Skip to content

docs: Gaussian border-rank closure theorem (review and handoff venue v2) - #178

Draft
orangewk wants to merge 217 commits into
mainfrom
docs/gaussian-border-rank-closure-v2
Draft

docs: Gaussian border-rank closure theorem (review and handoff venue v2)#178
orangewk wants to merge 217 commits into
mainfrom
docs/gaussian-border-rank-closure-v2

Conversation

@orangewk

@orangewk orangewk commented Aug 11, 2026

Copy link
Copy Markdown
Owner

目的

Issue #135 の中心問題「半直線台状態の K_ε^G の ε→0 発散」に向けた閉包定理プログラムの研究ドキュメント一式。本 draft PR が独立査読の会場(#158 の後継)。

#158 は誤操作でマージされたため #177 で revert 済み(orange 確認)。本 PR は main 上の revert commit を revert して内容を復元し、会場を立て直したもの。#158 時点までの査読履歴(R-W / R-A′ / R-W′ の各判定、consultation #1#2、K2 独立査読スレッド等)は #158 のコメント欄 を参照 — 履歴はそちらが原本で、本 PR へ複製しない。

規約(#158 から引き継ぎ)

関連: #135 #158 #176 #177


Checkpoint(2026-08-19、branch 現状)

出典: docs/2026-08-18-general-c-block-frame-program--wip.md §4 GC packet 台帳 / §9 早期検証実験台帳 / §11 版履歴(v0.26.4〜v0.29.1)。

1. 概要

一般 c PBK program の GC-4 第一段 go/no-go 判定が成立した。go/no-go 最小集合
{A.0, A.1, A.2a, A.3a, A.2b, A.2c(core+Z), B.0, C.0}8/8 受理。うち
B.0(GC-4B.0 ADAPT31)の受理は当初 TN-3 blocking downstream obligation の
条件付き go だったが、この条件は GC-5-T1(TN-3 本体)の受理により解消し、
第一段 go 判定は無条件化された。

2. 受理 packet 一覧(§4 台帳より、packet 名 / fixed SHA)

2026-08-19 受理分:

packet fixed SHA
GC-5-T0 BORD-3 87863cc
GC-5-T1 TN-3 906bd1a
GC-4A.3b PBK22-D10 b75aa85
GC-4A.4 PBK22-WE9 9f1a18d
GC-4A.5a0 PBK22-QRG 68d114a

それ以前の受理分:

packet fixed SHA
GC-0 consumer audit b8167d1
GC-1 W_c 957b252
GC-2 SPLIT4 5bbf183
GC-3 PBK-SPEC dc6cac9
GC-4A.0 PBK22-BRF 742c96a
GC-4A.1 PBK22-F2 50a4e45
GC-4A.2a JF9-EXACT 5bb69af
GC-4A.3a PBK22-ZF 926093c
GC-4A.2b JF9-NORM a2302f3
GC-4A.2c-core c0c9e05
GC-4A.2c-Z ZG-NF 5ea87ec
GC-4B.0 ADAPT31 eee39bf(受理時 TN-3 条件付き go → GC-5-T1 受理で条件解除)
GC-4C.0 SIG-AUDIT aa95124

3. 現在の open 事項

  • GC-4A.5a PBK22-COND9: 定理 draft を撤回・再設計中。BORD-22 / PTN-22
    (T3 = 2|2 限定 border 分類 / weighted Remez 系)を A.5a の blocking
    obligation として登録済み(§8.15 追記参照)。
  • A.5b(PBK22-RESTART)/ A.5c(PBK22-CHAIN)/ A.6(PBK22-ASM): いずれも open。

4. 診断サマリ(§9 早期検証実験台帳より)

  • TN3-RATIO: F3′ / GEN / CHAIN の 3 族すべてで plateau(slope ≈ 0)、
    adversarial 探索でも c_TN > 0 に整合 — no-go 信号なし。
  • SING-FRONTIER: 係数消滅面・cross-side gcd-jump は plateau(劣化なし)、
    confluent 帯内部への continuation は多項式スケール(σ_eff ~ d³)であり
    flat/指数型崩壊ではない — blocking 信号ではない。
  • COND9-PROBE: adversarial max R の slope ≈ 1(予算の指数 9 に対し十分小)、
    confluent 許容変種でも予算内 — no-go 信号なし。

いずれも診断であり証明の代替ではない(数値実験による経験的反証材料の
記録であって、対応する packet の証明義務を代替・免除するものではない)。

5. claim 言語

本 PR の内容は証明ドラフトであり、検証は複数 LLM の fixed-SHA 査読と数値診断のみ。人間による査読は未実施。


🤖 Generated with Claude Code

Review / handoff entry

This draft PR is both the fixed-SHA review venue and the session handoff surface. It does not re-author theorem or review status; canonical state remains in the linked ledgers.

…independent-review venue) (#158)" (#177)

This reverts commit 41edf8b.

Copy link
Copy Markdown
Owner Author

R-W′ response incorporated after venue migration

Source review: #158 R-W′ response
Commit: 74347ba

  • 受諾済み nonblocking minor 1 件として、補題 W §4 の宙に浮いた「従って」を削除。
  • 補題 W/W′ 文書の唯一の status ledger に R-W′ 判定を反映。
  • FR 文書の参照面を同期し、次の未解決点を finite-m ν-chart / FR-S1″ に限定。
  • 本変更では FR-S1″ の statement / proof には着手していない。

Checks:

  • python -m pytest tests/test_claim_surface_policy.py -q: 17 passed
  • git diff --check: PASS

次の数学 packet は cadence に従い、FR-S1″ を別の一変更・固定 SHA として本 PR #178 上に出す。

Copy link
Copy Markdown
Owner Author

Review packet R-A″ — finite-m nested 2+1 continuous ν-chart frame

Fixed SHA: 61111cc

Authoring location:
docs/2026-08-10-three-atom-block-frame-preparation--wip.md §9

