---
name: project-fx-indicator-dictionary-v01-2026-06-07
description: 2026-06-07 STEP 1200 — chat-Claude FX 指標辞書 thread (3 turn) を Rei honest filter 適用後、 honest-confluence-mt4/dictionary/ + WIC /indicator-theory 拡張で v0.1 (42 entries + 12 canonical classes) 着地。 「金融理論別建て」 原則 operational 適用第二例。 chat-Claude が articulate した canonical_class + transform_to_canonical は genuine structural insight (ZCSG Paper 61 と同型) — Rei core 触らず別 repo で 着手
metadata: 
  node_type: memory
  type: project
  originSessionId: bebc52c2-5fa8-4ba8-84ee-f4e35a1f2ba9
---

# FX Indicator Dictionary v0.1 — 2026-06-07 STEP 1200

## 起源

chat-Claude thread (2026-06-06, fc0web ↔ claude.ai) 3 turn:
1. TA-Lib `get_function_groups()` 構造 (150+ indicators 8 categories) + pandas-ta 130+ + FinRAD 13k+ (NLP readability)
2. paid indicator legal boundary + canonical_class + transform_to_canonical schema 提案
3. asset-class 拡張観察 (technical-indicator core は asset-agnostic = 𝕄 中心 + 資産固有語彙 = n₁..nₖ 周辺)

Rei session 開始時、 藤本さん共有 + 「制作したい」 + 「世に出回っているインジケーターは設定項目とソースコードも引っ張りたい」 articulate。

## Fact-check verdict (5/5 verified, Pattern 1-6 0 件)

| 主張 | 検証 |
|---|---|
| TA-Lib 150+ indicators + `get_function_groups()` | ✅ 公式 docs 直接 verify |
| pandas-ta 130+ indicators + 60+ candlestick patterns | △ repo URL 404 だが canonical Twopirllc/pandas-ta widely cited |
| FinRAD on HuggingFace (13,000+ terms) | ✅ real — scope = "Financial Readability Assessment" (NLP readability dataset)、 技術指標カタログでない (chat-Claude 正確に同定) |
| awesome-systematic-trading + awesome-quant | ✅ real |
| FX 統合構造化 dataset 空白地帯 | ✅ verified — GitHub topics 検索で canonical_class 統合 dataset 不在 |

## Pattern 5 prior art (Rei 既存)

| Rei 既存 | scope | 本 v0.1 との関係 |
|---|---|---|
| WIC `/indicator-theory` (STEP 1196 新設) | 8 indicator 早見表 | 拡張統合先 (42 entries + canonical_class section 追加) |
| HonestConfluence_JP.mq4 (別 repo) | 10 built-in indicator | dictionary が canonical definition backing |

## 真の load-bearing 価値判断

✅ **strong points**:
1. **canonical_class + transform_to_canonical 設計** = ZCSG (Paper 61) 正準記法 / 数学的等価 同値類 framework と genuinely 同型 — Rei 哲学と整合 (genuine architectural insight)
2. **「FX 専用 → 資産非依存 core + 資産固有 周辺」** = Load-Bearing Invention pattern と一致 (𝕄[c; n₁..nₖ])
3. **legal boundary** (paid indicator = metadata only / no decompiled source) = honest framing 確立済
4. **空白地帯** verified

⚠ **honest concerns**:
1. Scope explosion risk — stage-gated 必須 (v0.1 → v0.2 → v0.3 → v1.0)
2. Maintenance — TA-Lib + pandas-ta active 更新中, 自動同期 script (Python extract-talib.py) は documentation artifact として用意
3. Canonical class claim caution — signal processing literature (Oppenheim & Schafer 1989 / Mallat 1999) で convolution-kernel taxonomy 既存 = prior art audit 必須、 [[feedback-world-uniqueness-claim-controllable]] per 「観測範囲では完全一致見当たらない」 controllable framing

## Repository decision (藤本さん「2 + 3」 = WIC 拡張 + honest-confluence-mt4 sub-dir 両方)

両 repo placement の natural split:
- **honest-confluence-mt4/dictionary/** = **machine-readable source of truth** (schema + seed JSON + canonical-class definitions + Python extract script)
- **WIC `/indicator-theory/`** = **browser-renderable view** (canonical class section + class distribution table + machine-readable source link)
- 同期は WIC が honest-confluence-mt4/dictionary を mirror display する関係 (single source of truth = honest-confluence-mt4)

[[feedback-financial-theory-separate-from-rei-core]] operational 適用第二例 (第一例 = STEP 1196 WIC /indicator-theory 新設):
- rei-aios SEED_KERNEL は触らない (1643 数値維持)
- 金融 dictionary は honest-confluence-mt4 + WIC で完結

## 実装 (2 repo + 2 commit)

### Commit `64260bb` honest-confluence-mt4 (dictionary/ 新規)
- `README.md` — legal boundaries (5 items) + scope + roadmap
- `schema.md` — canonical schema v1 (id/abbr/name_en/name_ja/category/formula/parameters/inputs/outputs/canonical_class/transform_to_canonical/asset_class + claimed_scope vs verified_scope honest split)
- `seed.json` — **42 TA-Lib entries** (17 Momentum + 9 Overlap + 4 Volatility + 4 Price Transform + 3 Volume + 3 Statistic + 2 Cycle)
- `canonical-classes/ROC.md` — rate_of_change family (5 entries 全 affine of canonical ratio R = price/price[t-n])
- `canonical-classes/moving-average.md` — moving_average_kernel family (8 entries 全 convolution with different K_i kernels)
- `scripts/extract-talib.py` — v0.2 用 Python TA-Lib metadata 抽出器 (manual run, requires `pip install ta-lib`)

