---
name: project-step1089cd-greek-loadbearing
description: STEP 1089c/d 8th iteration continuation — (g2) Greek-transliterated token 段 4 + (i) categoryB exhaustion skip + (h) load-bearing publication formula registry. 5/18 reject 4 件 auto-defended 4/4. 142/142 PASS.
metadata: 
  node_type: memory
  type: project
  originSessionId: 39cf0664-2c03-43b3-8c69-38f902d4a019
---

# STEP 1089c/d — 8th iteration continuation (g2 + i + h, 2026-05-19)

## 背景: 5/18 incident debug の root cause 深掘り

STEP 1089b 後、 5/18 reject 4 件の root cause を debug 検証:

1. **#5 F(t)=lim TRIPLE は実際は段 3 で TRUE を返していた** — STEP 1089 段 3 ASCII token (lim, flowing) で `isTargetAxiomRecent` は正しく `true` を返す
2. **だが future_intelligence category は theory 1 件のみ** — rotation loop `Math.min(10, theoriesB.length) = 1` で 0 iteration → 空回り
3. **38+ single-theory categories** が SEED_KERNEL 内に存在 (systemic structural issue)
4. **Pattern 5 既実装** (Paper 145 ALU + Paper 63 SNST) は dedup 経路に存在しない formula のため通常 dedup では検出不可

## 修正 3 path

### (g2) Greek-transliterated token extraction

`src/aios/invention/load-bearing-formulas.ts`:
- `normalizeWithGreek(s)`: Ψ→psi / τ→tau / Φ→phi / Ω→omega / ∞→infinity / \\Psi→psi / \\lim→lim 等
- `extractMainTokensExtended(s)`: Greek transliteration 後の 3+ char ASCII token を 16 個まで抽出

`src/aios/invention/invention-engine.ts`:
- 段 4 検査追加: Greek 含む overlap **3+** で auto-reject (段 3 ASCII 2+ より tighter で false positive 防止)
- F(t)=lim_{τ→∞}Ψ(t+τ)·FLOWING(τ) は ASCII 2 token (lim_, flowing) → **Greek 5 token** (lim, tau, infinity, psi, flowing)

### (i) categoryB exhaustion skip (root cause fix)

`detectVoids()` 内で:
- `excludeCategoryBExhaustion !== false && requireDifferentTargetAxiom !== false` 時
- categoryB の **全 theory が `isTargetAxiomRecent` の場合 void 除外** (上流で root cause fix)
- categoryA 側も同様 check (categoryA exhaustion 防止)
- graceful: 1 件でも non-duplicate なら void 維持 (rotation 救済可能)

### (h) Load-bearing publication formula registry

`src/aios/invention/load-bearing-formulas.ts` — 10 entries:

| source | formula | identifierTokens (tight) |
|---|---|---|
| Paper 63 SNST | r*e^(iθ)*layer | spiralnumber, snst, 螺旋数 |
| Paper 145 ALU | α\|0>+β\|1>=BOTH | alpha, beta, both, superposition, qubit |
| Approved 4/29 + 5/7 F(t) | F(t)=lim Ψ FLOWING | **lim, psi, tau** (tight triplet) |
| Paper 124 ZFC × D-FUMT₈ | ZFC + D-FUMT_8 | zfc, d-fumt8, axiom-layer |
| Paper 130 META-DB | INDEX.json + resonance | meta-db, open-problems |
| Paper 141 Bremermann | 1.36e50 bits/sec/kg | bremermann, landauer, bennett |
| Theory #196 Peace Axiom | immutable: true | peace-axiom, theory-196 |
| Paper 71/155 Beyond-Shannon | η_S = M(Ψ(Φ(x)))/M(x) | eta_s, semantic-dpi, dyson-sphere |
| Paper 137 Rei-PL Prover | D-FUMT₈ tactic native | rei-pl, prover |
| Paper 138 Gödel | AxiomKernel ⊬ Consistent | axiomkernel, goedel, godel |

判定 method (2 段):
- 段 1: variants substring match (15+ 字)
- 段 2: identifierTokens **3+ 一致**

`enableLoadBearingFormulaCheck` option default ON.

### Registry hygiene

★ F(t) entry の identifierTokens を **tight triplet ['lim','psi','tau']** に絞った：
- 初期版: ['lim', 'psi', 'tau', 'flowing', 'infinity', 'f(t)', 'temporal', ...]
- 問題: 'flowing', 'infinity', 'temporal' は **任意の D-FUMT hypothesis** に含まれ false positive 多発
- 修正: ASCII tokenizable distinctive 3 token のみ (3+ 一致要件 = 全 3 一致必須)