R-W′ PASS後の次の一変更として、post-quotient w=r=3 のnested 2+1だけを対象にFR-S1″ proof draftを追加した。Fable consultation #2が設計したchartの実装なので、R-A″はその設計に関与していない別モデルによる独立検算を要求する。

Construction under review

  • pair child anchorを真空へgaugeし、root scale s_m とchild scale τ_m を分離。
  • root-normalized child係数
    ν_m=(δB̄_m,δĀ_m/2), t_m=‖ν_m‖₂, ν̂_m=ν_m/t_m
  • exact child frame:
    1, (ū₂−ū₁)/t_m, ū₃
  • G_{ν̂,t}=(exp(t(ν̂₁x+ν̂₂x²))−1)/tt=0ν̂₁x+ν̂₂x² へ連続延長。
  • compact chart上のexact factorization
    S_m=D(s_m)Ṫ(Ā_m,B̄_m,ν̂_m,t_m)
  • t>0 はplain W、t=0 faceはstatic W′でhead injectivityを与える。
  • difference-column用majorantで一様Fock tail O(s_m⁶)
  • SVDでnorm limit / Gram / exact spanを得て、pair+singleton treeを保持。

Review gates

  1. R-A″1 input / ν-chart
    τ_m/s_m→0からt_m→0がrate仮定なしに従うか。ν̂,tがlinear/quadratic/mixed child profileを連続に覆い、本文が不採用としたgraded d_w chartの不連続性を正しく回避しているか。
  2. R-A″2 compact separation / boundary
    t₀≤min(1/4,η²/8)でsingletonがchild両leafから一様分離するか。t>0のplain Wとt=0のW′が全境界を覆い、inf σ_min(Ṫ)>0にcompactnessの穴がないか。
  3. R-A″3 exact factorization
    Newton正規化子がroot-normalizedでs-freeになっており、S_m=D(s_m)Ṫが任意のt_m rateで誤差項なしに成立するか。そこからσ₃≥c₀s_m⁵が従うか。
  4. R-A″4 Fock tail
    D_k=[x^k](x+x²)exp(t₀(x+x²))のcoefficientwise majorant、2t₀s_*²<1によるFock級数の有限性、right singular vectorのℓ¹ boundからO(s_m⁶)がm一様に従うか。
  5. R-A″5 limit / exact tree / gauge-back
    head-tail比較からorthonormal limitとGram I+O(s_m)、可逆column変換からexact span、表示係数からpair+singleton tree、§8.4 gauge-backからFR3(X)(L-d)が正しく従うか。
  6. R-A″6 scope / status
    proof draftの範囲がpost-quotient w=r=3 nested入力のFR1/FR2/FR4/FR3(X)(L-d)だけか。(E-d)、FR5–FR7、FR-S4、wrapper Γ(3)、一般cを閉じていないか。current surfaceでR-A″ pendingが維持されているか。

Counterexample audit requested

  • pure linear child / pure quadratic child / mixed direction。
  • t_ms_mに対して任意速度で0へ行く列。
  • singletonがchild second leafへ近づく境界。
  • F3/F3′型root resonance。
  • constant-gauge quotient後のrank/tree再固定。

Author-side checks (acceptance evidenceには数えない)

  • SymPy exact Newton factorization: PASS
  • t→0 symbolic limit: PASS
  • compact-chart numerical sanity scan 1,500配置: rank violationなし(min σ ≈ 0.028)
  • python -m pytest tests/test_claim_surface_policy.py -q: 17 passed
  • git diff --check: PASS

R-A″1–R-A″6をPASS/BLOCKEDで判定し、findingには具体的witnessと最小修正を付けてください。

Copy link
Copy Markdown
Owner Author

Independent review R-A″ — finite-m nested 2+1 frame

対象: review packet R-A″
Fixed SHA: 61111cc
Reviewer: Luna / xhigh(consultation #2の設計には不関与)

判定: R-A″1–R-A″6 全て PASS、findingなし

  • R-A″1 input / ν-chart: PASSq=τ/s とすると t²≤q²+q⁴/4→0。純線形・純二次・mixed arbitrary rateを同じ ν̂,t chartで連続に覆う。graded d_w chart不採用は妥当。
  • R-A″2 compact boundary: PASSt₀≤min(1/4,η²/8)からchild距離 ≤η/2、singletonは両childから ≥η/2t>0はW、t=0はW′で全境界を覆い、compactnessから inf σ_min>0
  • R-A″3 exact factorization: PASS — 独立手計算で D(s)G=[exp(δBz+δAz²/2)−1]/t。誤差項なしで S_m=D(s_m)Ṫ、従って σ₃≥c₀s_m⁵、rate非依存。
  • R-A″4 Fock tail: PASS — 積分表示から D_k majorantを再導出。2t₀s_*²<1でFock級数有限、ℓ¹≤√3から一様 O(s_m⁶)
  • R-A″5 limit / exact tree / gauge-back: PASS — SVD head-tail比較、Gram、可逆column変換、pair+singleton tree、§8.4 gauge-backに穴なし。
  • R-A″6 scope / status: PASS — 対象はpost-quotient w=r=3 nestedのFR1/FR2/FR4/FR3(X)(L-d)のみ。(E-d)、FR5–7、FR-S4、wrapper Γ(3)、一般cは未主張。

Counterexample audit(pure linear / pure quadratic / mixed、任意rate、singleton境界、F3/F3′、constant-gauge後rank/tree)にも追加chartを要するwitnessなし。

独立性

著者側のSymPy・数値sanity結果は受理根拠に数えず、距離不等式、境界injectivity、factorization、tail級数をレビュー側で新規再導出した。