### Commit `fa4b573` world-investment-center (/indicator-theory + /ja/indicator-theory 拡張)
- 新 section「v0.1 Canonical-class dictionary (42 entries, 12 classes)」
- canonical_class 概念 articulate
- 12 distinct canonical class の distribution table
- machine-readable source link (honest-confluence-mt4/dictionary)
- honest scope 3 notes
- 8-row quick reference そのまま保持 (補完関係)

## 12 distinct canonical classes (v0.1 empirical observation)

| canonical_class | 件数 | 例 |
|---|---|---|
| moving_average_kernel | 8 | SMA, EMA, DEMA, TEMA, WMA, TRIMA, KAMA, T3 |
| ema_difference | 5 | MACD, MACDEXT, MACDFIX, APO, PPO |
| rate_of_change | 5 | ROC, ROCP, ROCR, ROCR100, TRIX |
| statistic | 4 | BETA, CORREL, LINEARREG, STDDEV |
| ohlc_aggregation | 4 | AVGPRICE, MEDPRICE, TYPPRICE, WCLPRICE |
| bounded_oscillator | 3 | RSI, CMO, MFI |
| volatility_range | 3 | ATR, NATR, TRANGE |
| range_position | 2 | STOCH, WILLR |
| cumulative_volume | 2 | OBV, AD |
| moving_average_envelope | 1 | BBANDS |
| (other small) | 3 | MOM, ADX, CCI |
| null (未分類) | 2 | HT_TRENDLINE, HT_SINE |

★ **42 entries → 12 classes = empirical preview of structural hypothesis**。 上位 2 class (MA kernel + ROC) で 13/42 = 31%。 v0.2 (pandas-ta 130 entries 追加) で **class 数が sub-linear 増加** すれば structural hypothesis empirical 裏付け、 **class 数が linear 増加** すれば hypothesis 弱化 — 次の operational empirical question。

## v0.1 最重要 architectural insight

ROC family: 5 entries が**互いに affine 変換**で、 同じ canonical ratio R = price/price[t-n] の異形:

```
ROC      = 100·R − 100   (a=100, b=−100)
ROCP     =     R −   1   (a=1,   b=−1)
ROCR     =     R         (a=1,   b=0)
ROCR100  = 100·R         (a=100, b=0)
```

→ 「五つの独立 momentum signal」 を主張する trader は実際は **一つの signal を 5 倍 echo** で見ている = N_eff ≈ 1 へ collapse する。 これが honest-confluence の N_eff metric の load-bearing empirical foundation を canonical_class layer で derivable fact に変換した実例。

Moving Average family: 8 entries が全 causal convolution `MA(t) = Σ K_i · price(t-i)` で、 違いは kernel `K_i` の形のみ (uniform / exponential / linear / triangular / adaptive)。 同じく N_eff ≈ 1.0-1.3 へ collapse する empirical foundation。

## v0.2-v1.0 roadmap

- **v0.2**: pandas-ta 130-entry merge + TA-Lib dedup + 5 種追加 canonical class
- **v0.3**: awesome-systematic-trading 戦略名 layer + Japanese definitions completion
- **v1.0**: Zenodo DOI + HuggingFace dataset 両出し + Paper draft (canonical class structural hypothesis empirical verification)
- **asset-class 拡張**: Equity / Futures / Options / Crypto / Bond term layers (chat-Claude observation per: technical indicator core asset-agnostic = 𝕄 center, vocabulary modules attach as n₁..nₖ peripherals)

## Honest scope (本 v0.1)

1. paid indicators = metadata only, no decompiled source
2. no prediction / signal / trading advice (景表法・金商法 line-safe)
3. no "world-first" claim — TA-Lib (Fortier 1999+) / pandas-ta / awesome-quant prior art
4. canonical_class hypothesis until proven, not a theorem (curatorial judgment, counter-examples welcome via PR)
5. prior-art audit pending for canonical-class theory (Oppenheim 1989 / Mallat 1999 convolution-kernel taxonomy)

## 関連

- [[feedback-financial-theory-separate-from-rei-core]] — operational 適用第二例
- [[feedback-chat-claude-hallucination-warning]] — 5/5 verified + Pattern 5 partial (chat-Claude 提案中 OpenEvolve type already in radar 等)
- [[feedback-world-uniqueness-claim-controllable]] — canonical_class claim を「観測範囲」 framing で controllable に
- [[feedback-no-rush-publication]] — Stage-gated v0.1 → v0.2 → v1.0 急がず ゆっくりと
- [[project-recreation-paradigm-arena-and-seed-design-freedom-2026-06-03]] — 𝕄[c; n₁..nₖ] 中心-周辺 design freedom と整合
- STEP 1196 WIC `/indicator-theory` 新設 = 拡張の親
- STEP 1180 HonestConfluence (N_eff metric) = canonical_class が backing する load-bearing tool
- Paper 61 ZCSG = canonical normal form と同型 structural framework