321 lines
12 KiB
Python
321 lines
12 KiB
Python
"""Review center read models.
|
||
|
||
This module keeps the new review/iteration semantics explicit:
|
||
- opportunity review describes whether the system found useful opportunities;
|
||
- paper trading review is the only place where PnL is treated as execution PnL;
|
||
- evidence attribution describes whether onchain/sentiment/LLM evidence helped.
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import json
|
||
from datetime import datetime, timedelta
|
||
|
||
from app.db.paper_trading import get_paper_trading_summary
|
||
from app.db.schema import get_conn
|
||
|
||
|
||
def _safe_int(value, default=0):
|
||
try:
|
||
return int(value or 0)
|
||
except Exception:
|
||
return default
|
||
|
||
|
||
def _safe_float(value, default=0.0):
|
||
try:
|
||
return float(value or 0)
|
||
except Exception:
|
||
return default
|
||
|
||
|
||
def _loads(value, default=None):
|
||
try:
|
||
if isinstance(value, (dict, list)):
|
||
return value
|
||
if isinstance(value, str) and value.strip():
|
||
return json.loads(value)
|
||
except Exception:
|
||
pass
|
||
return default if default is not None else {}
|
||
|
||
|
||
def _since(days):
|
||
return (datetime.now() - timedelta(days=max(1, min(_safe_int(days, 30), 365)))).isoformat()
|
||
|
||
|
||
def _bucket_count(rows, key, fallback="unknown"):
|
||
counts = {}
|
||
for row in rows:
|
||
value = (row.get(key) or fallback) if isinstance(row, dict) else fallback
|
||
counts[value] = counts.get(value, 0) + 1
|
||
return [{"name": k, "count": v} for k, v in sorted(counts.items(), key=lambda x: (-x[1], x[0]))]
|
||
|
||
|
||
def _dedupe_by_symbol(rows, limit=0):
|
||
items = []
|
||
seen = set()
|
||
for row in rows:
|
||
symbol = str(row.get("symbol") or "").strip().upper()
|
||
key = symbol or f"row:{row.get('id')}"
|
||
if key in seen:
|
||
continue
|
||
seen.add(key)
|
||
items.append(row)
|
||
if limit and len(items) >= int(limit):
|
||
break
|
||
return items
|
||
|
||
|
||
def _opportunity_review(conn, since):
|
||
rec_rows = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT id, symbol, rec_time, status, display_bucket, execution_status, action_status,
|
||
entry_triggered, strategy_version, signal_codes_json, signal_labels_json,
|
||
market_context_json, derivatives_context_json, sector_context_json
|
||
FROM recommendation
|
||
WHERE rec_time >= %s
|
||
ORDER BY rec_time DESC, id DESC
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
review_rows = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT *
|
||
FROM review_log
|
||
WHERE review_time >= %s
|
||
ORDER BY review_time DESC, id DESC
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
missed_rows_raw = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT *
|
||
FROM missed_explosions
|
||
WHERE detect_time >= %s
|
||
ORDER BY gain_pct DESC, detect_time DESC
|
||
LIMIT 200
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
missed_rows = _dedupe_by_symbol(missed_rows_raw, limit=20)
|
||
|
||
total = len(rec_rows)
|
||
executed_ids = {
|
||
int(r["recommendation_id"])
|
||
for r in conn.execute("SELECT recommendation_id FROM paper_trades WHERE opened_at >= %s", (since,)).fetchall()
|
||
if r.get("recommendation_id")
|
||
}
|
||
buy_now = [r for r in rec_rows if r.get("execution_status") == "buy_now"]
|
||
wait_pullback = [r for r in rec_rows if r.get("execution_status") == "wait_pullback"]
|
||
observe = [r for r in rec_rows if r.get("execution_status") == "observe" or r.get("display_bucket") == "watch_pool"]
|
||
invalid = [r for r in rec_rows if r.get("execution_status") == "invalid" or r.get("status") in ("expired", "invalid", "archived")]