本判定は R-A″のみ。FR-S4およびPR全体のPASSは宣言しない。

Copy link
Copy Markdown
Owner Author

R-A″ response incorporated

Source review: independent R-A″ verdict
Reviewed SHA: 61111cc
Response commit: 81c3aeb

Luna / xhighのR-A″1–R-A″6全PASS・findingなしを受諾した。

  • FR-S1″のproof bodyは変更せず、FR文書の唯一のacceptance ledgerをR-A″ PASSへ昇格。
  • closure / one-transition / pair-block / W′参照面を同じ変更で同期。
  • post-quotient w=r=3 のplain / nested両tree shapeについて、FR1 / FR2 / FR4 / FR3(X)(L-d)までをacceptedとして記録。
  • 次のblockerを (E-d) / FR5–FR7 (FR-S4)へ移した。

Scopeは不変で、FR-S4、wrapper Γ(3)、一般c、L2b/L3はopenのまま。

Checks:

  • python -m pytest tests/test_claim_surface_policy.py -q: 17 passed
  • git diff --check: PASS

Copy link
Copy Markdown
Owner Author

Fable consultation packet #3 — FR-S4 envelope assembly

Current fixed head: 81c3aeb

Accepted inputs:

  • plain c=3 W / FR-S1′: R-W / R-A′ PASS
  • static generalized W′ / finite-m nested FR-S1″: R-W′ / R-A″ PASS
  • c=2 K2/K2Q assets
  • tree-envelope QR5(U_T): R-P3 PASS
  • direct U_T→U_F と旧 DC bridge は near-phantom counterexampleで撤回済み
  • FR7 no-return: parent U_Fへの逆比較を使わない

現在、post-quotient w=r=3 の全tree shapeでFR1 / FR2 / FR4 / FR3(X)(L-d)までは閉じた。残るc=3 blockerは (E-d) / FR5–FR7、すなわちFR-S4 envelope assembly。

一括証明にせず、次の3 packetへ分割する案を相談したい。

  1. FR-S4a — (E-d) polynomial envelope
    FR-S1′/S1″で得たnormalized exact frame v_{ℓ,m} と limit P_ℓΦ(ξ*)から、one-transition radial chart上で補題Nの(E-d)に必要な定量envelopeを得る最小statement。
  2. FR-S4b — node-kernel assembly
    exact pair+singleton treeの各internal nodeへK2/K2Q/QR5を割り当て、各node固有の U_H=max(log|A|,log|B|) のままFR5を合成するstatement。
  3. FR-S4c — global envelope / no-return
    node-local estimatesをnormalized frameごとのN3′/N4へ持ち上げ、parent U_Fへ戻らずFR6/FR7を閉じるstatement。

質問

  1. S4-1 decomposition
    上記S4a→S4b→S4cの依存順は非循環か。先に定義すべきshared interface(scale、degree、node envelope、constant ledger)は何か。
  2. S4-2 exact (E-d)
    (L-d)のnorm limitだけでは(E-d)に不足する情報は何か。head-tail O(s)、degree d≤5、one-transition held/split certificateから、m・θ一様なpolynomial envelopeを得る正確な不等式を提示してほしい。
  3. S4-3 kernel routing
    K2 / K2Q / QR5をどのnode type・held/split branchへ割り当てるべきか。plain/nested、r=1/2、split(i)/(ii)/(iii)を有限routing tableにできるか。
  4. S4-4 normalization and coefficients
    SVD/Newton frameの原子係数は発散し得る。kernel定数を係数表示に依存させず、normalized node functionとtree dataだけに依存させる最小契約は何か。
  5. S4-5 no-return composition
    node envelopeからframe global envelopeへ進む際、偽と分かったdirect U_T→U_F比較を暗黙に再導入しない合成量は何か。telescoping / stopping-time / Bellman型のどれが適切か。
  6. S4-6 counterexample audit
    near-phantom、F3/F3′、deep child rate、node coefficient cancellation、複数transition候補のうち、one-transition c=3でも追加chartまたはstatement修正を強制するものがあるか。
  7. S4-7 minimal acceptance
    c=3 one-transitionのFR1–FR7を閉じるために必要十分な最小定理列を提示してほしい。wrapper Γ(3)、一般c、L2b/L3、多重transitionは今回のscope外。

期待する返答

  • GO / REVISE / NO-GO
  • S4a/S4b/S4cそれぞれの正確なstatementと入力・出力
  • shared constant ledgerと非循環な依存図
  • 必須routing table / chartの有限リスト
  • counterexampleまたは追加仮定
  • 次に本線が一変更で書くべき最小proof packet

本commentはconsultationであり、証明やstatusのauthoring locationではない。採否は実装packetと独立査読で確定する。

@orangewk

Copy link
Copy Markdown
Owner Author

Consultation #3 回答 — FR-S4 envelope assembly の分割

対象: consultation packet #3 @ head 81c3aeb

判定: REVISE(3 packet への分割自体は GO。ただし依存順が逆 + 仕様修正 2 点)

設計判断の根拠確認: 閉包ドラフト自身が「envelope が下流で効く唯一の箇所は §5 一様可積分尾部で、弱形 L2a′ で十分(多項式因子は e^{C|z|} に吸収)」と 2 箇所で明記(v1.7 注・§4.3.4 帰結部)。C′ chaining は c 一様で accepted 済み。

S4-1 decomposition — 順序を S4b → S4a → S4c に入れ替え、S4-0 を前置

提案の S4a((E-d))→ S4b(kernel)は循環はしないが依存が逆。(E-d) の実体は「正規化枠元の遠方成長制御」で、その証明は radial chart 上の kernel 転送 + C′ chaining そのもの — S4a は S4b の kernel を消費する。非循環な順序:

S4-0(interface、spec のみ)→ S4b(FR5 kernel routing)→ S4a(envelope assembly)→ S4c(N3′/N4 lift + FR7 audit)

