import analyzeIndicators from './analyze_indicators.js';
import { ok, fail, failFromError } from '../lib/result.js';
import { DetectPatternsInputSchema, DetectPatternsOutputSchema, PatternTypeEnum } from '../src/schemas.js';
import {
resolveParams,
} from './patterns/config.js';
import { detectSwingPoints, filterPeaks, filterValleys, type Candle } from './patterns/swing.js';
import {
linearRegressionWithR2,
near as nearFn,
pct as pctFn,
} from './patterns/regression.js';
import { type CandDebugEntry, type DetectContext } from './patterns/types.js';
import { buildStatistics } from './patterns/aftermath.js';
// --- 各パターン検出モジュール ---
import { detectDoubles } from './patterns/detect_doubles.js';
import { detectHeadAndShoulders } from './patterns/detect_hs.js';
import { detectTriangles } from './patterns/detect_triangles.js';
import { detectWedges } from './patterns/detect_wedges.js';
import { detectPennantsFlags } from './patterns/detect_pennants.js';
import { detectTriples } from './patterns/detect_triples.js';
/**
* detect_patterns - チャートパターン検出（完成済み＋形成中）
*
* 設計思想:
* - 目的: チャートパターンを検出し、統計的に信頼性の高いデータを提供
* - 特徴: swingDepth パラメータによる厳密なスイング検出でパターン品質を重視
* - ブレイク検出: ATR * 0.5 バッファ、最初の明確なブレイクで終点を確定
* - 用途: 「過去の成功率は？」「典型的な期間は？」「aftermath は？」
*
* オプション:
* - includeCompleted: true (デフォルト) → 完成済みパターンを検出
* - includeForming: true → 形成中パターンも検出（早期警告向け）
*/
type DetectIn = typeof DetectPatternsInputSchema extends { _type: infer T } ? T : any;
export default async function detectPatterns(
pair: string = 'btc_jpy',
type: string = '1day',
limit: number = 90,
opts: Partial<{
swingDepth: number;
tolerancePct: number;
minBarsBetweenSwings: number;
strictPivots: boolean;
patterns: Array<typeof PatternTypeEnum._type>;
requireCurrentInPattern: boolean;
currentRelevanceDays: number;
// 統合オプション
includeForming: boolean;
includeCompleted: boolean;
includeInvalid: boolean;
view: 'summary' | 'detailed' | 'full' | 'debug';
}> = {}
) {
try {
// --- パラメータ解決（patterns/config.ts から） ---
const { swingDepth, tolerancePct, minBarsBetweenSwings: minDist, autoScaled } = resolveParams(type, opts);
const strictPivots = (opts as any)?.strictPivots !== false; // 既定: 厳格
// 統合オプション
const includeForming = opts.includeForming ?? false;
const includeCompleted = opts.includeCompleted ?? true;
const includeInvalid = opts.includeInvalid ?? false;
const want = new Set(opts.patterns || []);
// 'triangle' が指定された場合は3種を含む互換挙動
if (want.has('triangle')) {
want.add('triangle_ascending' as any);
want.add('triangle_descending' as any);
want.add('triangle_symmetrical' as any);
}
const res = await analyzeIndicators(pair, type as any, limit);
if (!res?.ok) return DetectPatternsOutputSchema.parse(fail(res.summary || 'failed', 'internal')) as any;
const candles = res.data.chart.candles as Array<{ open: number; close: number; high: number; low: number; isoTime?: string }>;
if (!Array.isArray(candles) || candles.length < 20) {
return DetectPatternsOutputSchema.parse(ok('insufficient data', { patterns: [] }, { pair, type, count: 0 })) as any;
}
// 1) Swing points（patterns/swing.ts から）