|
||
executed = [r for r in rec_rows if int(r.get("id") or 0) in executed_ids]
|
||
|
||
outcomes = _bucket_count(review_rows, "outcome", "未复盘")
|
||
reviewed_effective = [r for r in review_rows if r.get("outcome") in ("爆发", "失败", "横盘")]
|
||
hit_count = sum(1 for r in reviewed_effective if r.get("outcome") == "爆发")
|
||
|
||
return {
|
||
"definition": "机会复盘只评价机会发现、确认和漏选，不代表交易收益。",
|
||
"summary": {
|
||
"total_opportunities": total,
|
||
"buy_now_count": len(buy_now),
|
||
"wait_pullback_count": len(wait_pullback),
|
||
"observe_count": len(observe),
|
||
"invalid_count": len(invalid),
|
||
"paper_executed_count": len(executed),
|
||
"reviewed_count": len(review_rows),
|
||
"effective_review_count": len(reviewed_effective),
|
||
"opportunity_hit_rate": round(hit_count / len(reviewed_effective) * 100, 2) if reviewed_effective else 0,
|
||
"missed_explosion_count": len(missed_rows),
|
||
},
|
||
"status_distribution": _bucket_count(rec_rows, "execution_status", "unknown"),
|
||
"outcome_distribution": outcomes,
|
||
"missed_explosions": missed_rows[:10],
|
||
"recent_reviews": review_rows[:12],
|
||
}
|
||
|
||
|
||
def _paper_review(conn, since, days):
|
||
summary = get_paper_trading_summary(days=days)
|
||
trades = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT *
|
||
FROM paper_trades
|
||
WHERE opened_at >= %s
|
||
ORDER BY opened_at DESC, id DESC
|
||
LIMIT 20
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
events = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT *
|
||
FROM paper_trade_events
|
||
WHERE event_time >= %s
|
||
ORDER BY event_time DESC, id DESC
|
||
LIMIT 30
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
exit_reasons = _bucket_count([t for t in trades if t.get("status") == "closed"], "exit_reason", "unknown")
|
||
event_types = _bucket_count(events, "event_type", "unknown")
|
||
return {
|
||
"definition": "模拟交易复盘是唯一收益口径，基于 paper_trades 的开仓、平仓、移动止盈事件。",
|
||
"summary": summary,
|
||
"exit_reasons": exit_reasons,
|
||
"event_types": event_types,
|
||
"recent_trades": trades,
|
||
"recent_events": events,
|
||
}
|
||
|
||
|
||
def _evidence_review(conn, since):
|
||
news_rows = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT source, symbol, importance, event_type, decision, tech_score, processed, detected_at, title
|
||
FROM event_news
|
||
WHERE detected_at >= %s
|
||
ORDER BY detected_at DESC, id DESC
|
||
LIMIT 80
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
onchain_rows = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT source, chain, symbol, signal_code, signal_label, direction, value_usd,
|
||
confidence, severity, detected_at
|
||
FROM onchain_events
|
||
WHERE detected_at >= %s
|
||
ORDER BY detected_at DESC, id DESC
|
||
LIMIT 80
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
raw_onchain_rows = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT source, chain, event_type, symbol_guess, mapped_symbol, mapping_status,
|
||
importance, detected_at, title
|
||
FROM onchain_raw_events
|
||
WHERE detected_at >= %s
|
||
ORDER BY importance DESC, detected_at DESC, id DESC
|
||
LIMIT 80
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
llm_rows = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT target_type, insight_type, status, model, prompt_version, updated_at
|
||
FROM llm_insights
|
||
WHERE updated_at >= %s
|
||
ORDER BY updated_at DESC, id DESC
|
||
LIMIT 80
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
|
||
mapped_raw = [r for r in raw_onchain_rows if r.get("mapping_status") == "mapped" or r.get("mapped_symbol")]
|
||
high_onchain = [r for r in onchain_rows if _safe_int(r.get("confidence")) >= 70 or str(r.get("severity") or "").upper() in ("A", "S")]