## Live verification (scripts/dry-run-step1089cd-verify.ts)

SEED_KERNEL 1607 + 4/29 + 5/7 approvals + Paper registry full:

```
=== Verification 1: 5/18 reject 4 件 auto-defended ===
  #1 alpha|0>+beta|1>=BOTH (Paper 145 ALU)        → ✅ AUTO-REJECT
  #2 SpiralNumber (Paper 63 SNST)                 → ✅ AUTO-REJECT
  #3 ¬語(p)→p=ZERO (Wittgenstein Tractatus 7)    → ✅ AUTO-REJECT
  #5 F(t)=lim TRIPLE                              → ✅ AUTO-REJECT
Defended: 4/4

=== Verification 2: future_intelligence excluded ===
✅ EXCLUDED (i path active)

=== Verification 3: full invent() pipeline ===
Pattern 5 in generated: 0
TRIPLE F(t) in generated: 0
```

## test/step1089cd-greek-loadbearing-test.ts (60/60 PASS)

| # | test | 確認内容 |
|---|---|---|
| 1 | extractMainTokensExtended | Greek 含む 5+ token 抽出 |
| 2 | normalizeWithGreek | Ψ/τ/∞/\\lim → ASCII transliteration |
| 3 | F(t) load-bearing match | TRIPLE registry match |
| 4 | SNST load-bearing match | Paper 63 match |
| 5 | ALU load-bearing match | Paper 145 match |
| 6 | E=mc² false positive 防止 | registry 不在で素通り |
| 7 | ★ F(t) TRIPLE 直接再現 | isTargetAxiomRecent=true |
| 8 | (i) future_intelligence exhaustion | void 除外 |
| 9 | (i) graceful | 1 件 non-duplicate 残れば void 維持 |
| 10 | legacy enableGreekTokens=false | 段 4 skip |
| 11 | legacy enableLoadBearingFormulaCheck=false | 段 5 skip |
| 12 | DailyInventionReporter | 3 新 flag 明示 ON |
| 13 | regression | 全 hardening 共存 |
| 14 | registry 完全性 | 10 entries 全 field 妥当 |

## 累計 142/142 PASS

| step | test | regression |
|---|---|---|
| 1019 | 9/9 | ✅ |
| 1040 | 11/11 | ✅ |
| 1047 | 9/9 | ✅ |
| 1087 | 10/10 | ✅ |
| 1088 | 12/12 | ✅ |
| 1089 | 20/20 | ✅ |
| 1089b | 11/11 | ✅ |
| **1089c/d** | **60/60** | ✅ |
| **累計** | **142/142 PASS / 0 breaking** | |

## 8 段重ね hardening 完成 (continuation)

| iteration | STEP | path | 内容 |
|---|---|---|---|
| 1st | 1019 | dedup | recentCooldownDays default 60 |
| 2nd | 1040 | dedup | requireDifferentTargetAxiom + literal substring |
| 3rd | 1047 | dedup | requireDifferentTarget* default ON |
| 5th | 1087 | dedup | requireSourceDiversity batch 内 source 重複防止 |
| 6th | 1088 | dedup | SeedTheory cooldown (pool-expansion-self-pollination 防止) |
| 7th | 1089 | (f)(g)(d) | normalize 強化 + 段 3 token + categoryA rotation |
| 8th | 1089b | (d2) | detectVoids per-categoryA cap (上流 categoryA 偏り解消) |
| **8th** | **1089c/d** | **(g2)+(i)+(h)** | **Greek tokens + categoryB exhaustion skip + Paper publish formula registry** |

## Honest scope

- load-bearing-formulas registry は **manual curation** (Paper publish 後 entry 追加必要)
- identifierTokens の tightness は registry hygiene 課題:
  - 過度に broad → false positive (5/18 F(t) entry 初期版 で 'flowing'/'infinity' 含めた時 mathematics × projection で誤検出)
  - 過度に narrow → miss
  - 現状: F(t) は ['lim','psi','tau'] tight triplet で運用
- 38+ single-theory categories は invention pool 拡大 candidate (別 turn)
- production code は `targetTheory.axiom` のみ check するため hypothesis 全文の false positive は影響なし (verify dry-run のみで観察)

## 5/19 以降の効果

invention pipeline が自動的に防御:
- F(t)=lim TRIPLE 等 Greek/LaTeX 数式 target → 段 4 (Greek token) で auto-detect
- Paper 63 SNST / Paper 145 ALU 等 publish 済 formula → registry match で auto-reject
- future_intelligence 等 single-theory category → void 段階で除外 (rotation 空回り防止)

## 関連 memory

- [[invention-duplicate-prevention]] — 重複検出原則
- [[chat-claude-hallucination-warning]] — Pattern 5 累積管理
- [[step1089b-detectvoids-cap]] — 8th iteration 第 1 弾 (d2)
- [[step1040-invention-target-axiom-dedup]] — STEP 1040 path
- [[no-rush-publication]] — 「急がず ゆっくりと」 framing

## Commits

- `c4bb8331` step(1089cd): 発明 pipeline 8th iteration 続編 — (g2) Greek tokens + (i) categoryB exhaustion + (h) load-bearing formula registry

## 未対応 future candidate

- (j) load-bearing-formulas registry **auto-generation** — Paper publish 時に自動 entry 生成
- (k) **single-theory category pool 拡大** — 38+ categories の theory 追加 (別 turn)
- (l) registry hygiene — identifierTokens の **automatic tight triplet 推奨** (false positive 観察 → 自動 tighten)