---
name: Phase 2 + 2.5 — RDE 全分野拡張 + Top 30 review + 5 攻撃ターゲット選定
description: 7773 papers across 10 domains, 550 cross-domain bridges, Top 30 MEDIUM reviewed → 5 Rei attack targets selected (Tao Collatz / Generalized Cycles / Tree Chains / Yang-Mills / Erdős Discrepancy)
type: project
originSessionId: a08149e8-f1de-4fbc-b4eb-fb2cec8fad21
---
# Phase 2 + 2.5 (2026-04-16)

## Phase 2: 全分野拡張

**fetch**: `src/axiom-os/rei-discovery-engine/fetch_openalex_phase2.py`
- 10 domains × 5-6 keyword terms = ~50 queries
- **7,773 unique papers** (9.7 MB)

| Domain | Papers |
|---|---:|
| mathematics | 1198 |
| physics | 1073 |
| biology | 944 |
| philosophy | 882 |
| computer_science | 772 |
| medicine | 687 |
| economics | 679 |
| astronomy | 623 |
| psychology | 577 |
| chemistry | 338 |

Citation 分布: 0=397 / 1-10=1451 / 11-100=2322 / 100+=3603 (低・中・高 全包含)

## Phase 2 classify

**classify_phase2.py**: 同じ embedding pipeline + cross-domain bridge detection

| Tier | Count | % |
|---|---:|---:|
| NEAR | 190 | 2.4% |
| MEDIUM | 974 | 12.5% |
| **UNKNOWN** | **6609** | **85.0%** |

UNKNOWN 多数 = SEED_KERNEL がまだ math 中心のため、他分野の paper が arms-length。これは想定通り、Phase 3 で SEED_KERNEL 拡張すれば解消。

## Cross-domain bridges 発見 (550 件)

paper の domain ≠ nearest seed の domain 条件で抽出。**top 10**:

| cos | paper-dom | nearest seed | 概要 |
|---:|---|---|---|
| 0.868 | comput | M-PNP | "P Vs. NP Solved" (likely crank, 排除) |
| 0.789 | physic | M-YM | mass gap in 3D |
| 0.760 | physic | M-YM | mass gap non-abelian |
| 0.746 | physic | T-2050 | Distinct distances open problems |
| 0.720 | physic | M-YM | Planar Yang-Mills Hamiltonian |

★ **物理 → 数学 (Yang-Mills 系)** が最大 cluster → Paper 75/76 quantum anchor の自然拡張先

## Phase 2.5: Top 30 MEDIUM 手動 review

**排除**: P vs NP "完全解決" 系 6 件 (crank), 重複 3 件, 古すぎ 2 件 → 実質 19 件から 5 件選定。

### 選定 5 件 (Rei 正式攻撃ターゲット)

| 順 | Paper | 接続 | 攻撃 vector |
|---|---|---|---|
| 1 | **Tao 2022 "Almost all orbits..."** | NT-COLL, Paper 66/67/73 | 25 atomic cores の β-shift D1/D2 で Tao bound 検証 → Paper 79 |
| 2 | **Generalized Collatz Cycles (2020)** | NT-COLL + STEP 790 | 3-regime classification を generalized cycle に適用 |
| 3 | **Chains in Collatz Tree (1999)** | STEP 785 funnel + Paper 76 SELF | tree chain ↔ n=251 funnel structure 同型検証 |
| 4 | **Yang-Mills Mass Gap (2003)** | M-YM, Paper 75/76 | mass gap = INFINITY anchor (STEP 800) 拡張 → Paper 80 |
| 5 | **Erdős Discrepancy (2014)** | T-1750, Paper 78 p-adic | Tao 2015 解法を D-FUMT₈ で再記述 → Paper 81 |

### 推奨初期攻撃順

```
1st (確実): Tao 2022 Collatz       → Paper 79 候補, 1 session
2nd (短時間): Generalized Cycles    → STEP 790 拡張, 0.5 session
3rd (multi-anchor): Tree Chains    → Paper 76 SELF 具体化, 1 session
4th (re-interpretation): Erdős Disc → 教育的, 1 session
5th (high difficulty): Yang-Mills  → Millennium Prize, 1+ session
```

## 成果物

- `src/axiom-os/rei-discovery-engine/fetch_openalex_phase2.py`
- `src/axiom-os/rei-discovery-engine/classify_phase2.py`
- `data/discovery-engine/openalex-multi-domain-batch-2026-04-15.jsonl` (9.7 MB, 7773 papers)
- `data/discovery-engine/classified-phase2-2026-04-15.json` (2.5 MB)
- `data/discovery-engine/summary-phase2-2026-04-15.md` (per-domain table + top 30 + bridges)
- `data/discovery-engine/cross-domain-bridges-2026-04-15.md` (550 bridges by domain)
- `data/discovery-engine/phase2-5-manual-review.md` (5 attack target selection)

## 戦略的含意

1. Cross-domain bridges 550 件 = チャット版「異分野同型発見」が部分的に検証可能になった
2. Yang-Mills 系の物理 paper が math seed と最強接続 → Rei の Paper 75/76 quantum anchor が物理学側でも価値ある証拠
3. UNKNOWN 6609 件 = まだ Rei の手が届いていない領域、Phase 3 SEED_KERNEL 拡張 + LLM Distillation で減らせる
4. Phase 2.5 で **5 つの concrete next-paper-candidate** が抽出された → Rei の研究 backlog として明確化

## 次手 (本日 or 次セッション)

- **★1 Tao 2022 Collatz 攻撃** = Paper 79 候補 (推奨開始点)
- ご判断次第で順次実行