|
||
actionable_news = [r for r in news_rows if r.get("decision") in ("recommend", "observe", "risk") or str(r.get("importance") or "").upper() in ("A", "S")]
|
||
llm_success = [r for r in llm_rows if r.get("status") == "success"]
|
||
|
||
return {
|
||
"definition": "多源归因只判断证据贡献：舆情、链上、LLM 是否帮助发现/解释机会，不直接生成交易收益。",
|
||
"summary": {
|
||
"news_count": len(news_rows),
|
||
"actionable_news_count": len(actionable_news),
|
||
"onchain_signal_count": len(onchain_rows),
|
||
"high_confidence_onchain_count": len(high_onchain),
|
||
"raw_onchain_count": len(raw_onchain_rows),
|
||
"mapped_raw_onchain_count": len(mapped_raw),
|
||
"llm_runs": len(llm_rows),
|
||
"llm_success_count": len(llm_success),
|
||
},
|
||
"news_sources": _bucket_count(news_rows, "source", "unknown"),
|
||
"news_decisions": _bucket_count(news_rows, "decision", "unprocessed"),
|
||
"onchain_sources": _bucket_count(onchain_rows, "source", "unknown"),
|
||
"onchain_signals": _bucket_count(onchain_rows, "signal_code", "unknown")[:12],
|
||
"raw_mapping": _bucket_count(raw_onchain_rows, "mapping_status", "unknown"),
|
||
"llm_status": _bucket_count(llm_rows, "status", "unknown"),
|
||
"recent_news": news_rows[:8],
|
||
"recent_onchain": onchain_rows[:8],
|
||
"recent_llm": llm_rows[:8],
|
||
}
|
||
|
||
|
||
def _iteration_review(conn, since):
|
||
logs = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT *
|
||
FROM strategy_iteration_log
|
||
WHERE created_at >= %s
|
||
ORDER BY created_at DESC, id DESC
|
||
LIMIT 12
|
||
""",
|
||
(since,),
|
||
).fetchall()]
|
||
candidates = [dict(r) for r in conn.execute(
|
||
"""
|
||
SELECT *
|
||
FROM strategy_rule_candidate
|
||
ORDER BY created_at DESC, id DESC
|
||
LIMIT 30
|
||
"""
|
||
).fetchall()]
|
||
for item in logs:
|
||
for field, fallback in (
|
||
("metrics_json", {}),
|
||
("findings_json", []),
|
||
("problems_json", []),
|
||
("actions_json", []),
|
||
("candidate_rules_json", []),
|
||
):
|
||
item[field.replace("_json", "")] = _loads(item.get(field), fallback)
|
||
return {
|
||
"definition": "策略迭代只产生候选假设和发布闸门结论，不直接等于收益提升。",
|
||
"summary": {
|
||
"iteration_count": len(logs),
|
||
"candidate_count": len(candidates),
|
||
"gray_count": sum(1 for c in candidates if c.get("status") == "gray"),
|
||
"active_count": sum(1 for c in candidates if c.get("status") == "active"),
|
||
"latest_release_decision": (logs[0].get("release_decision") if logs else "") or "hold",
|
||
"latest_release_reason": (logs[0].get("release_reason") if logs else "") or "",
|
||
},
|
||
"release_decisions": _bucket_count(logs, "release_decision", "unknown"),
|
||
"candidate_status": _bucket_count(candidates, "status", "candidate"),
|
||
"recent_logs": logs,
|
||
"recent_candidates": candidates[:12],
|
||
}
|
||
|
||
|
||
def get_review_center_dashboard(days=30):
|
||
days = max(1, min(_safe_int(days, 30), 365))
|
||
since = _since(days)
|
||
conn = get_conn()
|
||
try:
|
||
opportunity = _opportunity_review(conn, since)
|
||
paper = _paper_review(conn, since, days)
|
||
evidence = _evidence_review(conn, since)
|
||
iteration = _iteration_review(conn, since)
|
||
finally:
|
||
conn.close()
|
||
|
||
return {
|
||
"days": days,
|
||
"generated_at": datetime.now().isoformat(timespec="seconds"),
|
||
"principles": [
|
||
"机会归档不计算交易收益，只记录发现、确认、失效和漏选。",
|
||
"真实收益口径只来自模拟交易或未来真实交易账本。",
|
||
"链上、舆情、LLM 属于证据层，只做发现和解释，不直接改变推荐状态。",
|
||
"策略迭代只发布经过样本约束和灰度闸门验证的规则。",
|
||
],
|
||
"opportunity": opportunity,
|
||
"paper_trading": paper,
|
||
"evidence": evidence,
|
||
"iteration": iteration,
|
||
}
|
||
|
||
|
||
__all__ = ["get_review_center_dashboard"]
|