priority: must
    size: M
    tags: [lean, charges, tilt]
    labels: [area:lean, area:stability, type:feature, sev:high]
    blocked_by: [1226]
  - id: charge-abstraction-e10
    gh_issue: 1228
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m2
    title: "Adjudicate the three tilt spellings and the two nu"
    status: done
    progress: "Adjudicated in favour of `Tilt.tiltFamily`, which is the only one of the three spellings ever declared. `Exp.tiltChargeFamily` was planned for a slice that did not happen; `Walls/Threefold/TiltComparison.lean` supplies it as the theorem `tiltFamily_eq_exp` -- the tilt family IS `Exp.chargeFamily` at m = 2, reindexed by the surface chart after the (alpha, beta) transposition and pulled back along `threefoldTruncate` -- rather than as a second carrier, and `Exponential/Kernel.lean`'s forward reference is corrected to match. `tiltWallChargeFamily` existed only in `design-final.lean`; the implemented transport pulls back the UNTILTED `Threefold.chargeFamily`, and #1227's `exists_tilt_alignmentValue_ne_threefold` already separates the two with a computed witness, which is the false unification recorded as a negative result. On the slopes: both survive and neither is redundant -- `Wall.Threefold.nu` is the four-coordinate model and `BMT.nu` its realization on a polarised threefold, the Models/Examples separation MO1.06 exists to keep -- bridged by the pre-existing `ThreefoldWallTransport.nu_toNumClass`. New `AlgebraicGeometry/Numerical/Stability/ThreefoldTiltTransport.lean` composes that bridge with `nu_eq_tilt_slope`, so the geometric BMT slope is alpha times the tilt family's charge slope at the transported class. No new family, no third slope."
    summary: "Two duplications the charge work would otherwise ship, owned by no earlier epic. Three spellings of one tilt charge — `tiltWallChargeFamily` at the transport level, `Exp.tiltChargeFamily` at the kernel level, `Tilt.tiltFamily` from the rotation work — with no comparison between any pair. And two slopes, `Threefold.nu` (`Threefold/Basic.lean:208`) against `BMT.nu` (`BMT.lean:111`), unconfirmed against the tree. This exists as its own epic because the obligation is shared: attached as a footnote to two other issues, it gets dropped by both. A comparison theorem or a deletion; a docstring noting the similarity is not a discharge."
    priority: must
    size: S
    tags: [lean, charges, adjudication]
    labels: [area:lean, type:refactor, sev:high]
    blocked_by: [1221, 1227]
  - id: charge-abstraction-m3
    gh_milestone: 51
    kind: milestone
    title: "CA3 — The lattice root and the pairing bridge"
    status: planned
    lane: later
    acceptance:
      - "`Lattice.pairCharge` exists and both the period-domain and exponential charges are proved to be it, or the root is withdrawn and the bridge theorems are the whole delivery."
      - "`Mukai.Graded.pairing` at n = 2 is a comparison theorem, never a second definition."
      - "The Mukai pairing is generalised over its coefficient ring in place; no new name is introduced."
      - "Each new carrier is classified by the proof-witness test (`abstraction-tree.md` clause 2a) in the pull request, because `check_single_instantiation.py` does not scan these paths: a Lift needs no consumer count, a carrier with content names its consumers, an empty carrier is refused."
      - "Zero sorries; all gates green, including `check_roadmap.py --require-api`."
  - id: charge-abstraction-e11
    gh_issue: 1229
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m3
    title: "The pairing bridge, as theorems about the existing root"
    status: done
    progress: "Landed as three theorems beside `Mukai.realPairing` in `RealForm.lean`, with no new declaration: `realPairing_mul_eq_alternatingSum` (the n = 2 comparison, graded form written out rather than named), `realPairing_compression_rankOne` (the weighted bridge, `b H H` explicit and first power) and `realForm_compression_rankOne` (the factor of two is the `realForm` halving). The three refuted unifications are in the module docstring. One correction to the issue: `Exp.HDeg 2` is `Fin 3 -> R`, not definitionally `R x R x R`, so the comparison is stated on `Fin 3 -> R` coordinates with the repackaging explicit, as the issue's own MO1 note asks. All three depend only on propext, Classical.choice and Quot.sound."
    summary: "The n = 2 comparison, the weighted rank-one bridge and the `realForm` factor-of-two lemma, landed beside `Mukai.realPairing` in `RealForm.lean` as theorems about the root that already exists. Remediation proved what the audit had not checked: at n = 2 the proposed new graded form is literally `Mukai.realPairing (LinearMap.mul ℝ ℝ)` — same carrier, no weight, no transport — so shipping it as a definition would have added a second spelling of an existing pairing. Three refuted unifications go in the docstring; they are what justify any separate graded form existing at all. Deliberately sequenced BEFORE e12, inverting the audit's order."