const pivots = detectSwingPoints(candles as Candle[], { swingDepth, strictPivots });
// debug buffers
const debugSwings = pivots.map(p => ({ idx: p.idx, price: p.price, kind: p.kind, isoTime: (candles[p.idx] as any)?.isoTime }));
const debugCandidates: CandDebugEntry[] = [];
// --- 共有コンテキスト構築 ---
const ctx: DetectContext = {
candles,
pivots,
allPeaks: filterPeaks(pivots),
allValleys: filterValleys(pivots),
tolerancePct,
minDist,
want,
includeForming,
debugCandidates,
type,
swingDepth,
near: (a: number, b: number) => nearFn(a, b, tolerancePct),
pct: pctFn,
lrWithR2: linearRegressionWithR2,
};
// --- 各パターン検出を実行 ---
let patterns: any[] = [];
// 2) Double top/bottom
const doubles = detectDoubles(ctx);
patterns.push(...doubles.patterns);
// 3) Head & Shoulders
const hs = detectHeadAndShoulders(ctx);
patterns.push(...hs.patterns);
// 4) Triangles
const triangles = detectTriangles(ctx);
patterns.push(...triangles.patterns);
// 4b-4d) Wedges
const wedges = detectWedges(ctx);
patterns.push(...wedges.patterns);
// 5) Pennant & Flag + global dedup
const pennants = detectPennantsFlags(ctx);
patterns.push(...pennants.patterns);
// 6) Triple Top / Triple Bottom
const triples = detectTriples(ctx);
patterns.push(...triples.patterns);
// Optional filter: only patterns whose end is within N days from now (current relevance)
{
const requireCurrent = !!opts.requireCurrentInPattern;
const defaultDaysByType = (tf: string): number => {
if (tf === '1month') return 60; // ~2 months
if (tf === '1week') return 21; // ~3 weeks
return 7; // default for daily and intraday
};
const maxAgeDays = Number.isFinite(opts.currentRelevanceDays as any)
? Number(opts.currentRelevanceDays)
: defaultDaysByType(String(type));
if (requireCurrent && patterns.length) {
const nowMs = Date.now();
const inDays = (iso?: string) => {
if (!iso) return Infinity;
const t = Date.parse(iso);
if (!Number.isFinite(t)) return Infinity;
return Math.abs(nowMs - t) / 86400000;
};
patterns = patterns.filter((p: any) => inDays(p?.range?.end) <= maxAgeDays);
}
}
// Aftermath analysis + statistics（patterns/aftermath.ts へ抽出済み）
const { statistics } = buildStatistics(patterns, candles);
// includeForming / includeCompleted に基づくフィルタリング
let filteredPatterns = patterns;
if (!includeForming || !includeCompleted) {
filteredPatterns = patterns.filter((p: any) => {
const isForming = p.status === 'forming' || p.status === 'near_completion';
const isCompleted = p.status === 'completed' || p.status === 'invalid' || !p.status;
if (includeForming && isForming) return true;
if (includeCompleted && isCompleted) return true;
return false;
});
}
// includeInvalid に基づくフィルタリング
if (!includeInvalid) {
filteredPatterns = filteredPatterns.filter((p: any) => p.status !== 'invalid');
}
patterns = filteredPatterns;
// overlays: パターン範囲をそのまま帯描画できるように提供
const ranges = patterns.map((p: any) => ({ start: p.range.start, end: p.range.end, label: p.type }));
const warnings: any[] = [];
if (patterns.length <= 1) {
warnings.push({ type: 'low_detection_count', message: '検出数が少ないです。tolerancePct や minBarsBetweenSwings の調整を推奨します', suggestedParams: { tolerancePct: 0.03, minBarsBetweenSwings: 2 } });