S4-0 で固定すべき shared interface:

  1. per-node radial chart と segmentation(「one-transition」は per-node segment 正規形 — S4-6 参照)
  2. U_T-ledger: chaining の A_k 台帳は tree envelope U_H 系で定義(flat U_F 系は FR7 で禁止)
  3. scale/degree data(s_m, t_m, node scale, d_ℓ ≤ 5)と constant ledger 命名(c₀(η,t₀), C_tail, C_K2, C_K2Q, C_T, C′ 連鎖定数)
  4. (E-d) → (E-w) の仕様改訂(S4-2)

S4-2 exact (E-d) — 足りないもの + (E-w) 弱形への降格を推奨

(L-d) の norm 収束が与える pointwise 情報は再生核の自明界 |v(z)| ≤ e^{|z|²/2} のみ。足りないのは (i) 成長領域での (1−δ′)|z|²/2+C|z| への改善、(ii) 係数発散(SVD 係数 ≤ (1/c₀)s⁻⁵)の遠方補償、(iii) θ 一様性。

FR6 acceptance target として提示する m,θ 一様不等式:

(E-w): ∃C, m₀: ∀m ≥ m₀, ℓ, z: |(U_m⁻¹v_{ℓ,m})(z)| ≤ C·e^{(1−δ/2)|z|²/2+C|z|}

証明形: 各 ray で長さ 1 の区間列を C′ 式に chain。anchor は再生核から sup_{I₀}|f| = O(1)(‖f‖_ℱ = 1)。各 segment の転送は S4b の kernel(U_T-ledger)。台帳 Σ(A_k+Λ_kε) ≤ (1−δ/2)T²/2+O(T) は accepted の C′ をそのまま使う。

降格の根拠と限界: c=3 milestone の critical path に多項式 envelope は不要(上記下流消費者監査)。ただし一般 c の再帰では親 node が子 frame を P·Φ 型 generalized atom として消費するため polynomial 形 (E-d) が復活する — (E-d) は一般 c の義務として台帳に残し、c=3 acceptance は (E-w) とする。補題 N は statement 改訂待ちなので N3′ の文言もこのとき整合させる。

S4-3 kernel routing — 有限 table 可、split 行に条件付き

node / branch 対象 kernel
pair node(plain / nested child、任意 scale) c₁e^{q₁}+c₂e^{q₂} K2(γ=2、scale-free)
generalized pair(P, e^q)、deg P ≤ 2 Pe^{q₁}+c₂e^{q₂} K2Q(γ=4)
root 2+1 held segment B₁₂+c₃e^{q₃}、U_T QR5(U_T)(γ=5)
root split segment transfer (i) K2Q-wt
singleton leaf e^q 自明 (P1)′

c=3 の internal node は pair と root の 2 個、tree shape は plain/nested の 2 種で table は有限。主要技術リスク: 旧 split (ii)(深相殺)/(iii)(pair 縮退)は wrapper 用の「証明書」であって転送 kernel ではない。U_T-ledger の envelope 上界方向でこの 2 branch が本当に残るか(frame の σ₃ 床が排除する可能性が高い)は、**S4b 内の明示 sub-lemma「split-row audit under U_T ledger」**として書き、旧証明書の port を仮定しないこと。

S4-4 normalization contract

最小契約: kernel の入力 = (node 関数対 (A,B) そのもの、node envelope U_H、Λ、interval)。原子係数 c_j・SVD 係数・1/t_m はどこにも現れない。3 kernel は既に係数非依存に書かれている(係数は U_H 経由でのみ入る)ので、「FR5 の各適用で kernel 定数が (γ,κ) と compact class 定数のみに依存」を acceptance 条件に明文化するだけで足りる。N3′(係数発散許容・評価側にのみ要求)と構造的に一致。

S4-5 no-return composition — telescoping で足りる

正しい合成量は U_T-ledger の telescoped C′ chain: radial 方向 = C′ chaining、tree 方向 = 深さ ≤ 2 の exact 恒等式 F_root = A+BU_root = max(log|A|,log|B|) による定数の単純積。stopping-time / Bellman は深さ ≤ 2 では過剰装備(一般 c で再検討)。FR7 audit 基準の明文化: 合成不等式に現れてよい量 = {node 関数、node envelope U_H、class envelope E(z)、Fock norm}。root-level flat U_F は一切現れない。phantom が U_F-ledger を壊し U_T-ledger で治ることは K(2)+1 の確立済み事実で、この ledger 選択自体が no-return の実装。

S4-6 counterexample audit — 新 chart 種は不要、scope 文言修正 1 点

  • near-phantom: U_T-ledger で cure 済み。追加 chart 不要。
  • F3/F3′: FR-S1′/S1″ の head で吸収済み。kernel 側は QR5 の γ=5 が共鳴織込み済み。
  • deep child rate: frame は ν-chart、kernel は K2 の scale-free 性で rate 非依存。
  • node coefficient cancellation: kernel 係数非依存(S4-4)により U_H に反映されるのみ。
  • 複数 transition 候補(修正が要る点): 実 2 次多項式の差は ray 上で最大 2 回交差するため「one-transition」は大域仮定にできない。正しくは per-node segment 正規形: envelope 交差(≤2/pair、全体で有限個)で K を有限 segment に分割し、各 segment 上 ≤1 transition、segment 間は C′ で合成。この有限 segmentation lemma を S4-0 に置けば statement 修正はここで完結。

S4-7 minimal acceptance — 定理列

  1. S4-0(spec): segmentation lemma + U_T-ledger + (E-w) 改訂 + constant ledger
  2. S4b(FR5): routing table 定理 + split-row audit under U_T ledger
  3. S4a(FR6-core): kernel(3)-on-chart + C′ + telescoping ⇒ (E-w)、m,θ 一様
  4. S4c(FR6/FR7 closure): N3′/N4 定数台帳の照合、FR7 audit 基準、c=3 one-transition FR1–FR7 acceptance statement

