slug: feldman-zigzag-conjecture-nine-three bibkey: feldman2026missingzigzag doi: null url: https://arxiv.org/html/2609.26114v1 triage: theorem motivation_gids:
- D5/S3/Combinatorics/Zigzag/FeldmanConjectureNineThree.feldman_conjecture_nine_three
Feldman’s Conjecture 9.3: balanced zigzag choices
Problem
David V. Feldman, The Missing Zigzag: Cycles of Semitone Trichords and a
Conservation Law in Equal Temperament, arXiv:2609.26114v1, Section 5,
equation (2), Conjecture 9.3. For each integer t>=2, set n=3t. For every
k=2,...,n-2, choose one labelled directed form in ZMod n:
| Label | From | To |
|---|---|---|
| I | 1 | k-1 |
| II | k | 1-k |
| III | -1 | k |
| IV | k-1 | -k |
| V | -k | 1 |
| VI | 1-k | -1 |
At k=2, exactly II, III, IV, and V are permitted. Labels remain different
choices when their endpoint pairs coincide. Balance means integer
outdegree minus indegree zero at each nonzero residue; no closedness,
connectivity, Eulerian-circuit, or cycle condition is added. Let a(t) be
the number of these actual balanced labelled choices. The full conjecture is
a(2)=4; for every r>=2, a(2r)=4a(2r-1); and for every r>=1,
(2r)a(2r+1)=6(2r-1)a(2r).
Motivation
The finite choice object and balance predicate in Choices.lean directly
encode the stated count. The endpoint in
FeldmanConjectureNineThree.feldman_conjecture_nine_three carries all three
clauses in a single theorem. Its two companion closed-count theorems establish,
for every r>=1, a(2r)=4*6^(r-1)*choose(2r-2,r-1) and
a(2r+1)=2*6^r*choose(2r-1,r), which are stronger than the recurrences.
Gap
The mathematical candidate has a kernel-checked all-parameter proof and an
independent source review for its exact source checkpoint. The repository
has canonical joint first-Freeze state files for the eleven Zigzag modules.
The endpoint’s typed Scribe Proved claim binds the sole Conjecture 9.3 result
to this dossier, and its emitted Markdown carries the matching resolution
marker. Current required CI, merge into dev, and completion audit remain
pending. Resolution is derived from the typed claim, not hand-written here.
The bounded prior-art screen in the Library note found no matching later
public proof in its checked sources. It did not establish global novelty or
priority. The author’s pinned repository treats a stricter finite BadPlan12
and lacks this all-parameter balance-only theorem.
Route
Inversion pairs the nonchromatic classes. PathData records five reachable
frontier states in each of two signed sectors, with twelve distinct labelled
interior transitions per sector. ChoiceClassification and Retirement
prove that old interior nodes cannot be touched by later edges, and that the
remaining local balance equations force the recorded transitions. PathEncoding
turns paths into labelled choices; DecodedBalance recovers their balance
and injectivity. EvenPaths handles the antipodal terminal pair and
OddPaths the singleton terminal, proving both literal choice/path
equivalences for all positive r.
WeightedPaths counts the positive-sector path charges in a finite Laurent
polynomial. Its scalar recurrence has seeds f_0=2, f_1=4z and
f_m=2z f_(m-1)+3z^(-1) f_(m-2). The negative sector is reached through a
label-aware reflection, not an identification of its form labels.
LaurentCoefficients solves the recurrence by a finite antidiagonal sum and
extracts the two required coefficients. The endpoint composes these with
the exact even and odd equivalences and the elementary binomial identities.
Falsifier
An omitted balanced choice, lost multiplicity from coincident endpoints, later edge touching a retired vertex, missing antipodal or singleton terminal, failure of either inverse law, coefficient mismatch, or extra source restriction would break this correspondence. A recurrence for a proxy count without the equivalences would not settle the question.
Evidence
For the source checkpoint d2621fdf37c370557f924c11b26ff706d0d4964b,
the source review approved all eleven Lean owners and all 68 theorems;
designated independent tests reported kernel exit 0 over 1489 jobs and
consumer checks at t=2,3,4. The reported axiom closure contains only
propext, Classical.choice, and Quot.sound. These are source-stage
readings; the companion, joint first Freeze, and typed binding have separate
local checks. Current required CI and completion audit remain pending.
Independent finite table and boundary tests found literal
balanced counts 4,12,48 at n=6,9,12; they do not replace the all-r
proof.
Triage
theorem; Tier 1 external named Conjecture 9.3, preregistered before the
probe. This record anchors only that conjecture, not the broader Open
Problem 13.1. The theorem’s content admission and source fidelity have
independent source-stage approval. Delivery and KPI credit require current
required CI, merge into dev, and completion audit; this record claims neither.
ASSUMED-UNVERIFIED
The bounded literature screen excludes neither private nor unindexed work, and OpenAlex/Semantic Scholar were rate limited. Worldwide absence, priority, final publication, and post-checkpoint future CI are not verified by this record. The Lean kernel verifies the formal statements, not their mapping to the published prose; that mapping has a separate source review.