}
// --- サイズ抑制: debug 配列を上限でトリム（view未指定で返却が肥大化しやすいため） ---
// ただし accepted を優先的に残す（accepted → rejected の順で cap まで）
const cap = 200;
const swingsTrimmed = Array.isArray(debugSwings) ? debugSwings.slice(0, cap) : [];
let candidatesTrimmed: any[] = [];
if (Array.isArray(debugCandidates)) {
const acc = debugCandidates.filter((c: any) => !!c?.accepted);
const rej = debugCandidates.filter((c: any) => !c?.accepted);
candidatesTrimmed = [...acc, ...rej].slice(0, cap);
}
const debugTrimmed = {
swings: swingsTrimmed,
candidates: candidatesTrimmed,
};
// summary 生成: LLM が content から読み取れるように詳細を含める
const patternSummaries = patterns.map((p: any, idx: number) => {
const startDate = p.range?.start?.substring(0, 10) || '?';
const endDate = p.range?.end?.substring(0, 10) || '?';
let detail = `${idx + 1}. ${p.type} (パターン整合度: ${p.confidence})\n - 期間: ${startDate} ~ ${endDate}`;
// ウェッジパターンの場合、ブレイク情報を追加
if ((p.type === 'falling_wedge' || p.type === 'rising_wedge') && p.breakoutDirection && p.outcome) {
const directionJa = p.breakoutDirection === 'up' ? '上方' : '下方';
const outcomeJa = p.outcome === 'success' ? '成功' : '失敗';
const expectedDir = p.type === 'falling_wedge' ? '上方' : '下方';
const meaning = p.type === 'falling_wedge'
? (p.outcome === 'success' ? '強気転換' : '弱気継続')
: (p.outcome === 'success' ? '弱気転換' : '強気継続');
detail += `\n - ブレイク方向: ${directionJa}ブレイク（本来は${expectedDir}ブレイクが期待されるパターン）`;
detail += `\n - パターン結果: ${outcomeJa}（${meaning}）`;
}
// ネックラインがある場合
if (p.neckline && Array.isArray(p.neckline) && p.neckline.length >= 2) {
detail += `\n - ネックライン: ${Math.round(p.neckline[0]?.y || 0).toLocaleString()}円 → ${Math.round(p.neckline[1]?.y || 0).toLocaleString()}円`;
}
return detail;
}).join('\n\n');
// aftermath 統計をテキストに含める（LLM が structuredContent.data を読めない対策）
const statsText = statistics && Object.keys(statistics).length > 0
? '\n\n【統計情報】\n' + Object.entries(statistics).map(([k, v]) => `${k}: ${JSON.stringify(v)}`).join('\n')
: '';
const summaryText = `${pair.toUpperCase()} [${type}] ${limit}本から${patterns.length}件を検出（${patterns.map((p: any) => p.type).join('×1、')}×1）\n\n【検出パターン（全件）】\n${patternSummaries || 'なし'}${statsText}\n\nチャート連携: data.overlays を render_chart_svg.overlays に渡すと注釈/範囲を描画できます。\n\nパターン整合度について（形状一致度・対称性・期間から算出）:\n 0.8以上 = 理想的な形状（教科書的パターン）\n 0.7-0.8 = 標準的な形状（他指標と併用推奨）\n 0.6-0.7 = やや不明瞭（慎重に判断）\n 0.6未満 = 形状不十分`
+ `\n\n---\n📌 含まれるもの: チャートパターン検出（種類・整合度・期間）、ブレイク情報、統計`
+ `\n📌 含まれないもの: 出来高によるパターン確認、テクニカル指標値、板情報`
+ `\n📌 補完ツール: analyze_indicators（指標でパターンを裏付け）, get_flow_metrics（出来高確認）, get_orderbook（板情報）`;
const out = ok(
summaryText,
{ patterns, overlays: { ranges }, warnings, statistics },
{
pair,
type,
count: patterns.length,
effective_params: { swingDepth, minBarsBetweenSwings: minDist, tolerancePct, autoScaled },
visualization_hints: { preferred_style: 'line', highlight_patterns: patterns.map((p: any) => p.type).slice(0, 3) },
debug: debugTrimmed
}
);
return DetectPatternsOutputSchema.parse(out) as any;
} catch (e: unknown) {
return failFromError(e, { schema: DetectPatternsOutputSchema }) as any;
}
}