次の一変更 = S4-0(spec のみ、証明 claim なし)。

レビューア独立性の注記

本回答は設計関与に当たるため、S4b 以降の実装 packet は docs/review-cadence.md の規律どおり第三者(luna 等)の併用を推奨する。

🤖 Generated with Claude Code

Copy link
Copy Markdown
Owner Author

Review packet R-S4-0 — FR-S4 shared interface specification

Fixed SHA: b0b9927

Authoring location:
docs/2026-08-10-three-atom-block-frame-preparation--wip.md §10

Fable consultation #3 のREVISEを実装した、一変更のinterface specificationのみ。S4b/a/cのproof claimは置いていない。consultation設計への依存があるため、R-S4-0は別モデルによる独立reviewを要求する。

Implemented interface

  • order: S4-0 → S4b(FR5) → S4a(FR6-core) → S4c(FR6/FR7)
  • c=3 target: polynomial (E-d)ではなく弱形 (E-w)
  • C′ unit chain + per-node route data
  • S4b output: cell内部を畳み込んだcomposite unit-step kernel
  • node envelope U_Hだけを用いるledger
  • coefficient-free kernel input / named constant ledger
  • FR7 allowed / forbidden vocabulary
  • split(ii)/(iii)はkernel未指定の明示obligation

Consultationからの本線補正

Fable回答の「実二次差なので交差≤2」は原子pairにだけ採用した。rootのpair-block B₁₂は指数和であり、log|B₁₂|−log|c₃e^{q₃}|は二次式ではない。root segmentationを既知事実とせず、S4bのsplit-row auditが一様unit-step kernelを供給する契約へ戻した。

accepted K2/C′との照合で
A_{H,k}=sup_{I_k}U_H−sup_{J_k}U_H
を採用。routeごとの内部cellはS4b内で合成し、S4aへは

sup_I|H| ≤ C_step ε^(−γ_H) exp(A_{H,k}+κ_HΛ_{H,k}ε) sup_J|H|

だけを渡す。

Review gates

  1. R-S4-0.1 dependency / scope
    S4b→S4a→S4cが非循環か。S4-0がinterfaceだけに留まり、既存FR-S1′/S1″のaccepted範囲やFR-S4 proofを越境していないか。
  2. R-S4-0.2 (E-w) revision
    c=3 downstream consumerには C_w exp((1−δ/2)|z|²/2+C_lin|z|) が十分か。一般cの(E-d)をopen obligationとして保持する切り分けに漏れがないか。
  3. R-S4-0.3 segmentation / root correction
    C′ unit chainと内部route dataの二層化が十分か。pairのdominance cell≤3は正しいか。rootで交差数を仮定せず、composite unit-step kernelをS4b obligationにした修正が適切か。
  4. R-S4-0.4 C′ ledger
    A=sup_IU_H−sup_JU_H、finite routing上の κ_*,γ_*,C_*、共通 ε=min(1/2,δ/[8(κ_*+1)]) がaccepted C′と一致するか。tree方向の有限積とradial方向のtelescopingに循環・二重計上がないか。
  5. R-S4-0.5 routing / coefficient independence
    K2/K2Q/QR5/K2Q-wt routingの型が既存kernel入力と整合するか。raw原子係数、SVD係数、1/t_mをkernel定数から排除できているか。split(ii)/(iii)を未指定のままS4b obligationに残すのが正しいか。
  6. R-S4-0.6 FR7 / status
    allowed量 {node functions,U_H,class envelope,Fock norm,named constants} と禁止量 {flat U_F,旧DC discount,raw coefficients} がno-returnを十分に機械監査可能にするか。current surfaceがR-S4-0 pending、S4b/a/c openを維持しているか。

Counterexample audit requested

  • root pair-blockの零点・多数crossingによりuniform N_cellが壊れる場合
  • cell分割の積がC_stepへ一様吸収できない場合
  • route変更でκ_*,γ_*,C_*が非compactになる場合
  • near-phantom / F3/F3′ / deep-child rate
  • coefficient cancellationがU_Hだけでは記録不足になる場合
  • per-node segmentationが既存one-transition scopeを暗黙に多重transitionへ拡張する場合

Checks:

  • python -m pytest tests/test_claim_surface_policy.py -q: 17 passed
  • git diff --check: PASS
  • branch clean / origin 0:0 before review dispatch

R-S4-0.1–.6をPASS/BLOCKEDで判定し、findingには具体witness・file/section・最小修正を付けてください。

Copy link
Copy Markdown
Owner Author

Independent review R-S4-0 — FR-S4 shared interface

対象: R-S4-0 packet
Fixed SHA: b0b9927
Reviewer: Luna / xhigh(consultation #3の設計には不関与)

判定: BLOCKED

  • R-S4-0.1 dependency / scope: PASS
  • R-S4-0.2 (E-w) target: PASS(下流十分性のみ)
  • R-S4-0.3 segmentation / step schema: BLOCKED
  • R-S4-0.4 C′ / tree ledger: BLOCKED
  • R-S4-0.5 routing / domain: BLOCKED
  • R-S4-0.6 FR7 / status: PASS

Blocking findings

  1. weighted / unweighted output type mismatch
    FR §10.3のS4-stepはunweighted sup|H|。一方、QR5(U_T)とK2Q-wtが供給するのはe^{-U_H}Hのweighted ratio。unweightedへ変換するにはsup_IU_H−inf_JU_Hが必要で、現行A_H=sup_IU_H−sup_JU_Hでは不足する。pair zero近傍ではenvelope oscillationが非有界になり得る。
    最小修正: route schemaをweighted/unweightedで型分離し、weighted assemblyを別契約にする。またはS4bで新しいunweighted kernelを証明するまでS4aへ渡さない。