    priority: must
    size: S
    tags: [lean, lattice, bridge]
    labels: [area:lean, type:feature, sev:medium]
  - id: charge-abstraction-e12
    gh_issue: 1230
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m3
    title: "The lattice charge root"
    status: refuted
    summary: "WITHDRAWN 2026-09-30. `Lattice.pairCharge` and `Mukai.Graded.pairing` were proposed as a nonsymmetric root of `PeriodDomain.centralCharge` and `Mukai.expCharge`. Research found `expCharge` is already `centralCharge` at `realForm b`; neither leaf's API uses symmetry of `b`; for nonsymmetric `b`, `expCharge ≠ pairCharge (realPairing b)`; the remaining Lift (add/smul/zero/kernel) has no consumer; the odd graded form is a recorded negative result with no consumer. The bridge theorems of e11 are the whole delivery. Residual cleanup, if wanted: `cPair` in `Exponential/Divisorial.lean` may call `centralCharge` directly."
    priority: should
    size: M
    tags: [lean, lattice, root, withdrawn]
    labels: [area:lean, type:feature, gate:owner, sev:medium]
    blocked_by: [1229]
  - id: charge-abstraction-m4
    gh_milestone: null
    kind: milestone
    title: "CA4 — Noncommutative varieties"
    status: planned
    lane: parked
    acceptance:
      - "Each external blocker is either discharged or recorded as a dependency with the exact statement it needs."
      - "No `KuznetsovChargeData` carrier exists; each component reaches the tree by restriction along its inclusion."
      - "The cubic fourfold's charge lattice and its period-domain lattice remain two nodes; no comparison theorem between them is attempted."
  - id: charge-abstraction-e13
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m4
    title: "Cubic threefold Ku(X), by the rotated tilt charge"
    status: planned
    summary: "NOT FILED: blocked on an obligation outside this track. Bayer--Lahoz--Macrì--Stellari induce stability on Ku(X) from the (n,m) = (3,2) tilt charge rotated by 1/i, not from the BMT ch₃ charge. The wall-level half is e9. What is missing is the statement that the rotated charge is a stability function on the tilted heart, which no repair proved and which is not a wall-level fact: it needs `phaseTiltCharge_im_pos_of_phaseTors` (`TiltGeometry.lean:278`) and the `WeakUpperClosed` machinery (`:68`). Do not land a thin `ChargeFamily.restrict` alias first."
    priority: should
    size: L
    tags: [lean, noncommutative, blocked]
    labels: [area:lean, area:stability, type:feature, blocked]
  - id: charge-abstraction-e14
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m4
    title: "Cubic fourfold Ku(X): the charge on the algebraic lattice"
    status: planned
    summary: "NOT FILED: blocked on the geometry frontier. Plain `PeriodDomain.centralCharge` on the algebraic Mukai lattice of the component, signature (2,ρ), which for a very general cubic fourfold IS the A₂ lattice. No restriction is involved. Needs the Mukai lattice of Ku(X) and its signature, neither of which exists in the tree; `grep -rni kuznetsov DerivedAlgGeo` returns two disclaimers and no construction."
    priority: should
    size: L
    tags: [lean, noncommutative, blocked]
    labels: [area:lean, type:feature, blocked]
  - id: charge-abstraction-e15
    gh_issue: null
    gh_sub_issues: []
    kind: epic
    parent: charge-abstraction-m4
    title: "Cubic fourfold Ku(X): the period domain on the A2 complement"
    status: planned
    summary: "NOT FILED: blocked on the geometry frontier. `PeriodDomain.centralCharge` restricted along the inclusion of A₂^⊥, signature (2,20), with the `HasSignatureTwo` witness PROVED from sig H̃ = (4,20) plus A₂ positive definite of rank 2, via `QuadraticMap.sigPos_eq_add` (`SignatureAdditive.lean:202`). Separate from e14 and must stay separate: the two live on lattices that are orthogonal complements for a very general X, so no comparison theorem between them can exist, and filing them together would invite exactly the fusion the audit refuted. Landing it requires de-privatising `QuadraticMap.polar_restrict` (`SignatureAdditive.lean:62`) rather than copying it."
    priority: should
    size: L
    tags: [lean, noncommutative, blocked]
    labels: [area:lean, type:feature, blocked]

Exported from .claude/roadmap/charge-abstraction.yaml at the commit that removed it from the tree.

Home

Architecture

CI and runners

Decision records

Roadmaps

Clone this wiki locally