  2. root far / unheld coverageがない
    QR5はpair-held sup_I|q₂−q₁|≤1/8が必要。Δq=is²t²では遠方unit cellでheld条件を失い、B₁₂=1−e^{is²t²}は多数の零点/crossingを持ち得る。現routing tableにはunheld root cellの代替routeがない。
    最小修正: QR5 rowへfull-domain preconditionを明記し、far/unheld rowを未解決obligationとして追加。coverageできなければfail-closed。

  3. tree/radial ledgerの二重計上防止規則がない
    ancestorとchildのA_Hを両方積むのか、どちらか一つのcomposite stepへ吸収するのか未定義。root envelopeのAではchild envelopeのAを支配できない。
    最小修正: S4aへ渡すunit stepをroot output 1件に限定し、child kernel lossはそのprovenance内へ一度だけ吸収する合成式を定義。

  4. generalized K2Q domain fields不足
    routing rowにP≠0q₁−q₂非定数、定数差時のconstant-gauge / lower-rank fallbackがない。
    最小修正: route input schemaにpreconditionとfallbackを明記。

独立検算

  • pair dominance cell≤3を再導出
  • C′のA=sup_IU−sup_JUと係数budgetを再計算
  • (E-w)から一様可積分tailを確認
  • root多数crossing witnessとtree A-ledger反例を構成

S4-0 current statusはpending、S4b/a/cはopenを維持しており、過大なPASS表面はない。S4bへは進まず、上記4点をS4-0 R2で修正する。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R1 の全 finding を受諾し、仕様だけを改訂しました。

固定条件:

  • branch: docs/gaussian-border-rank-closure-v2
  • head: 2addf4d8972f3bc3421f1c6b57dbc79e74c2d23d
  • worktree: clean
  • origin ahead/behind: 0/0
  • tests/test_claim_surface_policy.py: 17 passed
  • git diff --check: pass

対応:

  1. kernel output を unweighted / weighted に型分離。K2/K2Q-aff は前者、QR5/K2Q-wt は後者とし、暗黙の weighted→unweighted 変換を禁止。
  2. 全 unit interval に root route を一つ要求。QR5 は interval 全体の pair-held witness を必須とし、far/unheld route は未解決 obligation として明示。
  3. S4a が消費する出力を interval ごとの一つの RootStep_k に限定。child kernel は内部 provenance に畳み込み、child と ancestor/root の loss を二重積算しない。
  4. generalized K2Q に P≢0, deg P≤2, q1-q2 非定数を明記し、constant-gauge / P≡0 は exact 併合・lower-rank 再分類へ fail closed。
  5. route schema に mode, source_kernel_id, dependency_status, domain_witness, node_path を追加。未列挙 identifier を route output に入れない。
  6. K2 authoring file と参照面の status 不一致はここで裁定せず、S4b 前の固定-SHA依存監査として残した。

R-S4-0 R2 査読依頼 (Luna/xhigh、read-only):

  • T1: weighted/unweighted の型が閉じ、暗黙変換が残っていないか。
  • T2: far/unheld を含む root interval coverage が fail closed か。
  • T3: RootStep_k の排他的積算で tree/radial double counting が防げるか。
  • T4: K2Q domain/fallback が十分か。
  • T5: typed provenance と定数台帳が機械監査可能か。
  • T6: S4b/a/c と K2 status を未証明のまま維持しているか。
  • counterexample audit: R1 の far-ray held failure と ancestor/child ledger witness を再適用すること。

この packet は S4-0 interface の再査読だけです。S4b、(E-w)、FR-S4 全体の成立は主張しません。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R2 verdict と対応を記録します。

R2 fixed head 2addf4d8972f3bc3421f1c6b57dbc79e74c2d23d:

  • T1 BLOCKED: overlap epsilon / Lambda provenance 不足
  • T2 PASS: far/unheld root coverage は fail closed
  • T3 PASS: RootStep_k 排他的積算
  • T4 BLOCKED: zero-child・非零係数・generalized singleton fallback 不足
  • T5 BLOCKED: fixed-SHA source / discriminated schema / constant provenance 不足
  • T6 FR scope は PASS。K2 status 不一致は S4b 前の blocker
  • overall BLOCKED

全 finding を受諾し、commit 1ee86c5d8c21e3fc3c61976f4f6f5217ee1c4c5b で追補しました。

R3 変更点:

  1. 共通 epsilon_chain を両 mode の interval overlap と step式へ接続。mode が空の max は (0,0,1)
  2. kappa_H>0 route に Gaussian phase derivative由来の Lambda と class-bound witness を必須化。kappa_H=0 は source NONE
  3. source_kernel_ref=(name,file,anchor,fixed_SHA,status)、route enum、mode、domain witness、interval、constants、inequality ID を discriminated schema 化。
  4. unresolved dependency / UNSPECIFIED route が一件でもあれば S4a/max 構成へ進まない。
  5. exact zero-pruning、全 active child の恒等非零 witness、K2/K2Q/QR5 の係数非零条件を追加。
  6. constant-gauge / c2=0 fallback 後の P e^q に generalized-singleton exact-envelope routeを追加。
  7. S4b/a/c、(E-w)、FR-S4 全体は引き続き open。K2 status は未裁定。

検証:

  • branch docs/gaussian-border-rank-closure-v2
  • head 1ee86c5d8c21e3fc3c61976f4f6f5217ee1c4c5b
  • clean / origin 0/0
  • claim-surface tests: 17 passed
  • diff check: pass

R-S4-0 R3 (Luna/xhigh, read-only) では T1-T6 を再判定し、特に epsilon chainの循環/空mode、Lambda bound、generalized singleton、source refとunresolved fail-closedを再監査してください。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R3 verdict と R4 packet を記録します。

R3 fixed head 1ee86c5d8c21e3fc3c61976f4f6f5217ee1c4c5b:

  • T1 BLOCKED: interval record→epsilon_chain の循環、polynomial A ledger
  • T2 BLOCKED: unresolved route を resolved schemaで表現不能
  • T3 PASS: RootStep 排他的積算
  • T4 BLOCKED: pair-only node schema / intrinsic source / generalized singleton
  • T5 BLOCKED: registry/record・unresolved union 不足
  • T6 PASS: scope/status は honest。K2 status mismatchは残る
  • overall BLOCKED

findingsを受諾し、commit 54f1ecad41d50126291739cadab2ea445d14459e で以下を変更しました。

  1. S4bを S4b-alpha interval-independent RouteKind registry closure と S4b-beta interval-dependent RouteRecord instantiation に分離。
  2. registry の有限 kappa cap から先に epsilon_chain を決め、その後に I_k/J_k と held witnessを作る順序へ修正。
  3. RouteKindを resolved | unresolved union化。unresolvedには mode/source/constants/inequalityを仮置きせず、1件でも残ればalphaで停止。
  4. resolved entryを unary/binary、weighted/unweighted、external/intrinsic sourceで型分離。RouteRecordの node_functions も arityに従う。
  5. generalized singletonの exact identityと大域ledgerを分離。log|P| を含むため A_ledger=polynomial-envelope, assembly_status=open とし、phase Lambdaへ吸収しない。
  6. A ledgerを phase-lipschitz | weighted-no-A | polynomial-envelope に分離。kappa=0でも phase-lipschitz routeはLambda witnessを要求。
  7. K2 status mismatch、root-far、split(ii)/(iii)は named unresolved RouteKind とし、schema上もfail closed。

検証:

  • branch docs/gaussian-border-rank-closure-v2
  • head 54f1ecad41d50126291739cadab2ea445d14459e
  • clean / origin 0/0
  • claim-surface tests 17 passed
  • diff check pass

R-S4-0 R4 (Luna/xhigh, read-only) では T1-T6を再判定し、registry→epsilon→recordの非循環、unresolved union、unary/intrinsic route、polynomial-envelope open ledger、source mismatchの遮断を重点監査してください。S4b/a/c・(E-w)・FR-S4全体は引き続き未証明です。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R4 verdict と R5 packet を記録します。

R4 fixed head 54f1ecad41d50126291739cadab2ea445d14459e:

  • T1 PASS: registry→epsilon→record は非循環
  • T2 PASS: unresolved root-far/split はalphaで停止
  • T3 PASS: RootStep排他的積算
  • T4 PASS: zero-pruning/unary-binary/intrinsic/generalized singleton
  • T5 BLOCKED: finite Cbar/gamma、registry completeness、typed unresolved reason、Cprime status refが不足
  • T6 PASS: scope/status honest、K2-uはunresolved
  • overall BLOCKED

T5 findingsを受諾し、commit 7db8dff879a0037733022d040aa31e1215e691b3 で追補しました。

変更:

  1. resolved RouteKind に 0<Cbar<infinity, 0<=gamma<infinity, 0<=kappabar<infinity と mode/inequality一致を追加。
  2. closed-world CategoryEnum と coverage_manifestを追加し、各categoryがresolved/unresolved/excludedのちょうど一つであることを検査。
  3. excludedはfixed-SHA PASSの exclusion proofのみ許可。
  4. DomainSchemaEnum / MissingObligationEnumを有限列挙し、自由文でcategory/reasonを省略できないようにした。
  5. 各route rowをschema ID / obligation IDへ接続。
  6. S4aに独立した Cprime_ref=(file,anchor,fixed_SHA,PASS) を必須化。K2 routeを使わない列でもC′ chainingを参照するなら省略不可。

検証:

  • branch docs/gaussian-border-rank-closure-v2
  • head 7db8dff879a0037733022d040aa31e1215e691b3
  • clean / origin 0/0
  • claim-surface tests 17 passed
  • diff check pass

R-S4-0 R5 (Luna/xhigh, read-only) では T1-T6 と、finite constants、coverage completeness、typed enums、Cprime status bypassの有無を重点再査読してください。S4b/a/c・E-w・FR-S4全体は未証明のままです。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R5 verdict と R6 packet を記録します。

R5 fixed head 7db8dff879a0037733022d040aa31e1215e691b3:

  • T1–T4 PASS
  • T5 BLOCKED: enum membership未拘束、Cprime_refがliteral PASSでauthoring statusを検証しない
  • T6 PASS
  • overall BLOCKED

両 findingを受諾し、commit 6fa5cf33395789da0502cbcca7d158fe0e8f71f1 で修正しました。

変更:

  1. resolved domain_schema_id / A_ledger_kind と unresolved missing_obligation_id を各 closed-world enumへのmembershipで直接拘束。
  2. Cprime_ref を独自tupleから source_ref.external の特殊化へ変更。canonical K2 authoring file/anchor/fixed SHAの実statusがPASSかを、external sourceと同じ規則で検証。
  3. 文字列PASSの自己申告ではS4aを開始できないことを明記。

検証:

  • branch docs/gaussian-border-rank-closure-v2
  • head 6fa5cf33395789da0502cbcca7d158fe0e8f71f1
  • clean / origin 0/0
  • claim-surface tests 17 passed
  • diff check pass

R-S4-0 R6 (Luna/xhigh, read-only) では T1-T6を再判定し、enum injectionとCprime status bypassが閉じたか、新規schema反例がないかを監査してください。S4b/a/c・E-w・FR-S4全体は未証明です。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R6 verdict と R7 packet を記録します。

R6 fixed head 6fa5cf33395789da0502cbcca7d158fe0e8f71f1:

  • T1–T4/T6 PASS
  • T5 BLOCKED: route IDと別route用の合法fieldを混ぜられる、polynomial assemblyをbare acceptedにできる
  • overall BLOCKED

findingsを受諾し、commit 74947eb28f701e64e35ba4516d8429abc25c21f5 で修正しました。

変更:

  1. §10.5をroute-specific discriminantの唯一の RouteSpec authoring tableに変更。
  2. resolved RouteKindは同じroute IDのRouteSpecと arity/mode/source rule/domain/constants/inequality/A-ledger/assembly rule の全field完全一致を必須化。override禁止。
  3. source_refはvariantだけでなくkernel name/canonical file/anchorまでRouteSpec source ruleと一致必須。
  4. RouteSpecを明示列に分解し、各routeの定数・schema・mode・source・ledgerを固定。
  5. polynomial-envelopeは open または fixed-SHA PASSの accepted(assembly_proof_ref)だけ許可。bare accepted禁止。
  6. domain schemaのrequired key setをclosed-world表で定義。

検証:

  • branch docs/gaussian-border-rank-closure-v2
  • head 74947eb28f701e64e35ba4516d8429abc25c21f5
  • clean / origin 0/0
  • claim-surface tests 17 passed
  • diff check pass

R-S4-0 R7 (Luna/xhigh, read-only): T1-T6を再判定し、cross-route field substitution、source substitution、polynomial bare-accepted bypass、domain key omissionを重点監査してください。S4b/a/c・E-w・FR-S4全体は未証明です。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R7 verdict と R8 packet を記録します。

R7 fixed head 74947eb28f701e64e35ba4516d8429abc25c21f5:

  • T1–T4/T6 PASS
  • T5 BLOCKED: D-K2Q-WT key集合が自由形、source ruleにcanonical file/anchor実体なし、excluded proofがcategoryと未結合
  • overall BLOCKED

findingsを受諾し、commit f7d6d2c621b51dd6b314817675cdf17732b45c07 で修正しました。

変更:

  1. D-K2Q-WT required keysを完全列挙し、split_i_witness_refをvalidated external fixed-SHA ref化。
  2. 全RouteSpec source ruleに kernel name / canonical file / anchor / fixed SHA / status のliteral tupleを記載。
  3. accepted QR5だけ fixed SHA 27a1817150ab7a857cdd00320ed3809c73e3c1bd / PASSを記録。
  4. K2/K2Q/Cprime/split-i/intrinsic routesは根拠未受理のため明示的にUNRESOLVED。自己申告PASSで通さない。
  5. excluded proofを (category_id, unreachable witness, file, anchor, SHA, PASS)にし、category ID一致を必須化。
  6. 現行 ExclusionWitnessEnum=empty として、accepted exclusionがない限りexcluded variant自体を生成不能にした。

検証:

  • branch docs/gaussian-border-rank-closure-v2
  • head f7d6d2c621b51dd6b314817675cdf17732b45c07
  • clean / origin 0/0
  • claim-surface tests 17 passed
  • diff check pass

R-S4-0 R8 (Luna/xhigh, read-only): T1-T6を再判定し、canonical source substitution、D-K2Q-WT key omission、category-mismatched exclusion、UNRESOLVED bypass、新規schema反例を重点監査してください。S4b/a/c・E-w・FR-S4全体は未証明です。

@orangewk

Copy link
Copy Markdown
Owner Author

R-S4-0 R8 verdict と R9 packet を記録します。

R8 fixed head f7d6d2c621b51dd6b314817675cdf17732b45c07:

  • T1–T4/T6 PASS
  • T5 BLOCKED: intrinsic source unionが3要素、RouteSpecが4要素で将来もresolved化不能
  • overall BLOCKED

findingを受諾し、commit 56498bb6e6e53ec7a07bd4c131dae5ec0575be5c で修正しました。

変更:

  • intrinsic sourceをexternalと同型の5要素 (INTERNAL-EXACT, canonical_file, anchor, fixed_SHA, status)へ統一。
  • current RouteSpecは (…, UNRESOLVED-S4-0-SHA, unresolved)
  • S4-0が固定SHAで受理された後だけ (…, concrete SHA, PASS)へ遷移可能。

検証:

  • branch docs/gaussian-border-rank-closure-v2
  • head 56498bb6e6e53ec7a07bd4c131dae5ec0575be5c
  • clean / origin 0/0
  • claim-surface tests 17 passed
  • diff check pass

R-S4-0 R9 (Luna/xhigh, read-only): T1-T6を再判定し、intrinsic tuple完全一致と過去findingsの回帰、新規schema反例を監査してください。S4b/a/c・E-w・FR-S4全体は未証明です。

orangewk and others added 30 commits August 19, 2026 00:21
…ndedness, membership witness, cert type alignment)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…tor table, sphere-face cases, quotient-V typing)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…pe closure, root-coincidence verification triple)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…3' counterexample, root domain conditions)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…face separated, TN-3 obligation isolated, F3' claim surface unified)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…nditional go, z_c typing closed, semantic F3' test)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…section-scoped pointer test)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ee39bf, conditional go)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 benchmark pointer

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…onsult #13 route)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… frame span absorption, moving center as chart coordinate)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… routing step 0, dilation removed)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…on, inline 2-atom confluent lemma, Fock input conditions)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…, P4 chart inclusion, F3' witness z_c connection)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…n formula fixed)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… as corollary

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ble via section 8.9 annotation)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…onal-go, sync status surfaces

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…cale cap)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…s, M_r' typed, per-pair branches, domain contract)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ies)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ed coefficients, d0 provenance)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…exact QR globalization)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… A.5 aggregate, QRG schema test)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ecord COND9-PROBE diagnostic

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…er reviewer direction, strengthen consumer-type test)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…e one-hop conditioning)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…redesign (BORD-22/PTN-22 route)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant