完成筹码断档与吸筹次数首轮实验

This commit is contained in:
baobao 2026-10-30 09:46:14 +08:00
parent 5dd0783640
commit 5081080b7a
40 changed files with 5743 additions and 0 deletions

View File

@ -2559,3 +2559,16 @@ QMT11:53:36至11:54:04短暂中断,自动重连恢复,外部原因未核实
- 本地及155候选镜像全量隔离回归通过,页面三道护栏通过。实际部署HTML与只读数据在Chrome重放,1440/390宽度、HTTP复制后粘贴50行、HTTPS剪贴板、空名单/故障禁用、失败反馈、刷新和排除残余持仓均通过;无控制台错误或业务写请求,不冒充真实登录验收。
- 17:09:03 QMT连续确认1485267、待处理0、待归属0,无新增拒绝或重连;资金年龄28.2秒、冻结0、四方2700股及微盘空仓,数量和现金对账均无差异。17:09:54确认继续推进1485281。微盘规则、模式及全局暂停保持原值。
- 记录见docs/选股取消主题限额与微盘名单复制_2026-10-29.md及docs/evidence/pms_theme_copy_20261029.json,当前状态已更新阶段接续。原有114行DEVLOG未提交内容与其他研究改动保留,CLAUDE.md未修改。
## 2026-10-30 筹码断档与吸筹次数首轮实验
按用户认可的方案,将现价上方稀疏与双峰断档分别研究,协议与参数在收益计算前冻结。155独立只读研究容器导出阶段快照、筹码分档和压力表,所有SQL单表,未继承业务挂载及交易凭证。行情截止10月29日,09:43核验本任务容器均已回收。无部署、参数调整或交易动作,未接触188真实PMS。
共14595个预定观察,10231个成熟,只有31个信号日和3个60日时间块。费用每边0.3%,成熟配对中上方稀疏21组胜率33.33%对19.05%,但平均净收益仍为负6.33%;上方双峰6组两边50%;下跌中至少2次独立吸筹14组两边57.14%。多次吸筹原始组54笔胜率46.30%,高于零次组38.57%,优势未在匹配后保留。15%拉升路径、组合条件与费用敏感性另列,未据小样本进行正式显著性判定。
套牢盘压力表只有26个数据日,较长历史依赖回填筹码与252日阶段回放。2286条分档核实ocr_above为严格高于现价的原始占比,ocr_115为成本达到现价115%及以上的远端占比。吸筹次数与固定价格门槛下套牢变化的描述性相关约负0.016,未证实清楚线性关系。粗分档、历史证券及行业覆盖、交易日历表为空等限制写入报告。
开发机10项配置隔离、断网专项测试通过;500条观察独立复算,97组配对逐项核对,98项源摘要一致;原始缓存重建筹码特征后摘要不变,四幅研究图检查通过。协议、代码、结果、审计和文件摘要位于research/chip_accumulation_20261030,当前状态见docs/筹码断档与吸筹研究接续_2026-10-30.md。研究原始数据和中间缓存留在开发机,不入Git。
尚未完成冻结后新样本的前瞻60日验证,也未完成真实交易链路验收;没有创建新的定时任务。保留原有114行DEVLOG未提交修改、复盘台账及其他未跟踪研究材料,仅提交本轮研究文件与本条记录。

View File

@ -10,6 +10,7 @@
| 全市场热度、涨跌幅与持有期标定 | [热度策略研究接续](全市场热度四维标定接续_2026-10-28.md),含回填限制、参数结果和复现入口;[热度反复活跃矩阵接续](热度反复活跃矩阵研究接续_2026-10-29.md),含前20日次数、持有期、数据可用性与阶段指数对照 |
| 综合分与聪明资金分阈值及市场阶段比较 | [评分策略研究接续](综合分与聪明资金分标定接续_2026-10-28.md),含810组结果、历史版本限制和复现入口;[反复活跃矩阵接续](反复活跃矩阵研究接续_2026-10-29.md),含20日次数、1至10日持有期与指数隔离 |
| 宏观价量网格、同步形成与日线趋势 | [宏观网格研究接续](宏观网格研究接续_2026-10-29.md),含三因素增量、八幅研究图、各风格与数据末日区间判断 |
| 筹码断档、套牢盘与吸筹次数 | [筹码吸筹研究接续](筹码断档与吸筹研究接续_2026-10-30.md),含两类断档、60日胜率与拉升路径、字段对拍和小样本限制 |
| PMS通用迁移与既有修复 | [开发接续](开发接续_2026-10-12.md),按主题查对应节;旧时点未完成项可能已有后续结论 |
| 架构与接口 | [README模块地图](../README.md)、[总体设计](../POSITION_MGMT_DESIGN.md);改接口再读[QMT协议](../QMT_WS_PROTOCOL.md)、[bionic接口](../BIONIC_PMS_INTERFACE.md)或[计划接口](../UPSTREAM_PLAN_API.md) |
| 后端、前端、测试、研究 | [后端规则](../app/AGENTS.md)、[页面规则](../app/web/AGENTS.md)、[脚本规则](../scripts/AGENTS.md)、[研究规则](../research/AGENTS.md),仅选择所涉及目录 |

View File

@ -0,0 +1,23 @@
# 筹码断档、套牢盘与吸筹次数研究接续
核验时间:2026年10月30日09:43,北京时间。行情与特征源截至10月29日。用户已授权按讨论方案开展研究,两类断档分别统计。
本轮完成首轮历史探索,尚未证实稳定的买入胜率优势。完整解释、原始分组、配对结果、拉升路径、联合条件及四幅研究图见[首轮报告](../research/chip_accumulation_20261030/筹码断档与吸筹次数首轮实验.md)。实验协议在计算收益前冻结,未据收益修改阈值。
| 主比较 | 完成配对 | 60日净胜率:处理组 / 对照组 | 当前判断 |
| --- | --- | --- | --- |
| 现价上方稀疏 | 21组 | 33.33% / 19.05% | 有局部正向差异,但处理组平均净收益仍为负6.33%;无法推广至全部稀疏股票 |
| 现价上方双峰断档 | 6组 | 50.00% / 50.00% | 样本不足,未看到胜率优势 |
| 下跌中至少2次独立吸筹 | 14组 | 57.14% / 57.14% | 原始分组的优势未在匹配后保留 |
费用为每边0.3%。下跌多次吸筹的原始组有54笔完成交易,胜率46.30%,零次组38.57%;这不是可直接归因于次数的增量。匹配后的60日内15%拉升比例也未显示优势。吸筹次数与固定价格门槛下未来20日套牢变化的描述性相关为负0.016,不能解释成因果。
数据覆盖是主要限制。gp_stock_trap_pressure仅26个数据日,accumulation_daily仅33日;较长历史来自266批阶段快照,其中252批回放、14批实际运行。2026年完整筹码表有197个数据日,但部分股票缺失、价格档过粗。共14595个预先抽取的观察,10231个成熟,4364个未满60日;成熟信号只有31日、3个时间块,不能据此宣称统计显著或独立样本外有效。连续每天的明确吸筹状态不当作多个独立事件。
字段核对发现ocr_above等于严格高于压力表收盘价的原始筹码百分比之和,ocr_115等于大于或等于收盘价115%的原始百分比之和。它表示远端套牢盘,不是上方15%以内的筹码。2286条分档按这个口径复算,误差仅为浮点误差。研究的归一化上方比例与原表tsp评分分别保留,没有回写业务表。
开发机通过10项配置隔离、禁止联网的专项测试;独立复算500条真实观察,核对97组配对门槛、98项源摘要,并用缓存重建特征验证摘要不变。trade_cal为空,以三个指数及最近四日全市场行情交叉核对日期,独立交易所日历验证仍未完成。测试回执、文件摘要和复现命令见报告及[交付清单](../research/chip_accumulation_20261030/manifest.json)。raw和cache保留在开发机并被Git忽略,复现也依赖既有宏观研究行情缓存,摘要均已登记。
155只用于独立只读容器导出,业务SQL单表,未继承交易凭证和业务挂载;09:43检查本任务容器已全部回收。未部署、未改参数、未生成交易动作,也未接触188真实PMS。不能把本地研究测试称为真实交易链路验收。
下一阶段仍欠冻结后新入样的前瞻60日观察、更完整的历史证券与行业可用时间、交易所日历及更细筹码档位。下跌、断档、多次吸筹的联合条件只有2笔或0笔成熟交易,暂不能检验叠加优势。保持本版本归档;新增数据或阈值应另建版本并区分历史探索和前瞻验证。本轮没有创建定时监控或修改生产调度。

View File

@ -0,0 +1,3 @@
raw/
cache/
__pycache__/

View File

@ -0,0 +1,75 @@
"""纯离线收益、配对与不确定性计算。"""
import numpy as np
def outcome(a,i,j,hold=60):
t=len(a['dates']);entry=i+1;due=entry+hold
r={'entry_index':entry,'due_index':due,'calendar_matured':bool(due<t),
'entry_filled':False,'completed':False,'entry_date':None,'exit_date':None,
'exit_delay':None,'gross':np.nan,'net30':np.nan,'net60':np.nan,
'path_complete':False,'hit15':np.nan,'up_before_down':np.nan,
'max_close_gain':np.nan,'max_close_loss':np.nan}
if entry>=t:return r
r['entry_date']=str(a['dates'][entry])
if not a['can_buy'][entry,j]:return r
r['entry_filled']=True;price=a['open'][entry,j]
if due>=t:return r
path=a['close'][entry:due,j]/price-1
r['path_complete']=bool(np.isfinite(path).all())
if r['path_complete']:
hi=np.flatnonzero(path>=.15);lo=np.flatnonzero(path<=-.10)
r.update(hit15=float(len(hi)>0),up_before_down=float(len(hi)>0 and (not len(lo) or hi[0]<lo[0])),
max_close_gain=float(path.max()),max_close_loss=float(min(0,path.min())))
exits=np.flatnonzero(a['can_sell'][due:,j])
if not len(exits):return r
end=due+int(exits[0]);gross=a['open'][end,j]/price
r.update(completed=True,exit_date=str(a['dates'][end]),exit_delay=end-due,gross=float(gross),
net30=float(gross*.997/1.003-1),net60=float(gross*.994/1.006-1))
return r
def match_pairs(frame,treatment,extra_pressure=False,day_count_caliper=None):
"""先只用信号日特征匹配。不会读取收益、成交或未来日期字段。"""
rows=[];unmatched=0;treated=0
fields=['ret60','log_cap','log_vol','log_turn','position60']
cal=np.array([.05,np.log(2),np.log(2),np.log(2),.20])
if extra_pressure:fields+=['overhead'];cal=np.r_[cal,.10]
if day_count_caliper is not None:fields+=['clear_days60'];cal=np.r_[cal,day_count_caliper]
f=frame.copy();f['treatment']=treatment
f=f[np.isfinite(f[fields]).all(axis=1)&f.industry.notna()&(f.industry!='')]
for (day,industry),g in f.groupby(['date','industry'],sort=True):
active=g[g.treatment].sort_values('code');control=g[~g.treatment].sort_values('code')
used=set()
for idx,row in active.iterrows():
treated+=1
c=control.loc[~control.index.isin(used)]
if c.empty:unmatched+=1;continue
diff=np.abs(c[fields].to_numpy(float)-row[fields].to_numpy(float))
valid=(diff<=cal+1e-12).all(axis=1)
if not valid.any():unmatched+=1;continue
distance=np.sum((diff/cal)**2,axis=1);distance[~valid]=np.inf
target=c.index[int(np.argmin(distance))];used.add(target)
rows.append({'treat_id':int(idx),'control_id':int(target),'date':day,'industry':industry,
'distance':float(distance.min()),'treat_code':row['code'],'control_code':c.loc[target,'code']})
return rows,{'treated_with_matching_fields':treated,'unmatched':unmatched,'matched':len(rows),'calipers':dict(zip(fields,cal.tolist()))}
def cluster_intervals(values,time_indices,treat_codes,control_codes,reps=1000,seed=20261030):
"""固定匹配后,分别整块重抽时间与联合重抽两端股票。区间仅描述性使用。"""
values=np.asarray(values,float);time_indices=np.asarray(time_indices,int)
ok=np.isfinite(values);values=values[ok];time_indices=time_indices[ok]
tc=np.asarray(treat_codes)[ok];cc=np.asarray(control_codes)[ok]
if not len(values):return {'n':0,'time_blocks':0,'time_interval':None,'stock_interval':None}
blocks=np.unique(time_indices//60);bi=np.searchsorted(blocks,time_indices//60)
stocks=np.unique(np.r_[tc,cc]);ti=np.searchsorted(stocks,tc);ci=np.searchsorted(stocks,cc)
rng=np.random.default_rng(seed);tb=[];sb=[]
for _ in range(reps):
if len(blocks)>1:
count=rng.multinomial(len(blocks),np.ones(len(blocks))/len(blocks));w=count[bi]
if w.sum():tb.append(float(np.sum(w*values)/w.sum()))
count=rng.multinomial(len(stocks),np.ones(len(stocks))/len(stocks));w=count[ti]*count[ci]
if w.sum():sb.append(float(np.sum(w*values)/w.sum()))
degenerate=bool(np.ptp(values)==0)
def interval(x):return np.quantile(x,[.025,.975]).tolist() if len(x)>50 and not degenerate else None
return {'n':len(values),'mean':float(values.mean()),'time_blocks':len(blocks),'unique_stocks':len(stocks),
'time_interval':interval(tb),'stock_interval':interval(sb),'formal_p_value':None,
'degenerate_observed_differences':degenerate,
'interval_note':'各配对观察差完全相同,经验重抽退化,不能给出有效不确定性区间。' if degenerate else '少量簇的重抽区间只作描述。',
'interpretation':'固定配对的描述性区间;少于8个时间块不进行正式显著性判断。'}

View File

@ -0,0 +1,151 @@
"""冻结特征后计算60日结果、预设匹配、分组及机制关系。"""
import datetime as dt
import numpy as np
import pandas as pd
from common import ROOT,read,dump,digest,protocol,install_offline
from analysis_core import outcome,match_pairs,cluster_intervals
def stats(g):
mature=g[g.calendar_matured];filled=mature[mature.entry_filled];done=filled[filled.completed]
wins=int((done.net30>0).sum());n=len(filled);unknown=n-len(done)
r={'signals':len(g),'matured_signals':len(mature),'filled_matured':n,'completed':len(done),'unresolved_exit':unknown,
'stocks':g.code.nunique(),'signal_dates':g.date.nunique(),'first_signal':g.date.min() if len(g) else None,
'last_signal':g.date.max() if len(g) else None,'wins':wins,
'win_rate':wins/len(done) if len(done) else None,
'win_rate_lower_missing_exit':wins/n if n else None,'win_rate_upper_missing_exit':(wins+unknown)/n if n else None,
'mean_net30':float(done.net30.mean()) if len(done) else None,'median_net30':float(done.net30.median()) if len(done) else None,
'mean_net60':float(done.net60.mean()) if len(done) else None,'median_net60':float(done.net60.median()) if len(done) else None,
'win_rate_cost60':float((done.net60>0).mean()) if len(done) else None,
'delayed_exits':int((done.exit_delay>0).sum()),'path_complete':int(filled.path_complete.sum())}
path=filled[filled.path_complete]
for k in ['hit15','up_before_down','max_close_gain','max_close_loss']:
r[k]=float(path[k].mean()) if len(path) else None
return r
def partial_corr(frame,x,y,controls):
cols=[x,y]+controls;f=frame[cols].replace([np.inf,-np.inf],np.nan).dropna()
if len(f)<30 or f[x].nunique()<2 or f[y].nunique()<2:return {'n':len(f),'partial_correlation':None}
mat=np.column_stack([np.ones(len(f)),f[controls].to_numpy(float)])
xv=f[x].to_numpy(float);yv=f[y].to_numpy(float)
xr=xv-mat@np.linalg.lstsq(mat,xv,rcond=None)[0];yr=yv-mat@np.linalg.lstsq(mat,yv,rcond=None)[0]
return {'n':len(f),'partial_correlation':float(np.corrcoef(xr,yr)[0,1]),'controls':controls,'interpretation':'描述性线性残差相关,非因果效应,未作显著性检验'}
def run_pairs(frame,key,subset,treat,pressure=False,day_caliper=None):
f=frame[subset].copy();pairs,counts=match_pairs(f,treat.loc[f.index],pressure,day_caliper)
out=[]
for p in pairs:
t=frame.loc[p['treat_id']];c=frame.loc[p['control_id']]
p.update(test=key,time_index=int(t['i']),treat_matured=bool(t.calendar_matured),control_matured=bool(c.calendar_matured),
treat_filled=bool(t.entry_filled),control_filled=bool(c.entry_filled),
treat_completed=bool(t.completed),control_completed=bool(c.completed),
treat_net30=float(t.net30),control_net30=float(c.net30),treat_net60=float(t.net60),control_net60=float(c.net60),
treat_days60=int(t.clear_days60),control_days60=int(c.clear_days60),
treat_ret60=float(t.ret60),control_ret60=float(c.ret60),
treat_path_complete=bool(t.path_complete),control_path_complete=bool(c.path_complete),
treat_hit15=float(t.hit15),control_hit15=float(c.hit15),
treat_up_first=float(t.up_before_down),control_up_first=float(c.up_before_down))
out.append(p)
q=pd.DataFrame(out)
result={'matching':counts,'all_pairs':len(q),'mature_pairs':0,'completed_pairs':0,'win_lift':None,'mean_return_lift':None}
if q.empty:return result,q
mature=q[q.treat_matured&q.control_matured]
complete=mature[mature.treat_completed&mature.control_completed]
result.update(mature_pairs=len(mature),completed_pairs=len(complete),
treat_entry_failures=int((~mature.treat_filled).sum()),control_entry_failures=int((~mature.control_filled).sum()),
unmatched_or_unfinished_pairs=len(mature)-len(complete))
path=mature[mature.treat_path_complete&mature.control_path_complete]
result['path_pairs']=len(path)
if len(path):
result['path_comparison']={
'treat_hit15':float(path.treat_hit15.mean()),'control_hit15':float(path.control_hit15.mean()),
'hit15_lift':float((path.treat_hit15-path.control_hit15).mean()),
'treat_up_first':float(path.treat_up_first.mean()),'control_up_first':float(path.control_up_first.mean()),
'up_first_lift':float((path.treat_up_first-path.control_up_first).mean())}
if len(complete):
delta=(complete.treat_net30>0).astype(float)-(complete.control_net30>0).astype(float)
result.update(treat_win=float((complete.treat_net30>0).mean()),control_win=float((complete.control_net30>0).mean()),
win_lift=float(delta.mean()),mean_return_lift=float((complete.treat_net30-complete.control_net30).mean()),
mean_treat_return=float(complete.treat_net30.mean()),mean_control_return=float(complete.control_net30.mean()),
median_treat_return=float(complete.treat_net30.median()),median_control_return=float(complete.control_net30.median()),
win_lift_cost60=float(((complete.treat_net60>0).astype(float)-(complete.control_net60>0).astype(float)).mean()),
mean_return_lift_cost60=float((complete.treat_net60-complete.control_net60).mean()),
mean_clear_days_difference=float((complete.treat_days60-complete.control_days60).mean()))
result['uncertainty']=cluster_intervals(delta,complete.time_index,complete.treat_code,complete.control_code)
for cfield in ['ret60','log_cap','log_vol','log_turn','position60']+(['overhead'] if pressure else []):
tt=frame.loc[complete.treat_id,cfield].to_numpy();cc=frame.loc[complete.control_id,cfield].to_numpy()
sd=float(f[cfield].std())
result.setdefault('matched_balance',{})[cfield]={'mean_difference':float((tt-cc).mean()),'standardized_difference':float((tt-cc).mean()/sd) if sd>0 else None}
lo=[];hi=[]
for _,pair in mature.iterrows():
# 无成交按现金零收益视为未盈利,仅用于信号层最坏/最好范围。
ts=(int(pair.treat_net30>0),)*2 if pair.treat_completed else ((0,0) if not pair.treat_filled else (0,1))
cs=(int(pair.control_net30>0),)*2 if pair.control_completed else ((0,0) if not pair.control_filled else (0,1))
lo.append(ts[0]-cs[1]);hi.append(ts[1]-cs[0])
result['signal_level_lift_bounds_unfilled_cash_zero']=[float(np.mean(lo)),float(np.mean(hi))]
period=[]
for month,g in complete.groupby(complete.date.str[:7]):
period.append({'month':month,'pairs':len(g),'win_lift':float(((g.treat_net30>0).astype(float)-(g.control_net30>0).astype(float)).mean()),
'return_lift':float((g.treat_net30-g.control_net30).mean())})
result['monthly']=period
return result,q
def main():
install_offline();p=protocol()
assert digest(ROOT/'implementation_notes.json')==(ROOT/'implementation_notes.sha256').read_text().split()[0]
f=pd.read_csv(ROOT/'cache/features_before_outcomes.csv.gz');a=dict(np.load(ROOT/'cache/prices.npz'))
# 固定特征摘要后才生成收益标签,便于证明分组未被收益倒推。
frozen={'frozen_at':dt.datetime.now(dt.timezone(dt.timedelta(hours=8))).isoformat(),
'features_sha256':digest(ROOT/'cache/features_before_outcomes.csv.gz'),'protocol_sha256':digest(ROOT/'protocol.json'),
'implementation_sha256':digest(ROOT/'implementation_notes.json'),'sample_sha256':digest(ROOT/'cache/samples_before_outcomes.csv.gz')}
if (ROOT/'feature_freeze.json').exists():
old=read(ROOT/'feature_freeze.json')
assert all(old[k]==v for k,v in frozen.items() if k!='frozen_at'),'已看收益之后不得改动冻结特征'
else:dump(ROOT/'feature_freeze.json',frozen)
results=[outcome(a,int(r.i),int(r.j)) for r in f.itertuples()]
f=pd.concat([f,pd.DataFrame(results)],axis=1)
f['log_cap']=np.log(f.market_cap);f['log_vol']=np.log(f.volatility60);f['log_turn']=np.log(f.turnover20)
f['frequency_group']=np.where(f.episodes60>=2,'2+',np.where(f.episodes60==1,'1','0'))
f['risk_intact']=~f.invalidated&~f.idling
f.to_csv(ROOT/'cache/observations.csv.gz',index=False,compression={'method':'gzip','mtime':0})
summaries=[]
def add(name,subset,groups):
for label,condition in groups:
g=f[subset&condition];summaries.append({'analysis':name,'group':label,**stats(g)})
allrows=pd.Series(True,index=f.index);chip=f.chip_valid;valley=chip&f.valley_evaluable;high=f.episodes60>=2
add('all',allrows,[('all',allrows)])
add('upper_gap',chip,[('no',f.upper_gap==False),('yes',f.upper_gap==True)])
add('valley_above',valley,[('no',f.valley_above==False),('yes',f.valley_above==True)])
add('valley_crossing',valley,[('no',f.valley_crossing==False),('yes',f.valley_crossing==True)])
add('valley_below',valley,[('no',f.valley_below==False),('yes',f.valley_below==True)])
for name,subset in [('frequency_all',allrows),('frequency_down',f.downtrend),('frequency_down_intact',f.downtrend&f.risk_intact)]:
add(name,subset,[(g,f.frequency_group==g) for g in ['0','1','2+']])
for kind,subset in [('upper_gap',chip),('valley_above',valley)]:
add(kind+'_joint_down',subset&f.downtrend,[(f'gap={g},high_freq={h}',(f[kind]==g)&(high==h)) for g in [False,True] for h in [False,True]])
add('overlap',valley,[(f'upper={u},valley={v}',(f.upper_gap==u)&(f.valley_above==v)) for u in [False,True] for v in [False,True]])
summary=pd.DataFrame(summaries);summary.to_csv(ROOT/'group_results.csv',index=False)
tests={};pairframes=[]
configurations=[('upper_gap',chip,f.upper_gap==True,True,None),('valley_above',valley,f.valley_above==True,True,None),
('frequency_down',f.downtrend,high,False,None),('frequency_down_same_days',f.downtrend,high,False,5),
('frequency_down_intact',f.downtrend&f.risk_intact,high,False,None),
('frequency20_down',f.downtrend,f.episodes20>=2,False,None)]
for key,mask,treat,pressure,day_caliper in configurations:
result,q=run_pairs(f,key,mask,treat,pressure,day_caliper);tests[key]=result
if not q.empty:pairframes.append(q)
print(key,'配对',result['completed_pairs'],'胜率差',result['win_lift'],flush=True)
(pd.concat(pairframes,ignore_index=True) if pairframes else pd.DataFrame(columns=['test','date','control_code','treat_id','control_id'])).to_csv(ROOT/'matched_pairs.csv.gz',index=False,compression={'method':'gzip','mtime':0})
# 控制同期价格变化,机制检验使用此前窗口吸筹次数,不引用未来收益。
mechanism=partial_corr(f,'prior20_episodes','pressure_change20_fixed_price',['return_last20','log_cap','log_vol','log_turn'])
relation={}
for target in ['overhead','overhead_near15','upper_mass']:
relation[target]=partial_corr(f[f.chip_valid],'episodes60',target,['ret60','position60','log_cap','log_vol','log_turn'])
coverage={'all_samples':len(f),'mature_samples':int(f.calendar_matured.sum()),'mature_stocks':int(f[f.calendar_matured].code.nunique()),
'pending_60_days':int((~f.calendar_matured).sum()),'latest_mature_signal':str(f.loc[f.calendar_matured,'date'].max()),
'mature_sample_dates':int(f[f.calendar_matured].date.nunique()),'mature_time_blocks':int((f.loc[f.calendar_matured,'i']//60).nunique()),
'chip_mature_samples':int((f.calendar_matured&f.chip_valid).sum()),'downtrend_mature_samples':int((f.calendar_matured&f.downtrend).sum()),
'historical_only':True,'formal_inference_performed':False,'reason':'已有历史包含回填与回放,且独立60日时间块不足8个。'}
dump(ROOT/'results.json',{'observed_at':dt.datetime.now(dt.timezone(dt.timedelta(hours=8))).isoformat(),
'coverage':coverage,'tests':tests,'mechanism':mechanism,'feature_relationships':relation,
'verdict':'仅提供历史探索和反例;现有历史不能确认稳定的60日胜率提升。'})
print(coverage,flush=True)
if __name__=='__main__':main()

View File

@ -0,0 +1,125 @@
{
"checks": {
"source_digests_verified": 98,
"exchange_calendar_matches": null,
"index_calendars_match": true,
"recent_dates_marketwide": true,
"independent_samples_recomputed": 500,
"duplicate_stock_dates": 0,
"minimum_sample_spacing": 60,
"features_unchanged_after_outcomes": true,
"group_totals_reconcile": true,
"matching_calipers_verified": 97
},
"source_quality": {
"trap": {
"rows": 135532,
"days": 26,
"stocks": 5226,
"start": "2026-09-10",
"end": "2026-10-29",
"missing_market_days": [
"2026-09-11",
"2026-09-15",
"2026-09-16",
"2026-09-17"
],
"null_rates": {
"ocr_above": 0.023824631821267303,
"ocr_115": 0.023824631821267303,
"tsp_dist": 0.023824631821267303,
"chip_peak_price": 0.023824631821267303,
"chip_peak_dev": 0.023824631821267303,
"chip_near_pct": 0.023824631821267303,
"rank_dist": 0.023824631821267303
},
"data_flags": {
"both": 132303,
"cyq_only": 3229
},
"loss_rate_identity_max_abs_error": 1.249000902703301e-16,
"ranges": {
"winner_rate": [
0.0,
100.38
],
"loss_rate": [
-0.0038,
1.0
],
"ocr_above": [
0.0,
100.14
],
"ocr_115": [
0.0,
99.85
],
"tsp": [
0.000192,
1.0
],
"tsp_pct": [
0.0192,
100.0
]
}
},
"daily": {
"rows": 184323,
"days": 33,
"stocks": 5597,
"start": "2026-09-04",
"end": "2026-10-29",
"missing_market_days": [
"2026-10-09"
],
"null_rates": {
"score": 0.0056476945362217415,
"score_basis": 1.0,
"raw_score": 1.0,
"state": 0.0056476945362217415,
"bias": 0.0056476945362217415,
"structure": 0.0056476945362217415,
"pos_pct": 0.0056476945362217415,
"pos_tag": 0.0056476945362217415,
"sub_low": 0.0056476945362217415,
"sub_vol": 0.0056476945362217415,
"sub_obv": 0.0056476945362217415,
"sub_close": 0.0056476945362217415,
"fund_adj": 0.0056476945362217415,
"mf_last_date": 0.06540149628641027,
"stock_name": 0.06540149628641027,
"is_st": 0.06540149628641027,
"is_delisting": 0.06540149628641027,
"dist_to_low_pct": 0.7148158395859442,
"idling": 0.12086934348941804,
"broke_below": 0.12086934348941804,
"acc_high_px": 0.7148158395859442,
"acc_low_px": 0.7148158395859442
},
"versions": {
"fund_flow_v2": 134152,
"fund_flow_v1": 50171
},
"quality": {
"OK": 182117,
"SHORT_HISTORY": 786,
"SUSPENDED": 745,
"STALE": 379,
"INSUFFICIENT_HISTORY": 296
}
},
"chip_dates": {
"start": "2025-11-10",
"end": "2026-10-29",
"days": 206,
"missing_2026_market_days": []
}
},
"limitations": [
"trade_cal为空;前三指数日期轴一致,最后4日由至少5000只股票的日线日期交叉核验,未完成独立交易所日历核验。"
],
"status": "passed_with_disclosed_limitations",
"machine": "开发机,unit_test_guard隔离配置并禁止联网"
}

View File

@ -0,0 +1,100 @@
"""在断网入口独立复算计数、日期、费用及分组,输出核验回执。"""
import collections
import numpy as np
import pandas as pd
from common import ROOT,OLD,read,dump,digest,protocol,install_offline
def main():
install_offline();p=protocol();a=dict(np.load(ROOT/'cache/prices.npz'));b=dict(np.load(ROOT/'cache/phase.npz'))
f=pd.read_csv(ROOT/'cache/observations.csv.gz');checks={}
source_count=0
for name,wanted in read(ROOT/'price_coverage.json')['source_sha256'].items():
assert digest(ROOT.parent/name)==wanted;source_count+=1
for metadata in ['phase_coverage.json','chip_coverage.json']:
for name,wanted in read(ROOT/metadata)['source_sha256'].items():
assert digest(ROOT/'raw'/name)==wanted;source_count+=1
checks['source_digests_verified']=source_count
calendar=read(ROOT/'raw/calendar.json.gz')['data'];assert calendar['ok']
days=sorted(str(r['cal_date'])[:10] for r in calendar['rows'] if str(r['is_open'])=='1' and str(r['cal_date'])[:10]<=p['data_cutoff'])
limitations=[]
if days:
assert list(a['dates'])==days;checks['exchange_calendar_matches']=True
else:
index=read(OLD/'raw/export.json')['data']['index_daily']
for code in ['000001.SH','399006.SZ','000300.SH']:
observed=sorted(r['timestamp'][:10] for r in index[code]['rows'] if p['price_source_start']<=r['timestamp'][:10]<=p['data_cutoff'])
assert list(a['dates'][a['dates']<=observed[-1]])==observed
recent=read(ROOT/'raw/recent_prices.json.gz')['data']['bars']['rows']
marketdays=collections.Counter(r['trade_date'][:10] for r in recent)
assert all(n>=5000 for n in marketdays.values())
assert list(a['dates'][a['dates']>'2026-10-23'])==sorted(marketdays)
days=list(a['dates'])
checks.update(exchange_calendar_matches=None,index_calendars_match=True,recent_dates_marketwide=True)
limitations.append('trade_cal为空;前三指数日期轴一致,最后4日由至少5000只股票的日线日期交叉核验,未完成独立交易所日历核验。')
rng=np.random.default_rng(31);ids=rng.choice(len(f),min(500,len(f)),replace=False)
for rid in ids:
r=f.iloc[rid];i,j=int(r.i),int(r.j)
for window,key in [(60,'episodes60'),(20,'episodes20')]:
v=b['events'][i-window:i,j];assert np.isfinite(v).all();assert int(v.sum())==r[key]
assert b['clear'][i-60:i,j].sum()==r.clear_days60
if r.completed:
e=int(r.entry_index);x=int(np.searchsorted(a['dates'],r.exit_date))
assert x>=e+60 and a['can_sell'][x,j]
assert not a['can_sell'][e+60:x,j].any()
net=(a['open'][x,j]*(1-.003)-a['open'][e,j]*(1+.003))/(a['open'][e,j]*(1+.003))
assert abs(net-r.net30)<1e-12
if r.path_complete:
path=a['close'][int(r.entry_index):int(r.due_index),j]/a['open'][int(r.entry_index),j]-1
assert len(path)==60 and np.isfinite(path).all();assert bool((path>=.15).any())==bool(r.hit15)
checks['independent_samples_recomputed']=len(ids)
assert not f.duplicated(['date','code']).any()
for code,g in f.groupby('code'):assert np.all(np.diff(np.sort(g.i))>=60)
checks['duplicate_stock_dates']=0;checks['minimum_sample_spacing']=60
freeze=read(ROOT/'feature_freeze.json')
assert freeze['features_sha256']==digest(ROOT/'cache/features_before_outcomes.csv.gz')
checks['features_unchanged_after_outcomes']=True
groups=pd.read_csv(ROOT/'group_results.csv')
for name,mask in [('frequency_all',pd.Series(True,index=f.index)),('frequency_down',f.downtrend),('frequency_down_intact',f.downtrend&f.risk_intact)]:
q=groups[groups.analysis==name]
assert q.signals.sum()==mask.sum()
assert q.completed.sum()==(mask&f.completed&f.calendar_matured).sum()
checks['group_totals_reconcile']=True
pairs=pd.read_csv(ROOT/'matched_pairs.csv.gz')
for key,g in pairs.groupby('test'):
assert not g.duplicated(['date','control_code']).any()
for r in g.itertuples():
t=f.loc[r.treat_id];c=f.loc[r.control_id]
assert t.date==c.date and t.industry==c.industry
assert abs(t.ret60-c.ret60)<=.05+1e-12
assert abs(t.position60-c.position60)<=.20+1e-12
for k in ['log_cap','log_vol','log_turn']:assert abs(t[k]-c[k])<=np.log(2)+1e-12
if key in ['upper_gap','valley_above']:assert abs(t.overhead-c.overhead)<=.10+1e-12
if key=='frequency_down_same_days':assert abs(t.clear_days60-c.clear_days60)<=5
checks['matching_calipers_verified']=len(pairs)
quality={}
for mode,datecol in [('trap','trade_date'),('daily','data_date')]:
d=pd.DataFrame(read(ROOT/f'raw/{mode}.json.gz')['data']['rows'])
d['day']=pd.to_datetime(d[datecol].astype(str)).dt.strftime('%Y-%m-%d')
field='symbol' if mode=='trap' else 'stock_code'
assert not d.duplicated(['day',field]).any()
within=[x for x in days if d.day.min()<=x<=d.day.max()]
q={'rows':len(d),'days':d.day.nunique(),'stocks':d[field].nunique(),'start':d.day.min(),'end':d.day.max(),
'missing_market_days':sorted(set(within)-set(d.day)),
'null_rates':{k:float(v) for k,v in d.isna().mean().items() if v>0}}
if mode=='trap':
q['data_flags']={str(k):int(v) for k,v in d.data_flag.value_counts(dropna=False).items()}
q['loss_rate_identity_max_abs_error']=float((d.loss_rate-(1-d.winner_rate/100)).abs().max())
q['ranges']={k:[float(d[k].min()),float(d[k].max())] for k in ['winner_rate','loss_rate','ocr_above','ocr_115','tsp','tsp_pct']}
else:
q['versions']={str(k):int(v) for k,v in d.algo_version.value_counts().items()}
q['quality']={str(k):int(v) for k,v in d.quality.value_counts().items()}
quality[mode]=q
dc=read(ROOT/'raw/chip_dates.json.gz')['data'];chip_dates=[]
for tab,r in dc.items():
assert r['ok'];chip_dates.extend(x['trade_date'] for x in r['rows'])
quality['chip_dates']={'start':min(chip_dates),'end':max(chip_dates),'days':len(chip_dates),
'missing_2026_market_days':sorted(set(x for x in days if '2026-01-05'<=x<=p['data_cutoff'])-set(chip_dates))}
dump(ROOT/'audit.json',{'checks':checks,'source_quality':quality,'limitations':limitations,'status':'passed_with_disclosed_limitations','machine':'开发机,unit_test_guard隔离配置并禁止联网'})
print('审计通过',checks,flush=True)
if __name__=='__main__':main()

View File

@ -0,0 +1,96 @@
{
"source_sha256": {
"chips_20260109.json.gz": "bdd585d23b7932ea9923fb4eaf55ca044b7692adf664c45d9ddd7502371131c8",
"chips_20260116.json.gz": "0148cb0e0afc478ebc506c0e9504b9b241a079b8ab665657e36358771393d7a3",
"chips_20260122.json.gz": "b81c26d9b02c47454ae999145e7e2baebfa7adb834d3ae8a6e94b3ed1f8da1c9",
"chips_20260123.json.gz": "fc030ce8f983ead18fc6b1ac0bbadff98dbf9bed37160a516fcfc2038f4312bd",
"chips_20260129.json.gz": "f9e42a975e189260fe57bfae784df97963fb2cd270364323da8b28fe18bc22c8",
"chips_20260130.json.gz": "2147bbfc5f8c20b961d24acd3484649b4570f4840485857ac5183faae469d9de",
"chips_20260205.json.gz": "aadfb6cce8e0e46126f808cd7476cdcb45b5c5f4745cbd8adf4612aea493b3d3",
"chips_20260206.json.gz": "76951fb0c608ff8dcf4e51af6ba940000762425ccc0c67abc95a24ac7adcf01d",
"chips_20260212.json.gz": "97553aa79a25837512f8628c3168229cceb8f938dabd1b6a972c78f167ad101c",
"chips_20260213.json.gz": "71e0df0a6b5259c05d8058ccdac34646013f4324503a4aa0d5c2f7e72f62a134",
"chips_20260227.json.gz": "5ecd01a1e904d5ebe6bde68feb3f10ceb05cf638a2ddc91b8fd8eb22c36dd711",
"chips_20260306.json.gz": "d2ea54032c3465e8f0e6d3a1f40cb51cfed054f89d533201a383028d2c99d86c",
"chips_20260312.json.gz": "7af258e7962d42905efa530839ef20a9eefdbfd1170bbc6497a3cff800b86c12",
"chips_20260313.json.gz": "a93c6d8e750cee9dc897f2fa55eadc4cb4bb5c7bf0f1a29769f9af08004d871d",
"chips_20260319.json.gz": "b53ff45fc72af71cca937de2b1b9bffb57ae7bd54c2a3e64b4516caf7cd1d579",
"chips_20260320.json.gz": "feee9aa4d5d68de95dc790827e12114c7af25110e55127716add82bd430cae3d",
"chips_20260326.json.gz": "2e76703d7f5394c171b442efa0b55835d2498d3e43b4bb428e95d9dd15dc087b",
"chips_20260327.json.gz": "7277cc7c1e42926381fabc08a442a9e052d6a0c6af701af2c14d0179fff98182",
"chips_20260401.json.gz": "6c84e646b3b7ad4321f12a4024195e18f6cf4e5acf9b3467553a362ad38bfc52",
"chips_20260403.json.gz": "4ab42a7a7410b3316b4d41401b93e5a7f1d5805af6d25d7299d65e943d98924d",
"chips_20260407.json.gz": "50fe3d82e74a888240e50a9bb43e39d2f87bf3251d02b83deba9c7570d9b3974",
"chips_20260410.json.gz": "8c6b5a381b94116aabfd631a2b338d4d8c83231695e507390f79b486a38871a5",
"chips_20260414.json.gz": "ba6b889e3753daaa653137bb4417c88bf769a4e29d03aaa6ee73309d3d3da840",
"chips_20260417.json.gz": "b968d36d7a27270945feba2f57457f7cb642574c09eeb8d5ba337cc63717c130",
"chips_20260421.json.gz": "a139c7a558265ea115fea0dd201036d355be09ce591761d5524ed40a6875e613",
"chips_20260424.json.gz": "e4246010d4a4b16a68d749f6dd0a2aaa4f1db93c9e63dd7fa6f6a3b552c03873",
"chips_20260428.json.gz": "9e2f708602d15d3b1e139de15dfb9a4e1fc48ef242d7347e07434917bc994b98",
"chips_20260430.json.gz": "c16f426b5cf50ac1681840e754abc4fd06d57350e08ea3df9d342d1863f953fc",
"chips_20260508.json.gz": "a40c96084d3790a61797874a72513635ef3bda0cbadba41a8fb36ccccfc394ee",
"chips_20260515.json.gz": "36727f5b488602868ecc52258c3400649536d9f6f6994dc76935468f86944b8d",
"chips_20260521.json.gz": "fd4943d7eaa4aae27e98fb89c8a23959fb31139e488edd71cffefc537edf7de0",
"chips_20260522.json.gz": "2f63a089d72fd278544e9eae8ebacaef52980e1e5aea5567870b5e735a0e69da",
"chips_20260528.json.gz": "0c9507c476c4a6d9ffc5c310388ca25e336a48e0067893acc7ad92e9f0d2b40d",
"chips_20260529.json.gz": "6ccc1c9a46d9bbd109bc816b2b2f250a96ba6c73861c323c1f202ad309f496af",
"chips_20260604.json.gz": "1de82d9009a1de442f3711edeaad30e2001f54e468ce0b49e93fefebd9d7e6cb",
"chips_20260605.json.gz": "95428dec2d19f948d3f70b66c8ccd28ef77053555c5b19e9051d29cb955993da",
"chips_20260611.json.gz": "4cf3f312653b8d447e804253b6c3c043467d633d2754c703117d1d0df03fafa4",
"chips_20260612.json.gz": "c53561b2ea5445eb88db22ead7dcae7982d5c4749aa3f82fbb26cab50237a0df",
"chips_20260618.json.gz": "63ce1277db759bf996a29cfd60bfb3e71c702b7e5ee3798a6674ce47d808c88a",
"chips_20260626.json.gz": "ca3c33ebf6482b4bf33aefae28c9e3741574429c9fa93984c8aa6fb4d1ce2e87",
"chips_20260703.json.gz": "7ec70d81aaa71df0754247d75517d827d149cf23afb302a2c16b183ce4951fdb",
"chips_20260710.json.gz": "1290cc690a266eec6f80f1224a28f40dbe14233fd30fa21e67cf5c1409e5f50e",
"chips_20260717.json.gz": "d6f8d60587a26fbc92c88cd2f3a92d95ba4444616e916aac66c25935ed047274",
"chips_20260724.json.gz": "06e0a892daac91273dcef5598b33c48b2455120dee2390609f62f08dc720c473",
"chips_20260731.json.gz": "1b6e9f5cd9877a65eea1c8f7021850545d8a36591caf41b6fab0f5a83ad8049e",
"chips_20260807.json.gz": "8d69d51f1620f61aadd974e2daf7854130038c622ce2127f69a7f7d9b69316ca",
"chips_20260814.json.gz": "1e92ef067c860287c42eb63be61ed58bbf245512c0f2a44157ad45c60bc3fc15",
"chips_20260821.json.gz": "5ab0ea678066a3bde80d0bdfacf19e611f764272a440c7538244a071d0399470",
"chips_20260827.json.gz": "e1f33113cc4a440edda245379e282c5ad0bbb7051f8492f03567032aac4459cf",
"chips_20260828.json.gz": "f2782085e6571ae974c249b4d8e8e9a981123c0a11db01db88dc448ecf706354",
"chips_20260901.json.gz": "15831753b678b6ae4c8a889d879c13a49511efa018dd1a85a01aa99314c246f0",
"chips_20260903.json.gz": "836ab57b65fa6db08ce7feff2eb05443c04f69d80ef1c04d9f06a55ce833495f",
"chips_20260904.json.gz": "61ef196f8decc891d3db368a72827109d8bfa0f65eb80585595fefd09b7f402c",
"chips_20260910.json.gz": "6bc5d4664f98dc8345cbfd92c25644d6e7f611a8f47c396f95745a3663a92694",
"chips_20260911.json.gz": "a1c3548f750be2d20b19385503cf2c7bdb78c7b008385d4b161c41cea5e687d5",
"chips_20260917.json.gz": "42973caa57028be0f9f6c1eefb9bcde39b87e1576eae785f64afb0c3db0d4b8c",
"chips_20260918.json.gz": "330a990fc9546c352ba92310e4cca999ac7bbc05b6879fd48e9d08554e7fdb8a",
"chips_20260924.json.gz": "85b14e201314fb40be83c4b44ac7192670ea9943e500cb14c9dcd6ced9986bbf",
"chips_20260930.json.gz": "54ba4456586b1480d3d3113e1d989c52737fe4d6e04012c01f49bc50d3dc8912",
"chips_20261009.json.gz": "b27abec954a6e40e834da2b9ff75639cd4319c66f47b71a4c7be6f3c3ca11b77",
"chips_20261016.json.gz": "3596f3ab440c9e83920cff496211d741e8c01de1f0f802e2ce3b58ae16fbd1b4",
"chips_20261023.json.gz": "520f50a6e315c7065573ac34fc1d357a24ef3faefcdf3df78bb105c82f0f33c2"
},
"audit": {
"raw_rows": 2665430,
"requested_stock_dates": 25973,
"OK": 18914,
"coarse_for_valley": 6615,
"invalid_values": 77
},
"samples": 14595,
"valid_chip_samples": 13247,
"valley_evaluable_samples": 9762,
"upper_gap_samples": 1488,
"above_valley_samples": 135,
"chip_created_later_than_data_day": 1305,
"fixed_price_change_pairs": 11908,
"trap_reconciliation_rows": 2276,
"trap_max_abs_difference_pp": {
"above_difference_pp": 2.65400239784195,
"far15_difference_pp": 2.69370725854829,
"near15_difference_pp": 95.79043591281744
},
"winner_summary_abs_gap_pp_quantiles": {
"0.5": 0.007819781978170681,
"0.9": 0.05977213671796733,
"0.99": 15.852687668885588
},
"weighted_cost_abs_error_quantiles": {
"0.5": 0.00036294927893165685,
"0.9": 0.0009103229545741609,
"0.99": 0.0018411818451217628
}
}

View File

@ -0,0 +1,52 @@
"""只使用当日筹码分布识别上方稀疏与双峰低密度区。"""
import numpy as np
def runs(mask):
out=[];start=None
for i,v in enumerate(list(mask)+[False]):
if v and start is None:start=i
if not v and start is not None:out.append((start,i));start=None
return out
def chip_features(prices,percent,close):
p=np.asarray(prices,float);w=np.asarray(percent,float)
result={'chip_valid':False,'valley_evaluable':False,'chip_reason':'invalid_values'}
if len(p)<10 or close<=0 or not np.isfinite(close) or not np.isfinite(p).all() or not np.isfinite(w).all() or np.any(p<=0) or np.any(w<0):return result
if len(np.unique(p))!=len(p):result['chip_reason']='duplicate_prices';return result
order=np.argsort(p);p,w=p[order],w[order];mass=w.sum()
result['raw_mass']=float(mass)
if not 98<=mass<=102:result['chip_reason']='mass_out_of_range';return result
w=w/mass;r=p/close
upper=float(w[(r>=1.02)&(r<=1.12)].sum())
step=float(np.median(np.diff(p))/close)
result.update(chip_valid=True,chip_reason='OK',upper_mass=upper,upper_gap=bool(upper<=.05),
overhead=float(w[r>1].sum()),overhead_near15=float(w[(r>1)&(r<=1.15)].sum()),
overhead_far15=float(w[r>1.15].sum()),winner_from_dist=float(w[r<=1].sum()),
weighted_cost=float(np.sum(p*w)),peak_price=float(p[np.argmax(w)]),step_ratio=step,
valley_above=False,valley_crossing=False,valley_below=False,valley_width=0.,valley_depth=None,
valley_low=None,valley_high=None)
if step>.02:
result['chip_reason']='coarse_for_valley';return result
result['valley_evaluable']=True
edges=np.linspace(.5,1.5,101);centers=(edges[1:]+edges[:-1])/2
h=np.histogram(r,bins=edges,weights=w)[0]
smooth=np.convolve(h,[.25,.5,.25],mode='same')
peaks=[i for i in range(2,98) if .70<=centers[i]<=1.30 and smooth[i]>smooth[i-1] and smooth[i]>=smooth[i+1] and h[i-2:i+3].sum()>=.05]
valleys=[]
for left,right in zip(peaks[:-1],peaks[1:]):
peak_density=min(h[left],h[right])
if peak_density<=0:continue
threshold=.25*peak_density
for lo,hi in runs(h[left+1:right]<=threshold):
lo+=left+1;hi+=left+1
if hi-lo<5:continue
low,high=float(edges[lo]),float(edges[hi])
position='above' if low>=1.02-1e-10 else ('below' if high<=1 else ('crossing' if low<=1<high else 'near_above'))
result['valley_'+position]=True
valleys.append({'position':position,'low':low,'high':high,'depth':float(h[lo:hi].mean()/peak_density)})
above=[v for v in valleys if v['position']=='above']
if above:
v=min(above,key=lambda v:(v['low'],v['high']))
result.update(valley_low=v['low'],valley_high=v['high'],valley_width=v['high']-v['low'],valley_depth=v['depth'])
result['valleys']=valleys
return result

View File

@ -0,0 +1,51 @@
"""离线公共函数。研究口径和原始快照均以摘要固定。"""
import gzip
import hashlib
import json
import re
import sys
from pathlib import Path
import numpy as np
ROOT=Path(__file__).resolve().parent
OLD=ROOT.parent/'macro_timing_20261026'
def read(path):
with gzip.open(path,'rt') if str(path).endswith('.gz') else open(path) as f:
return json.load(f)
def dump(path,value):
Path(path).write_text(json.dumps(value,ensure_ascii=False,indent=2,allow_nan=False)+'\n')
def digest(path): return hashlib.sha256(Path(path).read_bytes()).hexdigest()
def protocol():
assert digest(ROOT/'protocol.json')==(ROOT/'protocol.sha256').read_text().split()[0]
return read(ROOT/'protocol.json')
def offline(event,args):
if event in ('socket.connect','socket.getaddrinfo','socket.sendto','socket.sendmsg'):
raise RuntimeError('离线研究禁止联网')
def install_offline(): sys.addaudithook(offline)
def dot(s):
s=str(s)
return s[2:]+'.'+s[:2] if re.fullmatch(r'(SH|SZ|BJ)\d{6}',s) else s
def prefix(s): return s[-2:]+s[:6]
def previous_sum(x,window):
"""前window个市场交易日,不含当天;不足窗口为NaN。"""
x=np.asarray(x,dtype=float)
bad=np.isnan(x)
totals=np.concatenate([np.zeros((1,)+x.shape[1:]),np.cumsum(np.nan_to_num(x),axis=0)])
missing=np.concatenate([np.zeros((1,)+x.shape[1:]),np.cumsum(bad,axis=0)])
out=np.full(x.shape,np.nan)
out[window:]=np.where(missing[window:-1]-missing[:-window-1]==0,
totals[window:-1]-totals[:-window-1],np.nan)
return out
def sample_slot(code): return int(hashlib.sha256(code.encode()).hexdigest()[:8],16)%14
def finite_float(x): return float(x) if np.isfinite(x) else None

View File

@ -0,0 +1,60 @@
{
"interpretation": "源字段和冻结特征的口径诊断,未改变原始分组、阈值或收益计算。",
"distribution_missing_samples": 1308,
"distribution_invalid_samples": 40,
"coarse_valid_samples": 3485,
"coarse_upper_gap_samples": 209,
"winner_gap_over5pp_samples": 579,
"winner_gap_over5pp_coarse_samples": 156,
"trap_comparison_n": 2286,
"trap_close_mismatch_rows": 2,
"trap_close_max_difference": 0.2099999999999973,
"trap_abs_error_pp_quantiles": {
"raw_above_gt": {
"0.5": 7.105427357601002e-15,
"0.9": 2.842170943040401e-14,
"0.99": 4.263256414560601e-14,
"1.0": 7.105427357601002e-14
},
"raw_above_ge": {
"0.5": 1.4210854715202004e-14,
"0.9": 4.263256414560601e-14,
"0.99": 17.985000000000085,
"1.0": 45.98000000000002
},
"raw_far15_gt": {
"0.5": 4.440892098500626e-16,
"0.9": 7.105427357601002e-15,
"0.99": 2.1316282072803006e-14,
"1.0": 2.6899999999999977
},
"raw_far15_ge": {
"0.5": 4.440892098500626e-16,
"0.9": 7.105427357601002e-15,
"0.99": 2.1316282072803006e-14,
"1.0": 4.263256414560601e-14
},
"raw_near15": {
"0.5": 38.019999999999996,
"0.9": 72.13,
"0.99": 87.88900000000004,
"1.0": 95.80999999999999
}
},
"thin_export_dates": {
"20260522": {
"requested": 327,
"returned": 2
},
"20261016": {
"requested": 307,
"returned": 265
}
},
"chip_valid_reason_counts": {
"OK": 9762,
"coarse_for_valley": 3485,
"missing_distribution": 1308,
"invalid_values": 40
}
}

View File

@ -0,0 +1,50 @@
"""数据口径诊断。只比较源字段和冻结特征,不读取收益或修改分组。"""
from collections import defaultdict
import numpy as np
import pandas as pd
from common import ROOT,read,dump,dot,install_offline
def quantiles(values):
return {str(k):float(v) for k,v in pd.Series(values,dtype=float).abs().quantile([.5,.9,.99,1]).items()}
def main():
install_offline()
f=pd.read_csv(ROOT/'cache/features_before_outcomes.csv.gz')
trap={(r['trade_date'],dot(r['symbol'])):r for r in read(ROOT/'raw/trap.json.gz')['data']['rows']}
metrics=defaultdict(list);requested={};returned={}
for file in sorted((ROOT/'raw').glob('chips_*.json.gz')):
d=read(file)['data'];rows=defaultdict(list)
for r in d['distribution']['rows']:rows[(r['trade_date'],dot(r['symbol']))].append(r)
day=file.name[6:14];requested[day]=len(d['requested_symbols']);returned[day]=len(rows)
for key,rr in rows.items():
t=trap.get(key)
if not t or not t['close'] or t['ocr_above'] is None or t['ocr_115'] is None:continue
prices=np.array([r['price'] for r in rr],float);weights=np.array([r['percent'] for r in rr],float)
close=float(t['close'])
for name,mask,reference in [
('raw_above_gt',prices>close,t['ocr_above']),
('raw_above_ge',prices>=close,t['ocr_above']),
('raw_far15_gt',prices>close*1.15,t['ocr_115']),
('raw_far15_ge',prices>=close*1.15,t['ocr_115']),
('raw_near15', (prices>close)&(prices<=close*1.15),t['ocr_115'])]:
metrics[name].append(float(weights[mask].sum()-reference))
valid=f[f.chip_valid];gap=valid[valid.upper_gap==True]
winner=valid[valid.winner_summary_gap_pp.abs()>5]
reconciliation=pd.read_csv(ROOT/'trap_reconciliation.csv')
dump(ROOT/'data_diagnostics.json',{
'interpretation':'源字段和冻结特征的口径诊断,未改变原始分组、阈值或收益计算。',
'distribution_missing_samples':int((f.chip_reason=='missing_distribution').sum()),
'distribution_invalid_samples':int((~f.chip_valid&(f.chip_reason!='missing_distribution')).sum()),
'coarse_valid_samples':int((valid.step_ratio>.02).sum()),
'coarse_upper_gap_samples':int((gap.step_ratio>.02).sum()),
'winner_gap_over5pp_samples':len(winner),
'winner_gap_over5pp_coarse_samples':int((winner.step_ratio>.02).sum()),
'trap_comparison_n':len(metrics['raw_above_gt']),
'trap_close_mismatch_rows':int((reconciliation.close_difference.abs()>.005).sum()),
'trap_close_max_difference':float(reconciliation.close_difference.abs().max()),
'trap_abs_error_pp_quantiles':{k:quantiles(v) for k,v in metrics.items()},
'thin_export_dates':{k:{'requested':requested[k],'returned':v} for k,v in returned.items() if v<requested[k]*.9},
'chip_valid_reason_counts':{str(k):int(v) for k,v in f.chip_reason.value_counts(dropna=False).items()}})
print('源数据口径诊断已保存;未读取收益。')
if __name__=='__main__':main()

View File

@ -0,0 +1,7 @@
{
"frozen_at": "2026-10-30T09:36:03.682235+08:00",
"features_sha256": "200e4f80acc710bc7351290ba8e03aa7aa7e8ccdd0da4e4781767e834712be61",
"protocol_sha256": "afd62f6ab1b0cf74182df468c3a98bf244f135906be348460b912cd99d2aedaf",
"implementation_sha256": "a1993225778c5714eb073a245bc1a7a118d7079c389263a04e534cd6f2acac0c",
"sample_sha256": "3f8ccc9857c02f9637fbb7cb05ff8f08231972029f75303122aff1918115c299"
}

View File

@ -0,0 +1,60 @@
"""通过155独立只读研究容器导出。凭证只在环境中传递,禁止写业务表。"""
import gzip
import hashlib
import json
import shlex
import subprocess
import sys
from pathlib import Path
ROOT = Path(__file__).resolve().parent
SECRET_KEYS = ('PROXY_DB_URL', 'DB_MYSQL_URL')
WRAPPER = r'''
import json, os, subprocess, sys
r = json.load(sys.stdin)
i = json.loads(subprocess.check_output(['docker','inspect','pms-web']))[0]
v = dict(x.split('=',1) for x in i['Config']['Env'] if '=' in x)
if v.get('PMS_TABLE_PREFIX','') or v.get('PMS_STACK_PREFIX','') or i['Config']['Image'].startswith('pms-real'):
raise SystemExit('目标不是155模拟PMS配置,停止')
e = dict(os.environ)
a = ['docker','run','--rm','-i','--name','pms-chip-research-'+r['mode'].replace('_','-'),
'--label','pms.role=offline-research','--cpus','1','--memory','2g',
'--pids-limit','64','--read-only','--entrypoint','python']
for k in ['PROXY_DB_URL','DB_MYSQL_URL']:
if not v.get(k): raise SystemExit('缺源连接配置')
e[k] = v[k]; a += ['--env',k]
a += ['--env','PMS_TABLE_PREFIX=',i['Image'],'-',r['mode']]
print(json.dumps({'host':'155','prefix':'','source_image':i['Config']['Image'],
'image_id':i['Image'],'role':'readonly-research',
'business_mounts_inherited':False,'trading_credentials_inherited':False}),file=sys.stderr)
p = subprocess.run(a,input=r['code'].encode(),env=e)
raise SystemExit(p.returncode)
'''
def main():
mode = sys.argv[1]
if not all(c.isalnum() or c in '_-' for c in mode):
raise ValueError('模式格式错误')
path = ROOT/'raw'/(mode+'.json.gz')
path.parent.mkdir(exist_ok=True)
if path.exists():
json.load(gzip.open(path)); print(mode, '保留已有快照'); return
code = (ROOT/'remote.py').read_text()
if mode.startswith('chips_'):
day=mode.split('_')[1]
requested=json.loads((ROOT/'cache/chip_requests.json').read_text())[day]
code='REQUEST = '+repr(requested)+'\n'+code
with (path.parent/(mode+'.stderr')).open('wb') as log:
p = subprocess.run(['ssh','-o','BatchMode=yes','-o','ConnectTimeout=8',
'-p','2280','factor@192.168.16.155',
'python3 -c '+shlex.quote(WRAPPER)],
input=json.dumps({'mode':mode,'code':code}).encode(),
stdout=subprocess.PIPE,stderr=log,timeout=600)
if p.returncode:
raise RuntimeError(f'只读导出失败,退出码{p.returncode};未回显连接异常正文')
obj = json.loads(gzip.decompress(p.stdout))
path.write_bytes(p.stdout)
print(mode, 'bytes', len(p.stdout), 'sha256', hashlib.sha256(p.stdout).hexdigest())
if mode in ('probe','coverage'): print(json.dumps(obj,ensure_ascii=False,indent=2))
if __name__ == '__main__': main()

Binary file not shown.

After

Width:  |  Height:  |  Size: 83 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 63 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 88 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 54 KiB

View File

@ -0,0 +1,28 @@
"""记录运行依赖、源缓存和交付物摘要;不含凭证或原始数据正文。"""
import datetime as dt
import importlib.metadata
import platform
from common import ROOT,read,dump,digest,install_offline
def main():
install_offline()
packages={k:importlib.metadata.version(k) for k in ['numpy','pandas','scipy','matplotlib']}
(ROOT/'requirements.txt').write_text('\n'.join(k+'=='+v for k,v in packages.items())+'\n')
sources={}
for path in sorted((ROOT/'raw').glob('*.json.gz')):
sources[str(path.relative_to(ROOT))]={'sha256':digest(path),'bytes':path.stat().st_size}
parent_sources=read(ROOT/'price_coverage.json')['source_sha256']
artifacts={}
for path in sorted(ROOT.rglob('*')):
if not path.is_file() or path.name=='manifest.json':continue
relative=path.relative_to(ROOT)
if any(x in relative.parts for x in ['raw','cache','__pycache__']):continue
artifacts[str(relative)]={'sha256':digest(path),'bytes':path.stat().st_size}
dump(ROOT/'manifest.json',{'recorded_at':dt.datetime.now(dt.timezone(dt.timedelta(hours=8))).isoformat(),
'python':platform.python_version(),'packages':packages,'raw_sources':sources,
'price_sources_relative_to_research':parent_sources,'artifacts':artifacts,
'exclusions':'原始数据和中间缓存留在开发机;manifest不递归计算自身摘要。',
'environment':'开发机离线分析。155独立只读容器导出,无业务目录挂载及交易凭证。'})
print('交付清单已记录',len(sources),'份源快照,',len(artifacts),'份交付文件')
if __name__=='__main__':main()

View File

@ -0,0 +1,31 @@
analysis,group,signals,matured_signals,filled_matured,completed,unresolved_exit,stocks,signal_dates,first_signal,last_signal,wins,win_rate,win_rate_lower_missing_exit,win_rate_upper_missing_exit,mean_net30,median_net30,mean_net60,median_net60,win_rate_cost60,delayed_exits,path_complete,hit15,up_before_down,max_close_gain,max_close_loss
all,all,14595,10231,10183,10182,1,4993,44,2025-12-19,2026-10-23,3377,0.3316637202907091,0.3316311499558087,0.33172935284297356,-0.03478610843062442,-0.07780962491638327,-0.04056017351795038,-0.08332631639290211,0.32105676684344925,17,9692,0.39785390012381344,0.3156211308295501,0.19626120917317977,-0.17247434472636053
upper_gap,no,11759,7909,7875,7874,1,4903,41,2026-01-09,2026-10-23,2645,0.3359156718313437,0.3358730158730159,0.336,-0.03478976260561357,-0.07730100094646675,-0.0405638058330757,-0.08282073509356142,0.3251206502413005,13,7512,0.38777955271565495,0.3003194888178914,0.19120875774771415,-0.17093981266305447
upper_gap,yes,1488,1032,1020,1020,0,1296,41,2026-01-09,2026-10-23,283,0.2774509803921569,0.2774509803921569,0.2774509803921569,-0.06257209713797077,-0.130411707189515,-0.06817994196187817,-0.13561372453066667,0.26666666666666666,4,964,0.4024896265560166,0.2987551867219917,0.21627466796139683,-0.2166120016500042
valley_above,no,9627,6745,6713,6712,1,4026,41,2026-01-09,2026-10-23,2201,0.32792014302741357,0.32787129450320274,0.3280202591985699,-0.03582718341065308,-0.09069818900683047,-0.041595020619632106,-0.096137779015384,0.3176400476758045,7,6428,0.40494710640945863,0.305849408836341,0.20563959856874872,-0.1865276786366789
valley_above,yes,135,100,99,99,0,133,37,2026-01-09,2026-10-23,30,0.30303030303030304,0.30303030303030304,0.30303030303030304,-0.03686050521206846,-0.08005550655725147,-0.042622160923464776,-0.08555876280776897,0.29292929292929293,0,94,0.43617021276595747,0.3617021276595745,0.21547783096774378,-0.1538118469855404
valley_crossing,no,9757,6842,6809,6808,1,4041,41,2026-01-09,2026-10-23,2231,0.3277027027027027,0.32765457482743426,0.32780143927155236,-0.035769429999656654,-0.09023212948084569,-0.04153761269884984,-0.09567450753261036,0.31742068155111636,7,6519,0.40558367847829424,0.3067955207853965,0.20586561423300462,-0.18603449838892808
valley_crossing,yes,5,3,3,3,0,5,5,2026-03-06,2026-08-07,0,0.0,0.0,0.0,-0.20098854354523518,-0.18600692934167828,-0.20576835887465164,-0.19087636710222644,0.0,0,3,0.0,0.0,0.022772175309143477,-0.23311229859686447
valley_below,no,9678,6772,6741,6740,1,4035,41,2026-01-09,2026-10-23,2212,0.3281899109792285,0.328141225337487,0.3282895712802255,-0.03536681892790419,-0.0892519918628415,-0.04113741011142761,-0.0947002332558306,0.3178041543026706,7,6453,0.40492794049279407,0.30636913063691307,0.2054864761538279,-0.185360503904714
valley_below,yes,84,73,71,71,0,83,33,2026-01-09,2026-10-09,19,0.2676056338028169,0.2676056338028169,0.2676056338028169,-0.08097021822850688,-0.1908451474843762,-0.08646800252635972,-0.19568564224409646,0.2676056338028169,0,69,0.4492753623188406,0.3333333333333333,0.23336268246714745,-0.2511144955955527
frequency_all,0,11313,8062,8023,8023,0,4904,44,2025-12-19,2026-10-23,2688,0.3350367692882961,0.3350367692882961,0.3350367692882961,-0.032594579834655496,-0.07983751854520704,-0.03838175499930669,-0.08534207883515121,0.32543936183472516,14,7644,0.40920983778126635,0.3237833594976452,0.20503351080773505,-0.17639185330679752
frequency_all,1,3097,2073,2064,2063,1,2537,44,2025-12-19,2026-10-23,657,0.31846825012118274,0.3183139534883721,0.3187984496124031,-0.042000915686900085,-0.07525925993348259,-0.04773182063422589,-0.08079120810443619,0.3053805138148328,3,1954,0.3587512794268168,0.28607983623336747,0.16533366250147907,-0.15795268009211794
frequency_all,2+,185,96,96,96,0,184,43,2025-12-19,2026-10-23,32,0.3333333333333333,0.3333333333333333,0.3333333333333333,-0.06289534338647841,-0.06876968099112041,-0.06850125449922144,-0.07434045086939678,0.2916666666666667,0,94,0.2872340425531915,0.26595744680851063,0.12580325706640125,-0.15577112245454652
frequency_down,0,3287,2218,2209,2209,0,2793,44,2025-12-19,2026-10-23,852,0.38569488456315076,0.38569488456315076,0.38569488456315076,-0.030397601274800665,-0.04409897927547479,-0.036197919119339554,-0.0498173332682158,0.3721140787686736,4,2096,0.39217557251908397,0.32156488549618323,0.17173919198172766,-0.14781077490723682
frequency_down,1,1398,928,926,926,0,1316,44,2025-12-19,2026-10-23,350,0.3779697624190065,0.3779697624190065,0.3779697624190065,-0.030345753725446876,-0.05140654141807505,-0.0361463817303836,-0.05708118039613408,0.3617710583153348,1,870,0.40804597701149425,0.3505747126436782,0.17458343097078177,-0.1415098074918641
frequency_down,2+,105,54,54,54,0,105,35,2025-12-19,2026-10-23,25,0.46296296296296297,0.46296296296296297,0.46296296296296297,-0.018674731712418086,-0.02229585305873255,-0.02454517765235081,-0.02814463686706381,0.4074074074074074,0,53,0.39622641509433965,0.3584905660377358,0.1652674763583761,-0.12848323566271022
frequency_down_intact,0,2691,1832,1823,1823,0,2305,44,2025-12-19,2026-10-23,692,0.37959407569939657,0.37959407569939657,0.37959407569939657,-0.0319307932524109,-0.04869376246261237,-0.03772193929539635,-0.05438462972167024,0.3664289632473944,2,1732,0.39087759815242495,0.31812933025404155,0.1740004561946311,-0.15046927367380794
frequency_down_intact,1,685,428,427,427,0,661,44,2025-12-19,2026-10-23,127,0.297423887587822,0.297423887587822,0.297423887587822,-0.055946420184186926,-0.09668387760368802,-0.061593900875659933,-0.10208766025819027,0.28805620608899296,0,401,0.3915211970074813,0.3341645885286783,0.1725997455758315,-0.15995005258682635
frequency_down_intact,2+,61,24,24,24,0,61,27,2025-12-19,2026-10-23,9,0.375,0.375,0.375,-0.06296523770553535,-0.06391687999798934,-0.06857073069919516,-0.06951668011405548,0.2916666666666667,0,24,0.25,0.25,0.12352275137709319,-0.15851360246427051
upper_gap_joint_down,"gap=False,high_freq=False",4110,2745,2736,2736,0,3346,40,2026-01-09,2026-10-23,1085,0.3965643274853801,0.3965643274853801,0.3965643274853801,-0.02715525855974539,-0.03909058467563903,-0.032974972621056205,-0.04483889969220589,0.3815789473684211,4,2591,0.3971439598610575,0.32497105364724044,0.17317646368042827,-0.1439544398156561
upper_gap_joint_down,"gap=False,high_freq=True",92,43,43,43,0,92,30,2026-01-23,2026-10-23,24,0.5581395348837209,0.5581395348837209,0.5581395348837209,0.0027797285598961585,0.005604658417385888,-0.0032190614197440508,-0.00041103074802717465,0.4883720930232558,0,42,0.40476190476190477,0.35714285714285715,0.17845591152760865,-0.12020275423217264
upper_gap_joint_down,"gap=True,high_freq=False",208,52,50,50,0,205,28,2026-01-09,2026-10-23,13,0.26,0.26,0.26,-0.18055989322026378,-0.24732062090179974,-0.18546191602892684,-0.25182327027595525,0.24,1,45,0.3333333333333333,0.2222222222222222,0.146663859101299,-0.25601788323213387
upper_gap_joint_down,"gap=True,high_freq=True",4,2,2,2,0,4,3,2026-05-08,2026-08-07,0,0.0,0.0,0.0,-0.15620003960230072,-0.15620003960230072,-0.16124778698200076,-0.16124778698200076,0.0,0,2,0.5,0.5,0.08349007898788152,-0.20009003929355001
valley_above_joint_down,"gap=False,high_freq=False",2806,1775,1769,1769,0,2358,40,2026-01-09,2026-10-23,650,0.36743923120407007,0.36743923120407007,0.36743923120407007,-0.0399627010444463,-0.05855618875231605,-0.045705798920313886,-0.06418805738753208,0.3527416619559073,1,1688,0.3975118483412322,0.31398104265402843,0.17511100658048423,-0.15980317125996255
valley_above_joint_down,"gap=False,high_freq=True",57,29,29,29,0,57,29,2026-01-23,2026-10-23,13,0.4482758620689655,0.4482758620689655,0.4482758620689655,-0.028981263420900762,-0.021360671852235003,-0.03479005402678872,-0.027215050065290347,0.41379310344827586,0,28,0.32142857142857145,0.2857142857142857,0.14395458373019188,-0.13947829116484248
valley_above_joint_down,"gap=True,high_freq=False",86,63,62,62,0,85,31,2026-01-23,2026-10-23,16,0.25806451612903225,0.25806451612903225,0.25806451612903225,-0.08094023929040894,-0.08961426723906024,-0.08643820292710197,-0.0950603414423517,0.24193548387096775,0,57,0.3684210526315789,0.3157894736842105,0.15714491047986143,-0.15960575276329597
valley_above_joint_down,"gap=True,high_freq=True",1,0,0,0,0,1,1,2026-09-24,2026-09-24,0,,,,,,,,,0,0,,,,
overlap,"upper=False,valley=False",8355,5829,5808,5807,1,3890,40,2026-01-09,2026-10-23,1949,0.3356294127776821,0.33557162534435264,0.3357438016528926,-0.03181504165199738,-0.08571300151069017,-0.03760688013971514,-0.09118241371443447,0.3249526433614603,5,5570,0.4043087971274686,0.3066427289048474,0.20318742719002944,-0.18123413789699114
overlap,"upper=False,valley=True",128,96,95,95,0,126,37,2026-01-09,2026-10-23,30,0.3157894736842105,0.3157894736842105,0.3157894736842105,-0.027637546132053454,-0.07669206550687879,-0.03345437507136421,-0.08221544240393064,0.30526315789473685,0,90,0.4444444444444444,0.36666666666666664,0.22103981462814523,-0.14739367858127542
overlap,"upper=True,valley=False",1272,916,905,905,0,1108,41,2026-01-09,2026-10-23,252,0.27845303867403315,0.27845303867403315,0.27845303867403315,-0.06157139025320955,-0.13475470456709582,-0.06718522146701085,-0.139930741397864,0.27071823204419887,2,858,0.4090909090909091,0.3006993006993007,0.22155870647022471,-0.22089250604933727
overlap,"upper=True,valley=True",7,4,4,4,0,7,6,2026-01-23,2026-10-23,0,0.0,0.0,0.0,-0.2559057833624249,-0.23448199174851692,-0.2603570749108531,-0.23906144387179434,0.0,0,4,0.25,0.25,0.09033319860871059,-0.29822063608150257
1 analysis group signals matured_signals filled_matured completed unresolved_exit stocks signal_dates first_signal last_signal wins win_rate win_rate_lower_missing_exit win_rate_upper_missing_exit mean_net30 median_net30 mean_net60 median_net60 win_rate_cost60 delayed_exits path_complete hit15 up_before_down max_close_gain max_close_loss
2 all all 14595 10231 10183 10182 1 4993 44 2025-12-19 2026-10-23 3377 0.3316637202907091 0.3316311499558087 0.33172935284297356 -0.03478610843062442 -0.07780962491638327 -0.04056017351795038 -0.08332631639290211 0.32105676684344925 17 9692 0.39785390012381344 0.3156211308295501 0.19626120917317977 -0.17247434472636053
3 upper_gap no 11759 7909 7875 7874 1 4903 41 2026-01-09 2026-10-23 2645 0.3359156718313437 0.3358730158730159 0.336 -0.03478976260561357 -0.07730100094646675 -0.0405638058330757 -0.08282073509356142 0.3251206502413005 13 7512 0.38777955271565495 0.3003194888178914 0.19120875774771415 -0.17093981266305447
4 upper_gap yes 1488 1032 1020 1020 0 1296 41 2026-01-09 2026-10-23 283 0.2774509803921569 0.2774509803921569 0.2774509803921569 -0.06257209713797077 -0.130411707189515 -0.06817994196187817 -0.13561372453066667 0.26666666666666666 4 964 0.4024896265560166 0.2987551867219917 0.21627466796139683 -0.2166120016500042
5 valley_above no 9627 6745 6713 6712 1 4026 41 2026-01-09 2026-10-23 2201 0.32792014302741357 0.32787129450320274 0.3280202591985699 -0.03582718341065308 -0.09069818900683047 -0.041595020619632106 -0.096137779015384 0.3176400476758045 7 6428 0.40494710640945863 0.305849408836341 0.20563959856874872 -0.1865276786366789
6 valley_above yes 135 100 99 99 0 133 37 2026-01-09 2026-10-23 30 0.30303030303030304 0.30303030303030304 0.30303030303030304 -0.03686050521206846 -0.08005550655725147 -0.042622160923464776 -0.08555876280776897 0.29292929292929293 0 94 0.43617021276595747 0.3617021276595745 0.21547783096774378 -0.1538118469855404
7 valley_crossing no 9757 6842 6809 6808 1 4041 41 2026-01-09 2026-10-23 2231 0.3277027027027027 0.32765457482743426 0.32780143927155236 -0.035769429999656654 -0.09023212948084569 -0.04153761269884984 -0.09567450753261036 0.31742068155111636 7 6519 0.40558367847829424 0.3067955207853965 0.20586561423300462 -0.18603449838892808
8 valley_crossing yes 5 3 3 3 0 5 5 2026-03-06 2026-08-07 0 0.0 0.0 0.0 -0.20098854354523518 -0.18600692934167828 -0.20576835887465164 -0.19087636710222644 0.0 0 3 0.0 0.0 0.022772175309143477 -0.23311229859686447
9 valley_below no 9678 6772 6741 6740 1 4035 41 2026-01-09 2026-10-23 2212 0.3281899109792285 0.328141225337487 0.3282895712802255 -0.03536681892790419 -0.0892519918628415 -0.04113741011142761 -0.0947002332558306 0.3178041543026706 7 6453 0.40492794049279407 0.30636913063691307 0.2054864761538279 -0.185360503904714
10 valley_below yes 84 73 71 71 0 83 33 2026-01-09 2026-10-09 19 0.2676056338028169 0.2676056338028169 0.2676056338028169 -0.08097021822850688 -0.1908451474843762 -0.08646800252635972 -0.19568564224409646 0.2676056338028169 0 69 0.4492753623188406 0.3333333333333333 0.23336268246714745 -0.2511144955955527
11 frequency_all 0 11313 8062 8023 8023 0 4904 44 2025-12-19 2026-10-23 2688 0.3350367692882961 0.3350367692882961 0.3350367692882961 -0.032594579834655496 -0.07983751854520704 -0.03838175499930669 -0.08534207883515121 0.32543936183472516 14 7644 0.40920983778126635 0.3237833594976452 0.20503351080773505 -0.17639185330679752
12 frequency_all 1 3097 2073 2064 2063 1 2537 44 2025-12-19 2026-10-23 657 0.31846825012118274 0.3183139534883721 0.3187984496124031 -0.042000915686900085 -0.07525925993348259 -0.04773182063422589 -0.08079120810443619 0.3053805138148328 3 1954 0.3587512794268168 0.28607983623336747 0.16533366250147907 -0.15795268009211794
13 frequency_all 2+ 185 96 96 96 0 184 43 2025-12-19 2026-10-23 32 0.3333333333333333 0.3333333333333333 0.3333333333333333 -0.06289534338647841 -0.06876968099112041 -0.06850125449922144 -0.07434045086939678 0.2916666666666667 0 94 0.2872340425531915 0.26595744680851063 0.12580325706640125 -0.15577112245454652
14 frequency_down 0 3287 2218 2209 2209 0 2793 44 2025-12-19 2026-10-23 852 0.38569488456315076 0.38569488456315076 0.38569488456315076 -0.030397601274800665 -0.04409897927547479 -0.036197919119339554 -0.0498173332682158 0.3721140787686736 4 2096 0.39217557251908397 0.32156488549618323 0.17173919198172766 -0.14781077490723682
15 frequency_down 1 1398 928 926 926 0 1316 44 2025-12-19 2026-10-23 350 0.3779697624190065 0.3779697624190065 0.3779697624190065 -0.030345753725446876 -0.05140654141807505 -0.0361463817303836 -0.05708118039613408 0.3617710583153348 1 870 0.40804597701149425 0.3505747126436782 0.17458343097078177 -0.1415098074918641
16 frequency_down 2+ 105 54 54 54 0 105 35 2025-12-19 2026-10-23 25 0.46296296296296297 0.46296296296296297 0.46296296296296297 -0.018674731712418086 -0.02229585305873255 -0.02454517765235081 -0.02814463686706381 0.4074074074074074 0 53 0.39622641509433965 0.3584905660377358 0.1652674763583761 -0.12848323566271022
17 frequency_down_intact 0 2691 1832 1823 1823 0 2305 44 2025-12-19 2026-10-23 692 0.37959407569939657 0.37959407569939657 0.37959407569939657 -0.0319307932524109 -0.04869376246261237 -0.03772193929539635 -0.05438462972167024 0.3664289632473944 2 1732 0.39087759815242495 0.31812933025404155 0.1740004561946311 -0.15046927367380794
18 frequency_down_intact 1 685 428 427 427 0 661 44 2025-12-19 2026-10-23 127 0.297423887587822 0.297423887587822 0.297423887587822 -0.055946420184186926 -0.09668387760368802 -0.061593900875659933 -0.10208766025819027 0.28805620608899296 0 401 0.3915211970074813 0.3341645885286783 0.1725997455758315 -0.15995005258682635
19 frequency_down_intact 2+ 61 24 24 24 0 61 27 2025-12-19 2026-10-23 9 0.375 0.375 0.375 -0.06296523770553535 -0.06391687999798934 -0.06857073069919516 -0.06951668011405548 0.2916666666666667 0 24 0.25 0.25 0.12352275137709319 -0.15851360246427051
20 upper_gap_joint_down gap=False,high_freq=False 4110 2745 2736 2736 0 3346 40 2026-01-09 2026-10-23 1085 0.3965643274853801 0.3965643274853801 0.3965643274853801 -0.02715525855974539 -0.03909058467563903 -0.032974972621056205 -0.04483889969220589 0.3815789473684211 4 2591 0.3971439598610575 0.32497105364724044 0.17317646368042827 -0.1439544398156561
21 upper_gap_joint_down gap=False,high_freq=True 92 43 43 43 0 92 30 2026-01-23 2026-10-23 24 0.5581395348837209 0.5581395348837209 0.5581395348837209 0.0027797285598961585 0.005604658417385888 -0.0032190614197440508 -0.00041103074802717465 0.4883720930232558 0 42 0.40476190476190477 0.35714285714285715 0.17845591152760865 -0.12020275423217264
22 upper_gap_joint_down gap=True,high_freq=False 208 52 50 50 0 205 28 2026-01-09 2026-10-23 13 0.26 0.26 0.26 -0.18055989322026378 -0.24732062090179974 -0.18546191602892684 -0.25182327027595525 0.24 1 45 0.3333333333333333 0.2222222222222222 0.146663859101299 -0.25601788323213387
23 upper_gap_joint_down gap=True,high_freq=True 4 2 2 2 0 4 3 2026-05-08 2026-08-07 0 0.0 0.0 0.0 -0.15620003960230072 -0.15620003960230072 -0.16124778698200076 -0.16124778698200076 0.0 0 2 0.5 0.5 0.08349007898788152 -0.20009003929355001
24 valley_above_joint_down gap=False,high_freq=False 2806 1775 1769 1769 0 2358 40 2026-01-09 2026-10-23 650 0.36743923120407007 0.36743923120407007 0.36743923120407007 -0.0399627010444463 -0.05855618875231605 -0.045705798920313886 -0.06418805738753208 0.3527416619559073 1 1688 0.3975118483412322 0.31398104265402843 0.17511100658048423 -0.15980317125996255
25 valley_above_joint_down gap=False,high_freq=True 57 29 29 29 0 57 29 2026-01-23 2026-10-23 13 0.4482758620689655 0.4482758620689655 0.4482758620689655 -0.028981263420900762 -0.021360671852235003 -0.03479005402678872 -0.027215050065290347 0.41379310344827586 0 28 0.32142857142857145 0.2857142857142857 0.14395458373019188 -0.13947829116484248
26 valley_above_joint_down gap=True,high_freq=False 86 63 62 62 0 85 31 2026-01-23 2026-10-23 16 0.25806451612903225 0.25806451612903225 0.25806451612903225 -0.08094023929040894 -0.08961426723906024 -0.08643820292710197 -0.0950603414423517 0.24193548387096775 0 57 0.3684210526315789 0.3157894736842105 0.15714491047986143 -0.15960575276329597
27 valley_above_joint_down gap=True,high_freq=True 1 0 0 0 0 1 1 2026-09-24 2026-09-24 0 0 0
28 overlap upper=False,valley=False 8355 5829 5808 5807 1 3890 40 2026-01-09 2026-10-23 1949 0.3356294127776821 0.33557162534435264 0.3357438016528926 -0.03181504165199738 -0.08571300151069017 -0.03760688013971514 -0.09118241371443447 0.3249526433614603 5 5570 0.4043087971274686 0.3066427289048474 0.20318742719002944 -0.18123413789699114
29 overlap upper=False,valley=True 128 96 95 95 0 126 37 2026-01-09 2026-10-23 30 0.3157894736842105 0.3157894736842105 0.3157894736842105 -0.027637546132053454 -0.07669206550687879 -0.03345437507136421 -0.08221544240393064 0.30526315789473685 0 90 0.4444444444444444 0.36666666666666664 0.22103981462814523 -0.14739367858127542
30 overlap upper=True,valley=False 1272 916 905 905 0 1108 41 2026-01-09 2026-10-23 252 0.27845303867403315 0.27845303867403315 0.27845303867403315 -0.06157139025320955 -0.13475470456709582 -0.06718522146701085 -0.139930741397864 0.27071823204419887 2 858 0.4090909090909091 0.3006993006993007 0.22155870647022471 -0.22089250604933727
31 overlap upper=True,valley=True 7 4 4 4 0 7 6 2026-01-23 2026-10-23 0 0.0 0.0 0.0 -0.2559057833624249 -0.23448199174851692 -0.2603570749108531 -0.23906144387179434 0.0 0 4 0.25 0.25 0.09033319860871059 -0.29822063608150257

View File

@ -0,0 +1,9 @@
{
"frozen_at": "2026-10-30T09:18:48.988949+08:00",
"stage": "未计算任何未来收益前的实现细化",
"count_vs_duration_sensitivity": "在吸筹次数配对中附加明确吸筹天数差不超过5天;不满足不放宽。",
"mechanism": "将t-20日筹码价格乘以当日复权因子再除以t日复权因子,统一到t日价格坐标;比较固定t日收盘阈值之上的筹码占比变化。解释变量为t-20日前20日吸筹次数,控制t-20至t的收益。",
"bootstrap_stock": "联合重抽所有出现的股票,配对权重为处理股与对照股抽样次数的乘积。两类重抽区间分列,不解释为完整市场阶段覆盖。",
"industry": "读取截止日之前全部行业历史记录,再按trade_date选择当时最近一条;披露回填时间,不以最新分类直接覆盖旧日期。",
"label_families": "三个正式主检验均只对应60日净胜率;其余均探索,不据此升级结论。"
}

View File

@ -0,0 +1 @@
a1993225778c5714eb073a245bc1a7a118d7079c389263a04e534cd6f2acac0c implementation_notes.json

View File

@ -0,0 +1,532 @@
{
"recorded_at": "2026-10-30T09:45:26.502311+08:00",
"python": "3.12.14",
"packages": {
"numpy": "2.3.5",
"pandas": "2.2.3",
"scipy": "1.16.3",
"matplotlib": "3.10.7"
},
"raw_sources": {
"raw/calendar.json.gz": {
"sha256": "62b8a2e4a4781e4eb3e230417892e7c6555e2f42ef8cfd5ba96ea5a3bf8863ac",
"bytes": 144
},
"raw/calendar_probe.json.gz": {
"sha256": "e1bae1a7bc9d6c0249b26ca2a47add841fba5bee6fd42577ff4ecff240633b52",
"bytes": 458
},
"raw/chip_dates.json.gz": {
"sha256": "e9314df115cf0e68b81c59a8e8dd203c53b2b09f7757fe8f718f8dfe5020c222",
"bytes": 1442
},
"raw/chips_20260109.json.gz": {
"sha256": "bdd585d23b7932ea9923fb4eaf55ca044b7692adf664c45d9ddd7502371131c8",
"bytes": 423545
},
"raw/chips_20260116.json.gz": {
"sha256": "0148cb0e0afc478ebc506c0e9504b9b241a079b8ab665657e36358771393d7a3",
"bytes": 409166
},
"raw/chips_20260122.json.gz": {
"sha256": "b81c26d9b02c47454ae999145e7e2baebfa7adb834d3ae8a6e94b3ed1f8da1c9",
"bytes": 223518
},
"raw/chips_20260123.json.gz": {
"sha256": "fc030ce8f983ead18fc6b1ac0bbadff98dbf9bed37160a516fcfc2038f4312bd",
"bytes": 223437
},
"raw/chips_20260129.json.gz": {
"sha256": "f9e42a975e189260fe57bfae784df97963fb2cd270364323da8b28fe18bc22c8",
"bytes": 218812
},
"raw/chips_20260130.json.gz": {
"sha256": "2147bbfc5f8c20b961d24acd3484649b4570f4840485857ac5183faae469d9de",
"bytes": 226589
},
"raw/chips_20260205.json.gz": {
"sha256": "aadfb6cce8e0e46126f808cd7476cdcb45b5c5f4745cbd8adf4612aea493b3d3",
"bytes": 215512
},
"raw/chips_20260206.json.gz": {
"sha256": "76951fb0c608ff8dcf4e51af6ba940000762425ccc0c67abc95a24ac7adcf01d",
"bytes": 208344
},
"raw/chips_20260212.json.gz": {
"sha256": "97553aa79a25837512f8628c3168229cceb8f938dabd1b6a972c78f167ad101c",
"bytes": 215249
},
"raw/chips_20260213.json.gz": {
"sha256": "71e0df0a6b5259c05d8058ccdac34646013f4324503a4aa0d5c2f7e72f62a134",
"bytes": 221426
},
"raw/chips_20260227.json.gz": {
"sha256": "5ecd01a1e904d5ebe6bde68feb3f10ceb05cf638a2ddc91b8fd8eb22c36dd711",
"bytes": 428913
},
"raw/chips_20260306.json.gz": {
"sha256": "d2ea54032c3465e8f0e6d3a1f40cb51cfed054f89d533201a383028d2c99d86c",
"bytes": 425528
},
"raw/chips_20260312.json.gz": {
"sha256": "7af258e7962d42905efa530839ef20a9eefdbfd1170bbc6497a3cff800b86c12",
"bytes": 229177
},
"raw/chips_20260313.json.gz": {
"sha256": "a93c6d8e750cee9dc897f2fa55eadc4cb4bb5c7bf0f1a29769f9af08004d871d",
"bytes": 206404
},
"raw/chips_20260319.json.gz": {
"sha256": "b53ff45fc72af71cca937de2b1b9bffb57ae7bd54c2a3e64b4516caf7cd1d579",
"bytes": 206301
},
"raw/chips_20260320.json.gz": {
"sha256": "feee9aa4d5d68de95dc790827e12114c7af25110e55127716add82bd430cae3d",
"bytes": 218369
},
"raw/chips_20260326.json.gz": {
"sha256": "2e76703d7f5394c171b442efa0b55835d2498d3e43b4bb428e95d9dd15dc087b",
"bytes": 239609
},
"raw/chips_20260327.json.gz": {
"sha256": "7277cc7c1e42926381fabc08a442a9e052d6a0c6af701af2c14d0179fff98182",
"bytes": 215036
},
"raw/chips_20260401.json.gz": {
"sha256": "6c84e646b3b7ad4321f12a4024195e18f6cf4e5acf9b3467553a362ad38bfc52",
"bytes": 203606
},
"raw/chips_20260403.json.gz": {
"sha256": "4ab42a7a7410b3316b4d41401b93e5a7f1d5805af6d25d7299d65e943d98924d",
"bytes": 213769
},
"raw/chips_20260407.json.gz": {
"sha256": "50fe3d82e74a888240e50a9bb43e39d2f87bf3251d02b83deba9c7570d9b3974",
"bytes": 241134
},
"raw/chips_20260410.json.gz": {
"sha256": "8c6b5a381b94116aabfd631a2b338d4d8c83231695e507390f79b486a38871a5",
"bytes": 235640
},
"raw/chips_20260414.json.gz": {
"sha256": "ba6b889e3753daaa653137bb4417c88bf769a4e29d03aaa6ee73309d3d3da840",
"bytes": 239906
},
"raw/chips_20260417.json.gz": {
"sha256": "b968d36d7a27270945feba2f57457f7cb642574c09eeb8d5ba337cc63717c130",
"bytes": 211146
},
"raw/chips_20260421.json.gz": {
"sha256": "a139c7a558265ea115fea0dd201036d355be09ce591761d5524ed40a6875e613",
"bytes": 221162
},
"raw/chips_20260424.json.gz": {
"sha256": "e4246010d4a4b16a68d749f6dd0a2aaa4f1db93c9e63dd7fa6f6a3b552c03873",
"bytes": 237016
},
"raw/chips_20260428.json.gz": {
"sha256": "9e2f708602d15d3b1e139de15dfb9a4e1fc48ef242d7347e07434917bc994b98",
"bytes": 227523
},
"raw/chips_20260430.json.gz": {
"sha256": "c16f426b5cf50ac1681840e754abc4fd06d57350e08ea3df9d342d1863f953fc",
"bytes": 200793
},
"raw/chips_20260508.json.gz": {
"sha256": "a40c96084d3790a61797874a72513635ef3bda0cbadba41a8fb36ccccfc394ee",
"bytes": 472571
},
"raw/chips_20260515.json.gz": {
"sha256": "36727f5b488602868ecc52258c3400649536d9f6f6994dc76935468f86944b8d",
"bytes": 476050
},
"raw/chips_20260521.json.gz": {
"sha256": "fd4943d7eaa4aae27e98fb89c8a23959fb31139e488edd71cffefc537edf7de0",
"bytes": 230488
},
"raw/chips_20260522.json.gz": {
"sha256": "2f63a089d72fd278544e9eae8ebacaef52980e1e5aea5567870b5e735a0e69da",
"bytes": 2235
},
"raw/chips_20260528.json.gz": {
"sha256": "0c9507c476c4a6d9ffc5c310388ca25e336a48e0067893acc7ad92e9f0d2b40d",
"bytes": 224169
},
"raw/chips_20260529.json.gz": {
"sha256": "6ccc1c9a46d9bbd109bc816b2b2f250a96ba6c73861c323c1f202ad309f496af",
"bytes": 223742
},
"raw/chips_20260604.json.gz": {
"sha256": "1de82d9009a1de442f3711edeaad30e2001f54e468ce0b49e93fefebd9d7e6cb",
"bytes": 214689
},
"raw/chips_20260605.json.gz": {
"sha256": "95428dec2d19f948d3f70b66c8ccd28ef77053555c5b19e9051d29cb955993da",
"bytes": 226380
},
"raw/chips_20260611.json.gz": {
"sha256": "4cf3f312653b8d447e804253b6c3c043467d633d2754c703117d1d0df03fafa4",
"bytes": 204013
},
"raw/chips_20260612.json.gz": {
"sha256": "c53561b2ea5445eb88db22ead7dcae7982d5c4749aa3f82fbb26cab50237a0df",
"bytes": 234730
},
"raw/chips_20260618.json.gz": {
"sha256": "63ce1277db759bf996a29cfd60bfb3e71c702b7e5ee3798a6674ce47d808c88a",
"bytes": 440744
},
"raw/chips_20260626.json.gz": {
"sha256": "ca3c33ebf6482b4bf33aefae28c9e3741574429c9fa93984c8aa6fb4d1ce2e87",
"bytes": 425774
},
"raw/chips_20260703.json.gz": {
"sha256": "7ec70d81aaa71df0754247d75517d827d149cf23afb302a2c16b183ce4951fdb",
"bytes": 443516
},
"raw/chips_20260710.json.gz": {
"sha256": "1290cc690a266eec6f80f1224a28f40dbe14233fd30fa21e67cf5c1409e5f50e",
"bytes": 389661
},
"raw/chips_20260717.json.gz": {
"sha256": "d6f8d60587a26fbc92c88cd2f3a92d95ba4444616e916aac66c25935ed047274",
"bytes": 441983
},
"raw/chips_20260724.json.gz": {
"sha256": "06e0a892daac91273dcef5598b33c48b2455120dee2390609f62f08dc720c473",
"bytes": 447479
},
"raw/chips_20260731.json.gz": {
"sha256": "1b6e9f5cd9877a65eea1c8f7021850545d8a36591caf41b6fab0f5a83ad8049e",
"bytes": 460822
},
"raw/chips_20260807.json.gz": {
"sha256": "8d69d51f1620f61aadd974e2daf7854130038c622ce2127f69a7f7d9b69316ca",
"bytes": 403646
},
"raw/chips_20260814.json.gz": {
"sha256": "1e92ef067c860287c42eb63be61ed58bbf245512c0f2a44157ad45c60bc3fc15",
"bytes": 444832
},
"raw/chips_20260821.json.gz": {
"sha256": "5ab0ea678066a3bde80d0bdfacf19e611f764272a440c7538244a071d0399470",
"bytes": 463626
},
"raw/chips_20260827.json.gz": {
"sha256": "e1f33113cc4a440edda245379e282c5ad0bbb7051f8492f03567032aac4459cf",
"bytes": 233754
},
"raw/chips_20260828.json.gz": {
"sha256": "f2782085e6571ae974c249b4d8e8e9a981123c0a11db01db88dc448ecf706354",
"bytes": 220929
},
"raw/chips_20260901.json.gz": {
"sha256": "15831753b678b6ae4c8a889d879c13a49511efa018dd1a85a01aa99314c246f0",
"bytes": 230258
},
"raw/chips_20260903.json.gz": {
"sha256": "836ab57b65fa6db08ce7feff2eb05443c04f69d80ef1c04d9f06a55ce833495f",
"bytes": 216978
},
"raw/chips_20260904.json.gz": {
"sha256": "61ef196f8decc891d3db368a72827109d8bfa0f65eb80585595fefd09b7f402c",
"bytes": 205013
},
"raw/chips_20260910.json.gz": {
"sha256": "6bc5d4664f98dc8345cbfd92c25644d6e7f611a8f47c396f95745a3663a92694",
"bytes": 194575
},
"raw/chips_20260911.json.gz": {
"sha256": "a1c3548f750be2d20b19385503cf2c7bdb78c7b008385d4b161c41cea5e687d5",
"bytes": 212678
},
"raw/chips_20260917.json.gz": {
"sha256": "42973caa57028be0f9f6c1eefb9bcde39b87e1576eae785f64afb0c3db0d4b8c",
"bytes": 212738
},
"raw/chips_20260918.json.gz": {
"sha256": "330a990fc9546c352ba92310e4cca999ac7bbc05b6879fd48e9d08554e7fdb8a",
"bytes": 223906
},
"raw/chips_20260924.json.gz": {
"sha256": "85b14e201314fb40be83c4b44ac7192670ea9943e500cb14c9dcd6ced9986bbf",
"bytes": 229693
},
"raw/chips_20260930.json.gz": {
"sha256": "54ba4456586b1480d3d3113e1d989c52737fe4d6e04012c01f49bc50d3dc8912",
"bytes": 228909
},
"raw/chips_20261009.json.gz": {
"sha256": "b27abec954a6e40e834da2b9ff75639cd4319c66f47b71a4c7be6f3c3ca11b77",
"bytes": 213165
},
"raw/chips_20261016.json.gz": {
"sha256": "3596f3ab440c9e83920cff496211d741e8c01de1f0f802e2ce3b58ae16fbd1b4",
"bytes": 169050
},
"raw/chips_20261023.json.gz": {
"sha256": "520f50a6e315c7065573ac34fc1d357a24ef3faefcdf3df78bb105c82f0f33c2",
"bytes": 214183
},
"raw/coverage.json.gz": {
"sha256": "60a52da90950c4fc96c7e1a7b5d0b49f2d4550290e948da7a3b56f25b451deda",
"bytes": 47168
},
"raw/daily.json.gz": {
"sha256": "9fe19c5b23260c3b785a6e54d628a85f9873e7482d72b374986fbd23b3869c45",
"bytes": 6805366
},
"raw/history_probe.json.gz": {
"sha256": "9e1a02217586cbfcaa993e3f996a9d66f5e4d7a51f10d165fd193b74f00aad7a",
"bytes": 3895
},
"raw/industry.json.gz": {
"sha256": "4a72321fe0cc184e39b29d542a82d52510a575ca11645819b1b8802cb1eeaba2",
"bytes": 1996
},
"raw/industry_history.json.gz": {
"sha256": "38a5a20eba0c1e31ae2da9be1b0858dc9018a6322409efd3ff2aea2fdfc9940a",
"bytes": 311731
},
"raw/phase_202509.json.gz": {
"sha256": "da9db1bfb3a8b06999c3f6739c7c7c41fd990492a4c73daf10c74bf7dab3fef7",
"bytes": 5231640
},
"raw/phase_202510.json.gz": {
"sha256": "3070512bf010d0ae33fad2c143d70ec7ba1fbe81d4a49a6c6c4dc1350fe95f6e",
"bytes": 10296680
},
"raw/phase_202511.json.gz": {
"sha256": "33e9b958c5f6c36dfae93d68154320568c38f59b2c2fee5297157910a3a35966",
"bytes": 12675684
},
"raw/phase_202512.json.gz": {
"sha256": "26142648da7c81dca4905d92ec9dd19bc109d87408bd92cb9b475b2e89e078f2",
"bytes": 15827525
},
"raw/phase_202601.json.gz": {
"sha256": "694302d33295e37a1436d4664f619f13c2b9a871aa2da9ec4789d6b112f6d91e",
"bytes": 14533249
},
"raw/phase_202602.json.gz": {
"sha256": "4fed71b43c4cd70b30ed101970f830709e4a9abdc3667e7d5fbef09deafa6b1c",
"bytes": 10266484
},
"raw/phase_202603.json.gz": {
"sha256": "15d3a59100289448ead0ea5291e1410b64a7db1129b2556af53834c929a9cce5",
"bytes": 16382557
},
"raw/phase_202604.json.gz": {
"sha256": "54beff86ede5be27f498990884f077406284c70442883db24f3a6115c49b2b91",
"bytes": 16288040
},
"raw/phase_202605.json.gz": {
"sha256": "f96d4f11bc01ad5e469be020dd83687cc418c86724b349e21ff84b62fd59a90d",
"bytes": 14160246
},
"raw/phase_202606.json.gz": {
"sha256": "d3d0b0a800f423657c2664403b73f089c1e7af2a78af19b4b859f414effa0dc3",
"bytes": 16872488
},
"raw/phase_202607.json.gz": {
"sha256": "081069758b6e92343f3e783c88321b1f8c559cde802ff887fcd6eac5d09ab25f",
"bytes": 19007492
},
"raw/phase_202608.json.gz": {
"sha256": "6bb7324dd6aa1a3ac2208e75c63caa5c95dddd1a7dc91ec7af03dd3a3a3a86a2",
"bytes": 17942453
},
"raw/phase_202609.json.gz": {
"sha256": "f7768fe75571f0eaf524076cafcd2220507d5d1149b735cfff74e77422d613a0",
"bytes": 18392081
},
"raw/phase_202610.json.gz": {
"sha256": "a4f638e53eb57be4fe06a3a666e999a35028bd7e15c385cdde2dcaa037354d3b",
"bytes": 14365827
},
"raw/probe.json.gz": {
"sha256": "8aa462c6fc76d1487fdee3ab6f05b07152a1b4c75885b16bfa5e987f2b365ef3",
"bytes": 3841
},
"raw/recent_prices.json.gz": {
"sha256": "58f9c166f362a42dc9324cc5eb104738911046f8ef213221dc7d88e12c6eb29a",
"bytes": 1922479
},
"raw/trap.json.gz": {
"sha256": "dc3faea89cf5ad65cf21b9c33dd499a05f98426d00891783ba8f83d56e0ec409",
"bytes": 11753091
}
},
"price_sources_relative_to_research": {
"macro_timing_20261026/raw/export.json": "a1b6678eb5aa0578b0a5f679304bfb65ef4a965a1260c6bed8d45bb672c0d67c",
"chip_accumulation_20261030/raw/recent_prices.json.gz": "58f9c166f362a42dc9324cc5eb104738911046f8ef213221dc7d88e12c6eb29a",
"macro_timing_20261026/raw/timed_micro100/202506.json.gz": "d468048839a210443494d5cef6bd5becb2ab20a2b6e92084fa86914d273d51b6",
"macro_timing_20261026/raw/timed_micro100/202507.json.gz": "2989587abecb15d9a6ac2478db581124945389b8b15e2f1b686560c5d4a4e402",
"macro_timing_20261026/raw/timed_micro100/202508.json.gz": "5a5487b283263ed4ac080b46b2d8d915c4eae357ce5ec24d6aac33c957d719f5",
"macro_timing_20261026/raw/timed_micro100/202509.json.gz": "385e49c8d7a9d2e4c5bceb2da2b888d6a7c881f1d4b70f91d83a17e21c2291ee",
"macro_timing_20261026/raw/timed_micro100/202510.json.gz": "1552c624ed590d4d1476aba37757922b76725d3921ad4add2f4c9240aa93a343",
"macro_timing_20261026/raw/timed_micro100/202511.json.gz": "405b743eae531569990ea047deb32540b46795a18e8f9ee00147448210190ec7",
"macro_timing_20261026/raw/timed_micro100/202512.json.gz": "691e978610282b446212a14238c938c74045e498e2d5d292aad3a3c2bcb65b42",
"macro_timing_20261026/raw/timed_micro100/202601.json.gz": "389c3a1dbd32bd59649cee6f4ac299a050516eb8a19074b4e7c15e0bf2982432",
"macro_timing_20261026/raw/timed_micro100/202602.json.gz": "ba45a8aa755a651f7afa4d5bb397de12f2a44d62b770defc74b8e79113adf4b4",
"macro_timing_20261026/raw/timed_micro100/202603.json.gz": "47ead146abbc2c5c594722c31b067dc24461dfbb948e1e2a7007ecc82900b6fd",
"macro_timing_20261026/raw/timed_micro100/202604.json.gz": "93da94833fc47d0dcb3a58624439ba270a4be9cdbed2e7b437dff8e77ddfb37f",
"macro_timing_20261026/raw/timed_micro100/202605.json.gz": "4a4824cc7130e044032bdee96fd84469825dfe1b95250aa5709f4435393a79ec",
"macro_timing_20261026/raw/timed_micro100/202606.json.gz": "b1d7909eafc803d62f10e64d8eb66f6b98833f2bec40f38d0ee24ed8a5c419be",
"macro_timing_20261026/raw/timed_micro100/202607.json.gz": "ccbab654f6dde42cd84fb341554949024da2a524db96cde0b636302e356c77f4",
"macro_timing_20261026/raw/timed_micro100/202608.json.gz": "3c78629a223fe9c30476a55767884db308bff902a9b35950dffca411ecd26a0f",
"macro_timing_20261026/raw/micro100/202609.json.gz": "933bc305bf6b8efde4df2ef137cd2216c49731a6ab8b6c7933b53cba4fbe91e4",
"macro_timing_20261026/raw/micro100/202610.json.gz": "0ef40efa23ce31262f02e0d4232d2a33559d5d34229ae824ee841cb030bee019",
"macro_timing_20261026/raw/timed_micro100/small_gap_candidates.json": "b2bda0470acd52bffbe691ae013b0897b2afb4718a5d4db6738b56e36dc98346",
"macro_timing_20261026/raw/timed_micro100/gap_legacy.json.gz": "93a4f719d4aecf73862837fb7713cd21bb0f075c35bb3649d38b9f8d483bbfe2",
"macro_timing_20261026/raw/micro100/reference.json.gz": "911e1ab9f4295eb67fdd00d53eb4ba5587104101065601f35d506c495471563d"
},
"artifacts": {
".gitignore": {
"sha256": "370f284f736a499f325e2b2560c103a74ec9d256005d13c8d20050f424fb738b",
"bytes": 25
},
"analysis_core.py": {
"sha256": "a28d51993112366bd44b23a24980b0ef4391a7ff27f531791fc9a1d64ff5e84f",
"bytes": 4872
},
"analyze.py": {
"sha256": "476423ec5bf59bc2255f20e271e87e6fa1b0480dedbbca07d6680a55b8748a76",
"bytes": 12192
},
"audit.json": {
"sha256": "2ecc75819199a011b89371ecff4a73e65f59efa0f01dafed76eb33adaf313a8d",
"bytes": 3417
},
"audit.py": {
"sha256": "b37973732d66ea520f09d0c494c125ab430ee9c19552ac693878964b0f468ebd",
"bytes": 6594
},
"chip_coverage.json": {
"sha256": "391cb053876ba86e50ed05a717cbfc5136f871184c480f1e55316916e5bc1060",
"bytes": 6988
},
"chip_features.py": {
"sha256": "9e713e7f384c128f162ea93ec287a4fd9c2771c1469f779f05c8f131492984d7",
"bytes": 2885
},
"common.py": {
"sha256": "a7b04a1b4fbfbfc2b5a121690a3dee97ffc630a87fb0e6afb78f4184b6f06f44",
"bytes": 1746
},
"data_diagnostics.json": {
"sha256": "d9716e361c11e4831c785639483bd0b6fb12b59dc680ea7b70a93feab8d9ea4e",
"bytes": 1636
},
"diagnose_data.py": {
"sha256": "3e542f3f3cece86a17b6f2418550fd56218a4fac1f8974327523fe2cb2981f0f",
"bytes": 3132
},
"feature_freeze.json": {
"sha256": "c48f279698a064171f67d3a5f216bfe77639248e571eae4daa47eb4545a0ed6e",
"bytes": 414
},
"fetch.py": {
"sha256": "ac3f5299c70a711f8c05fc616a21c66cbdf68bd920b80f04e295ad44c2adf26c",
"bytes": 2868
},
"figures/下跌中的吸筹次数.png": {
"sha256": "c00901b16e6721619a25f26be59d8a7b661fbe712b25a4df3d4781a3e7c79afa",
"bytes": 85281
},
"figures/两种断档示例.png": {
"sha256": "6473840cb28d82a1b5af9d2222d9b434567d89d04c06c92122641c52a715cdc5",
"bytes": 64229
},
"figures/匹配后的胜率差.png": {
"sha256": "621279a267eb36ee5bc4f9f716828eceec3d10278c9fa71819394ea0de480023",
"bytes": 90341
},
"figures/数据覆盖.png": {
"sha256": "9fa54d4075c19586b4a7bb621912be7d2172d9c1c5a6b5202f056d6c7bd80790",
"bytes": 55191
},
"finalize.py": {
"sha256": "c0c7e0bed96accf5b1901a649750f24ae0caf012f96d547d1a73ece076bf92b9",
"bytes": 1652
},
"group_results.csv": {
"sha256": "8f965d9aadb55831a3adabf890639db3fb0b4c99c2f8dc4e788965804e6cfe27",
"bytes": 9459
},
"implementation_notes.json": {
"sha256": "a1993225778c5714eb073a245bc1a7a118d7079c389263a04e534cd6f2acac0c",
"bytes": 966
},
"implementation_notes.sha256": {
"sha256": "27bada4353a589e3f63d74ac3d0212a74944608a8e1fb9b991880c400a80650a",
"bytes": 92
},
"matched_pairs.csv.gz": {
"sha256": "263f44f6d69665c6a02be6def69983d2beeb287314140068c26e03494bd57364",
"bytes": 6999
},
"phase_coverage.json": {
"sha256": "1bc73d4c09fa13024dc064f06358d7dfa52de0a1b408c141ab1761f72a1f27aa",
"bytes": 1949
},
"prepare_chips.py": {
"sha256": "05d347c982a377ebd8231309b718cac0c5f0e6ed5d9a596b4515ac6e30584b10",
"bytes": 5833
},
"prepare_phase.py": {
"sha256": "9d070c04330d26513859141fdbd73dc490d6e5a4c2cc780b96aff9e496b24486",
"bytes": 8384
},
"prepare_prices.py": {
"sha256": "364de6eee3f15b8edb2fa07ef477a6f94185bc08b0810d60aa1aa7c2adc2ec6b",
"bytes": 6345
},
"price_coverage.json": {
"sha256": "75c233c917a9123017c357d9dcba3a51f60d777ed4295d39478002a41367ebbe",
"bytes": 5593
},
"protocol.json": {
"sha256": "afd62f6ab1b0cf74182df468c3a98bf244f135906be348460b912cd99d2aedaf",
"bytes": 5487
},
"protocol.sha256": {
"sha256": "2e568442ec219daadca57e3d747fa29c47f30efc77ba36dd8c2780e7be84a224",
"bytes": 80
},
"remote.py": {
"sha256": "f553999807b8c7df56a08a9d0720a6cbea46cd329008c126bd4b326977b35785",
"bytes": 11149
},
"report.py": {
"sha256": "d80ba5ddd5f10e312d02583247ac59de6e6350978413646909034bf987ebfc6d",
"bytes": 24073
},
"requirements.txt": {
"sha256": "80fc930320a723026c0853dcdb15849a20300176e77cee80550a96280834b23e",
"bytes": 60
},
"results.json": {
"sha256": "239303288e0738d95995d6967027f5f465cf51f3254d847a6b9bb1444a8eae82",
"bytes": 19887
},
"test_research.py": {
"sha256": "d2f9561813123e8b3868ff6ce165199d6e576a66f4dd52211bdf7bbcb54854db",
"bytes": 3950
},
"trap_reconciliation.csv": {
"sha256": "24ff5dddbb038280bb60dee6c7cd05065ef3de387d036a804921fb6429932dd4",
"bytes": 230367
},
"validation.json": {
"sha256": "1205ec9ce5f82b439502eb46e971236014ec48d26e308947bd21da820a47c053",
"bytes": 1184
},
"筹码断档与吸筹次数首轮实验.md": {
"sha256": "2fb6d3ebbf9d7f1dc2530d5147fbb393638728022304c24437366e150bcb9b21",
"bytes": 15848
}
},
"exclusions": "原始数据和中间缓存留在开发机;manifest不递归计算自身摘要。",
"environment": "开发机离线分析。155独立只读容器导出,无业务目录挂载及交易凭证。"
}

View File

@ -0,0 +1,46 @@
{
"source_sha256": {
"phase_202509.json.gz": "da9db1bfb3a8b06999c3f6739c7c7c41fd990492a4c73daf10c74bf7dab3fef7",
"phase_202510.json.gz": "3070512bf010d0ae33fad2c143d70ec7ba1fbe81d4a49a6c6c4dc1350fe95f6e",
"phase_202511.json.gz": "33e9b958c5f6c36dfae93d68154320568c38f59b2c2fee5297157910a3a35966",
"phase_202512.json.gz": "26142648da7c81dca4905d92ec9dd19bc109d87408bd92cb9b475b2e89e078f2",
"phase_202601.json.gz": "694302d33295e37a1436d4664f619f13c2b9a871aa2da9ec4789d6b112f6d91e",
"phase_202602.json.gz": "4fed71b43c4cd70b30ed101970f830709e4a9abdc3667e7d5fbef09deafa6b1c",
"phase_202603.json.gz": "15d3a59100289448ead0ea5291e1410b64a7db1129b2556af53834c929a9cce5",
"phase_202604.json.gz": "54beff86ede5be27f498990884f077406284c70442883db24f3a6115c49b2b91",
"phase_202605.json.gz": "f96d4f11bc01ad5e469be020dd83687cc418c86724b349e21ff84b62fd59a90d",
"phase_202606.json.gz": "d3d0b0a800f423657c2664403b73f089c1e7af2a78af19b4b859f414effa0dc3",
"phase_202607.json.gz": "081069758b6e92343f3e783c88321b1f8c559cde802ff887fcd6eac5d09ab25f",
"phase_202608.json.gz": "6bb7324dd6aa1a3ac2208e75c63caa5c95dddd1a7dc91ec7af03dd3a3a3a86a2",
"phase_202609.json.gz": "f7768fe75571f0eaf524076cafcd2220507d5d1149b735cfff74e77422d613a0",
"phase_202610.json.gz": "a4f638e53eb57be4fe06a3a666e999a35028bd7e15c385cdde2dcaa037354d3b"
},
"batches": 266,
"rows": 1466056,
"dates": [
20250918,
20261029
],
"live_days": 14,
"replay_days": 252,
"unique_episodes": 5743,
"sample_audit": {
"scheduled": 16484,
"incomplete_phase_history": 854,
"ineligible": 1035
},
"samples_before_outcomes": 14595,
"sample_dates": 44,
"sample_stocks": 4993,
"with_chip_date": 13654,
"missing_industry": 0,
"industry_backfill_stock_samples": 302,
"count60_distribution": {
"0": 11313,
"1": 3097,
"2": 184,
"3": 1
},
"downtrend_samples": 4790,
"requests_days": 62
}

View File

@ -0,0 +1,83 @@
"""计算当日筹码特征、固定价格坐标下的变化,并独立核对现存压力表。"""
from collections import defaultdict
import numpy as np
import pandas as pd
from common import ROOT,read,dump,digest,dot,protocol,install_offline
from chip_features import chip_features
def main():
install_offline();protocol();a=dict(np.load(ROOT/'cache/prices.npz'))
di={d:i for i,d in enumerate(a['dates'])};ci={c:j for j,c in enumerate(a['codes'])}
samples=pd.read_csv(ROOT/'cache/samples_before_outcomes.csv.gz',keep_default_na=False)
records={};raw_dist={};audit=defaultdict(int);sources={};checks=[]
requests=read(ROOT/'cache/chip_requests.json')
files=sorted((ROOT/'raw').glob('chips_*.json.gz'))
assert {f.name[6:14] for f in files}==set(requests),'所需筹码导出日期未齐'
trap=read(ROOT/'raw/trap.json.gz')['data'];assert trap['ok']
trapmap={(r['trade_date'],dot(r['symbol'])):r for r in trap['rows']}
for f in files:
data=read(f)['data'];sources[f.name]=digest(f)
assert data['requested_symbols']==requests[f.name[6:14]],'导出股票清单不匹配'
dist=data['distribution'];summary=data['summary'];assert dist['ok'] and summary['ok']
groups=defaultdict(list)
for r in dist['rows']:groups[(r['trade_date'],dot(r['symbol']))].append(r)
sm={(r['trade_date'],dot(r['symbol'])):r for r in summary['rows']}
audit['raw_rows']+=len(dist['rows']);audit['requested_stock_dates']+=len(data['requested_symbols'])
for key,rows in groups.items():
d,c=key
if d not in di or c not in ci:continue
i,j=di[d],ci[c];close=a['raw_close'][i,j]
prices=np.array([r['price'] for r in rows]);weights=np.array([r['percent'] for r in rows],float)
feat=chip_features(prices,weights,close);feat.pop('valleys',None)
times=[r['create_time'] for r in rows if r.get('create_time')]
feat.update(chip_created_min=min(times) if times else '',chip_created_max=max(times) if times else '',
chip_rows=len(rows),date=d,code=c)
if key in sm:
s=sm[key];feat['summary_winner']=s['winner_rate'];feat['summary_cost']=s['weight_avg']
if feat['chip_valid'] and s['winner_rate'] is not None:
feat['winner_summary_gap_pp']=feat['winner_from_dist']*100-s['winner_rate']
if feat['chip_valid'] and s['weight_avg']:
feat['cost_summary_relative_error']=feat['weighted_cost']/s['weight_avg']-1
if feat['chip_valid']:
raw_dist[key]=(prices.astype(float),weights/weights.sum())
if key in trapmap:
t=trapmap[key]
checks.append({'date':d,'code':c,'close_difference':float(close-t['close']) if t['close'] else None,
'above_difference_pp':feat['overhead']*100-t['ocr_above'] if t['ocr_above'] is not None else None,
'far15_difference_pp':feat['overhead_far15']*100-t['ocr_115'] if t['ocr_115'] is not None else None,
'near15_difference_pp':feat['overhead_near15']*100-t['ocr_115'] if t['ocr_115'] is not None else None,
'raw_mass':feat['raw_mass']})
records[key]=feat;audit[feat['chip_reason']]+=1
print(f.name,'筹码股票日',len(groups),flush=True)
rows=[]
for _,r in samples.iterrows():
key=(r['date'],r['code']);f=records.get(key,{'chip_valid':False,'valley_evaluable':False,'chip_reason':'missing_distribution'})
x=dict(r);x.update(f);x['pressure_change20_fixed_price']=np.nan
prev=(r['previous20_date'],r['code'])
if key in raw_dist and prev in raw_dist:
i,j=int(r['i']),int(r['j']);ip=di[prev[0]];oldp,oldw=raw_dist[prev]
factor=a['adj'][ip,j]/a['adj'][i,j]
oldmass=oldw[oldp*factor>a['raw_close'][i,j]].sum()
x['pressure_change20_fixed_price']=float(f['overhead']-oldmass)
x['return_last20']=float(a['close'][i,j]/a['close'][ip,j]-1)
rows.append(x)
frame=pd.DataFrame(rows)
frame.to_csv(ROOT/'cache/features_before_outcomes.csv.gz',index=False,compression={'method':'gzip','mtime':0})
pd.DataFrame(records.values()).to_csv(ROOT/'cache/chip_features_all.csv.gz',index=False,compression={'method':'gzip','mtime':0})
pd.DataFrame(checks).to_csv(ROOT/'trap_reconciliation.csv',index=False)
valid=frame[frame.chip_valid]
cov={'source_sha256':sources,'audit':dict(audit),'samples':len(frame),'valid_chip_samples':len(valid),
'valley_evaluable_samples':int(frame.valley_evaluable.sum()),'upper_gap_samples':int(valid.upper_gap.sum()) if len(valid) else 0,
'above_valley_samples':int(valid.valley_above.sum()) if len(valid) else 0,
'chip_created_later_than_data_day':int((valid.chip_created_max.str[:10]>valid.date).sum()) if len(valid) else 0,
'fixed_price_change_pairs':int(frame.pressure_change20_fixed_price.notna().sum()),'trap_reconciliation_rows':len(checks)}
if len(checks):
q=pd.DataFrame(checks)
cov['trap_max_abs_difference_pp']={k:float(q[k].abs().max()) for k in ['above_difference_pp','far15_difference_pp','near15_difference_pp']}
if len(valid):
cov['winner_summary_abs_gap_pp_quantiles']={str(k):float(v) for k,v in valid.winner_summary_gap_pp.abs().quantile([.5,.9,.99]).items()}
cov['weighted_cost_abs_error_quantiles']={str(k):float(v) for k,v in valid.cost_summary_relative_error.abs().quantile([.5,.9,.99]).items()}
dump(ROOT/'chip_coverage.json',cov)
print('有效筹码样本',len(valid),'上方稀疏',cov['upper_gap_samples'],'上方双峰断档',cov['above_valley_samples'],flush=True)
if __name__=='__main__':main()

View File

@ -0,0 +1,116 @@
"""按冻结批次组装吸筹历史,再按预定日程生成与收益无关的观察样本。"""
from collections import defaultdict
import numpy as np
import pandas as pd
from common import ROOT,read,dump,digest,dot,prefix,protocol,install_offline,previous_sum,sample_slot
def main():
install_offline();p=protocol();a=dict(np.load(ROOT/'cache/prices.npz'))
dates=a['dates'];codes=a['codes'];di={d:i for i,d in enumerate(dates)};ci={c:j for j,c in enumerate(codes)}
shape=a['close'].shape
seen=np.zeros(shape,bool);valid=np.zeros(shape,bool)
events=np.full(shape,np.nan);clear=np.full(shape,np.nan);scores=np.full(shape,np.nan)
risk=np.zeros(shape,bool);idling=np.zeros(shape,bool);phase=np.full(shape,'',dtype='U20')
episode_seen={};sources={};batches=[];details_bad=0
coverage=read(ROOT/'raw/coverage.json.gz')['data']['batches']['rows']
catalog={r['batch_id']:r for r in coverage}
head=next(r for r in coverage if r['status']=='HEAD' and r['generation_id']==p['phase_generation'])
published={};parent=head['parent_batch_id'];expected_sequence=head['sequence_no']
while parent:
b=catalog[parent];assert b['status']=='COMPLETE' and b['sequence_no']==expected_sequence
assert b['generation_id']==p['phase_generation'] and b['data_date']<=20261029
published[parent]=b;parent=b['parent_batch_id'];expected_sequence-=1
assert expected_sequence==-1
for f in sorted((ROOT/'raw').glob('phase_*.json.gz')):
sources[f.name]=digest(f);data=read(f)['data'];assert data['ok']
for b in data['batches']:
assert b['batch_id'] in published
assert b['content_hash']==published[b['batch_id']]['content_hash']
batches.append(b)
for r in data['rows']:
d=str(r['data_date']);d=d[:4]+'-'+d[4:6]+'-'+d[6:];c=dot(r['stock_code'])
if d not in di or c not in ci:continue
i,j=di[d],ci[c];assert not seen[i,j],(d,c);seen[i,j]=True
ok=r['phase_quality']=='OK' and r['details'].get('bars_lag')==0
valid[i,j]=ok;phase[i,j]=r['phase'];risk[i,j]=r['phase']=='invalidated';idling[i,j]=bool(r['details'].get('idling'))
if ok:
events[i,j]=0;clear[i,j]=int('明确吸筹' in str(r['details'].get('state') or ''))
if r['phase_score'] is not None:scores[i,j]=r['phase_score']
ep=r['episode_id'];rec=r['details'].get('recognized_date')
if ep:
assert rec is not None and rec<=r['data_date']
key=(c,ep)
if key in episode_seen:assert episode_seen[key]['recognized_date']==rec
else:episode_seen[key]={'code':c,'episode_id':ep,'recognized_date':rec,'first_snapshot':r['data_date']}
print(f.name,'完成',flush=True)
assert set(b['batch_id'] for b in batches)==set(published),'导出未覆盖冻结代次的所有已发布批次'
for e in episode_seen.values():
ds=str(e['recognized_date']);d=ds[:4]+'-'+ds[4:6]+'-'+ds[6:]
if d not in di:continue
i,j=di[d],ci[e['code']]
assert e['first_snapshot']==e['recognized_date'],'事件被倒填或缺少识别日快照'
assert valid[i,j],'事件识别日质量异常'
events[i,j]+=1
c60=previous_sum(events,60);c20=previous_sum(events,20);d60=previous_sum(clear,60);d20=previous_sum(clear,20)
since=np.full(shape,np.nan);last=np.full(len(codes),-10000,dtype=int)
for i in range(len(dates)):
since[i]=np.where(last>=0,i-last,np.nan)
last[events[i]>0]=i
ind=read(ROOT/'raw/industry_history.json.gz')['data'];assert ind['ok']
industries=np.full(shape,'',dtype='U64');industry_created=np.full(shape,'',dtype='U32')
hist=defaultdict(list)
for r in ind['rows']:
c=dot(r['ts_code'])
if c in ci and r.get('industry'):hist[c].append(r)
for c,rows in hist.items():
rows=sorted(rows,key=lambda r:r['trade_date']);j=ci[c];k=0;chosen=None
for i,d in enumerate(dates):
while k<len(rows) and rows[k]['trade_date']<=d:chosen=rows[k];k+=1
if chosen:
industries[i,j]=chosen['industry'];industry_created[i,j]=str(chosen.get('created_at') or '')
week=[pd.Timestamp(str(d)).isocalendar()[:2] for d in dates]
weekly=[i for i in range(len(dates)-1) if week[i]!=week[i+1] and dates[i]>=p['sampling']['first_week']]
slots=np.array([sample_slot(c) for c in codes]);last_sample=np.full(len(codes),-10000)
records=[];audit=defaultdict(int)
chip_cov=read(ROOT/'raw/chip_dates.json.gz')['data']
chip_dates={r['trade_date'] for tab in chip_cov.values() if tab['ok'] for r in tab['rows']}
requests=defaultdict(list)
for w,i in enumerate(weekly):
js=np.flatnonzero(slots==w%14)
for j in js:
audit['scheduled']+=1
if i-last_sample[j]<60:audit['cooldown']+=1;continue
if not a['eligible'][i,j]:audit['ineligible']+=1;continue
if not valid[i,j] or not np.isfinite(c60[i,j]):audit['incomplete_phase_history']+=1;continue
feature=[a['ret60'][i,j],a['slope61'][i,j],a['volatility60'][i,j],a['turnover20'][i,j],a['position60'][i,j],a['market_cap'][i,j]]
if not np.isfinite(feature).all() or min(feature[2],feature[3],feature[5])<=0:audit['incomplete_price_history']+=1;continue
last_sample[j]=i;d=str(dates[i]);c=str(codes[j]);rec={'date':d,'code':c,'i':int(i),'j':int(j),'week_index':w,
'episodes60':int(c60[i,j]),'episodes20':int(c20[i,j]),'clear_days60':int(d60[i,j]),'clear_days20':int(d20[i,j]),
'recency':int(since[i,j]) if np.isfinite(since[i,j]) else None,'phase':str(phase[i,j]),'phase_score':float(scores[i,j]),
'invalidated':bool(risk[i,j]),'idling':bool(idling[i,j]),'ret60':float(feature[0]),'slope61':float(feature[1]),
'volatility60':float(feature[2]),'turnover20':float(feature[3]),'position60':float(feature[4]),'market_cap':float(feature[5]),
'downtrend':bool(feature[0]<=-.10 and feature[1]<0),'industry':str(industries[i,j]),
'industry_created_at':str(industry_created[i,j]),'raw_close':float(a['raw_close'][i,j]),'chip_date_available':d in chip_dates,
'previous20_date':str(dates[i-20]),'prior20_episodes':int(c20[i-20,j]) if np.isfinite(c20[i-20,j]) else None,
'prior20_clear_days':int(d20[i-20,j]) if np.isfinite(d20[i-20,j]) else None}
records.append(rec)
if d in chip_dates:
requests[d.replace('-','')].append(prefix(c))
prev=str(dates[i-20])
if prev in chip_dates:requests[prev.replace('-','')].append(prefix(c))
frame=pd.DataFrame(records);frame.to_csv(ROOT/'cache/samples_before_outcomes.csv.gz',index=False,compression={'method':'gzip','mtime':0})
dump(ROOT/'cache/chip_requests.json',{k:sorted(set(v)) for k,v in requests.items()})
np.savez_compressed(ROOT/'cache/phase.npz',seen=seen,valid=valid,events=events,clear=clear,count60=c60,count20=c20,
days60=d60,days20=d20,recency=since,score=scores,risk=risk,idling=idling,phase=phase)
pd.DataFrame(episode_seen.values()).to_csv(ROOT/'cache/episodes.csv.gz',index=False,compression={'method':'gzip','mtime':0})
dump(ROOT/'phase_coverage.json',{'source_sha256':sources,'batches':len(batches),'rows':sum(b['expected_rows'] for b in batches),
'dates':[min(b['data_date'] for b in batches),max(b['data_date'] for b in batches)],
'live_days':sum(b['source']=='live' for b in batches),'replay_days':sum(b['source']=='replay' for b in batches),
'unique_episodes':len(episode_seen),'sample_audit':dict(audit),'samples_before_outcomes':len(frame),
'sample_dates':frame.date.nunique(),'sample_stocks':frame.code.nunique(),'with_chip_date':int(frame.chip_date_available.sum()),
'missing_industry':int((frame.industry=='').sum()),'industry_backfill_stock_samples':int((frame.industry_created_at.str[:10]>frame.date).sum()),
'count60_distribution':{str(k):int(v) for k,v in frame.episodes60.value_counts().sort_index().items()},
'downtrend_samples':int(frame.downtrend.sum()),'requests_days':len(requests)})
print('预收益样本',len(frame),'筹码导出日期',len(requests),flush=True)
if __name__=='__main__':main()

View File

@ -0,0 +1,106 @@
"""复用已核验的全市场原始行情;不读取或筛选未来收益。"""
import bisect
import collections
import re
import numpy as np
import pandas as pd
from common import ROOT,OLD,read,dump,digest,dot,protocol,install_offline
def main():
install_offline();p=protocol();sources={}
def tracked(path):
sources[str(path.relative_to(ROOT.parent))]=digest(path);return read(path)
exp=tracked(OLD/'raw/export.json')['data']['index_daily']['000001.SH']['rows']
calendar=sorted({r['timestamp'][:10] for r in exp})
recent=tracked(ROOT/'raw/recent_prices.json.gz')['data']
recent_dates=sorted({r['trade_date'] for r in recent['bars']['rows']})
calendar=sorted(set(calendar)|set(recent_dates))
dates=[d for d in calendar if p['price_source_start']<=d<=p['data_cutoff']]
di={d:i for i,d in enumerate(dates)}
monthly=[]
for ym in sorted({d[:7].replace('-','') for d in dates}):
folder='timed_micro100' if ym<='202608' else 'micro100'
f=OLD/f'raw/{folder}/{ym}.json.gz'
monthly.append((f,tracked(f)['data']))
monthly.append((ROOT/'raw/recent_prices.json.gz',recent))
codes=sorted({r['ts_code'] for _,data in monthly for r in data['bars']['rows'] if r['ts_code'].endswith(('.SH','.SZ'))})
ci={c:j for j,c in enumerate(codes)};shape=(len(dates),len(codes))
a={k:np.full(shape,np.nan) for k in ['raw_open','raw_close','open','close','low','high','adj','volume','market_cap','turnover','up','down']}
risk=np.zeros(shape,bool);seen=set();repairs=[]
repairlist=tracked(OLD/'raw/timed_micro100/small_gap_candidates.json')
legacy=tracked(OLD/'raw/timed_micro100/gap_legacy.json.gz');assert legacy['ok']
lm={(r['trade_date'][:10],r['ts_code']):r for r in legacy['rows']}
wanted={(r['date'],r['code']) for r in repairlist}
for path,data in monthly:
assert all(data[k]['ok'] for k in ['bars','st','limits'])
for r in data['bars']['rows']:
d,c=r['trade_date'][:10],r['ts_code']
if d not in di or c not in ci:continue
i,j=di[d],ci[c]
assert (i,j) not in seen,(d,c,path)
seen.add((i,j))
for dest,src in [('raw_open','open'),('raw_close','close'),('volume','vol'),('market_cap','total_mv'),('turnover','turnover_rate'),('adj','adj_factor')]:
if r.get(src) is not None:a[dest][i,j]=r[src]
adj=a['adj'][i,j]
if adj>0:
for k in ['open','close','low','high']:
if r.get(k) and r[k]>0:a[k][i,j]=r[k]*adj
for r in data['limits']['rows']:
d,c=r['trade_date'][:10],r['ts_code']
if d not in di or c not in ci:continue
for k in ['up','down']:
if r.get(k+'_limit'):a[k][di[d],ci[c]]=r[k+'_limit']
for r in data['st']['rows']:
d,c=r['trade_date'][:10],r['ts_code']
if d in di and c in ci:risk[di[d],ci[c]]=True
for d,c in sorted(wanted):
if d not in di or c not in ci:continue
i,j=di[d],ci[c]
if (i,j) in seen:continue
r=lm[(d,c)];adj=r.get('adj_factor')
if adj and r.get('open') and r.get('close'):
for dest,src in [('raw_open','open'),('raw_close','close'),('volume','vol'),('market_cap','total_mv'),('turnover','turnover_rate'),('adj','adj_factor')]:
if r.get(src) is not None:a[dest][i,j]=r[src]
for k in ['open','close','low','high']:
if r.get(k):a[k][i,j]=r[k]*adj
repairs.append({'date':d,'code':c})
ref=tracked(OLD/'raw/micro100/reference.json.gz')['data']
listing={r['stock_code']:r['list_date'] for r in ref['listings']['rows']}
names=collections.defaultdict(list)
for r in ref['names']['rows']:names[r['ts_code']].append(r)
eligible=np.zeros(shape,bool);missing_listing=[];missing_name=0
for j,c in enumerate(codes):
ld=listing.get(c)
if not ld:missing_listing.append(c);continue
for i,d in enumerate(dates):
if bisect.bisect_right(calendar,d)-bisect.bisect_left(calendar,ld)<120:continue
nr=[r for r in names[c] if r['start_date']<=d and (not r['end_date'] or d<=r['end_date']) and (not r['ann_date'] or r['ann_date']<=d)]
if not nr:missing_name+=1;continue
name=max(nr,key=lambda r:r['start_date'])['name']
eligible[i,j]=not risk[i,j] and not re.search('ST|退',name,re.I) and a['close'][i,j]>0
tradable=(a['open']>0)&(a['volume']>0)&np.isfinite(a['up'])&np.isfinite(a['down'])
can_buy=tradable&(a['raw_open']<a['up']-.005)&(a['raw_open']>a['down']+.005)&~risk
can_sell=(a['open']>0)&(a['volume']>0)&np.isfinite(a['down'])&(a['raw_open']>a['down']+.005)
cl=pd.DataFrame(a['close'])
ret60=(cl/cl.shift(60)-1).to_numpy()
x=np.arange(61,dtype=float);x-=x.mean()
slope=np.full(shape,np.nan)
logs=np.log(a['close'])
for i in range(60,len(dates)):
w=logs[i-60:i+1];slope[i]=np.sum(w*x[:,None],axis=0)/np.sum(x*x)
volatility=np.log(cl/cl.shift()).rolling(60,min_periods=60).std().to_numpy()
turn20=pd.DataFrame(a['turnover']).rolling(20,min_periods=20).mean().to_numpy()
position=np.full(shape,np.nan)
for i in range(60,len(dates)):
w=a['close'][i-60:i]
valid=np.isfinite(w).all(axis=0)&np.isfinite(a['close'][i])
position[i,valid]=(w[:,valid]<a['close'][i,valid]).mean(axis=0)
np.savez_compressed(ROOT/'cache/prices.npz',**a,dates=np.array(dates),codes=np.array(codes),eligible=eligible,
can_buy=can_buy,can_sell=can_sell,ret60=ret60,slope61=slope,volatility60=volatility,turnover20=turn20,position60=position)
dump(ROOT/'price_coverage.json',{'source_sha256':sources,'dates':[dates[0],dates[-1]],'market_days':len(dates),'stocks':len(codes),
'eligible_stock_days':int(eligible.sum()),'missing_listing_codes':missing_listing,'missing_historical_name_stock_days':missing_name,
'legacy_repairs':repairs,'calendar_extension':'2026-10-26至29取全市场原始日线共同日期,后续与trade_cal交叉核验',
'survivorship_limit':'已退市证券和名称历史覆盖未完全认证;不能解释为全体历史A股无偏样本。'})
print('价格面板',shape,'合格股票日',int(eligible.sum()),flush=True)
if __name__=='__main__':main()

View File

@ -0,0 +1,172 @@
{
"source_sha256": {
"macro_timing_20261026/raw/export.json": "a1b6678eb5aa0578b0a5f679304bfb65ef4a965a1260c6bed8d45bb672c0d67c",
"chip_accumulation_20261030/raw/recent_prices.json.gz": "58f9c166f362a42dc9324cc5eb104738911046f8ef213221dc7d88e12c6eb29a",
"macro_timing_20261026/raw/timed_micro100/202506.json.gz": "d468048839a210443494d5cef6bd5becb2ab20a2b6e92084fa86914d273d51b6",
"macro_timing_20261026/raw/timed_micro100/202507.json.gz": "2989587abecb15d9a6ac2478db581124945389b8b15e2f1b686560c5d4a4e402",
"macro_timing_20261026/raw/timed_micro100/202508.json.gz": "5a5487b283263ed4ac080b46b2d8d915c4eae357ce5ec24d6aac33c957d719f5",
"macro_timing_20261026/raw/timed_micro100/202509.json.gz": "385e49c8d7a9d2e4c5bceb2da2b888d6a7c881f1d4b70f91d83a17e21c2291ee",
"macro_timing_20261026/raw/timed_micro100/202510.json.gz": "1552c624ed590d4d1476aba37757922b76725d3921ad4add2f4c9240aa93a343",
"macro_timing_20261026/raw/timed_micro100/202511.json.gz": "405b743eae531569990ea047deb32540b46795a18e8f9ee00147448210190ec7",
"macro_timing_20261026/raw/timed_micro100/202512.json.gz": "691e978610282b446212a14238c938c74045e498e2d5d292aad3a3c2bcb65b42",
"macro_timing_20261026/raw/timed_micro100/202601.json.gz": "389c3a1dbd32bd59649cee6f4ac299a050516eb8a19074b4e7c15e0bf2982432",
"macro_timing_20261026/raw/timed_micro100/202602.json.gz": "ba45a8aa755a651f7afa4d5bb397de12f2a44d62b770defc74b8e79113adf4b4",
"macro_timing_20261026/raw/timed_micro100/202603.json.gz": "47ead146abbc2c5c594722c31b067dc24461dfbb948e1e2a7007ecc82900b6fd",
"macro_timing_20261026/raw/timed_micro100/202604.json.gz": "93da94833fc47d0dcb3a58624439ba270a4be9cdbed2e7b437dff8e77ddfb37f",
"macro_timing_20261026/raw/timed_micro100/202605.json.gz": "4a4824cc7130e044032bdee96fd84469825dfe1b95250aa5709f4435393a79ec",
"macro_timing_20261026/raw/timed_micro100/202606.json.gz": "b1d7909eafc803d62f10e64d8eb66f6b98833f2bec40f38d0ee24ed8a5c419be",
"macro_timing_20261026/raw/timed_micro100/202607.json.gz": "ccbab654f6dde42cd84fb341554949024da2a524db96cde0b636302e356c77f4",
"macro_timing_20261026/raw/timed_micro100/202608.json.gz": "3c78629a223fe9c30476a55767884db308bff902a9b35950dffca411ecd26a0f",
"macro_timing_20261026/raw/micro100/202609.json.gz": "933bc305bf6b8efde4df2ef137cd2216c49731a6ab8b6c7933b53cba4fbe91e4",
"macro_timing_20261026/raw/micro100/202610.json.gz": "0ef40efa23ce31262f02e0d4232d2a33559d5d34229ae824ee841cb030bee019",
"macro_timing_20261026/raw/timed_micro100/small_gap_candidates.json": "b2bda0470acd52bffbe691ae013b0897b2afb4718a5d4db6738b56e36dc98346",
"macro_timing_20261026/raw/timed_micro100/gap_legacy.json.gz": "93a4f719d4aecf73862837fb7713cd21bb0f075c35bb3649d38b9f8d483bbfe2",
"macro_timing_20261026/raw/micro100/reference.json.gz": "911e1ab9f4295eb67fdd00d53eb4ba5587104101065601f35d506c495471563d"
},
"dates": [
"2025-06-03",
"2026-10-29"
],
"market_days": 343,
"stocks": 5249,
"eligible_stock_days": 1689379,
"missing_listing_codes": [
"001220.SZ",
"001232.SZ",
"001233.SZ",
"001237.SZ",
"001246.SZ",
"001248.SZ",
"001257.SZ",
"001280.SZ",
"001312.SZ",
"001325.SZ",
"001365.SZ",
"001369.SZ",
"001393.SZ",
"001396.SZ",
"001399.SZ",
"301449.SZ",
"301513.SZ",
"301531.SZ",
"301569.SZ",
"301583.SZ",
"301599.SZ",
"301638.SZ",
"301655.SZ",
"301660.SZ",
"301666.SZ",
"301667.SZ",
"301669.SZ",
"301677.SZ",
"301680.SZ",
"301682.SZ",
"301683.SZ",
"301686.SZ",
"301687.SZ",
"301688.SZ",
"301689.SZ",
"301696.SZ",
"301697.SZ",
"301699.SZ",
"301707.SZ",
"301716.SZ",
"301717.SZ",
"301718.SZ",
"601091.SH",
"601112.SH",
"601123.SH",
"603092.SH",
"603248.SH",
"603284.SH",
"603293.SH",
"603352.SH",
"603376.SH",
"603402.SH",
"603407.SH",
"603435.SH",
"603448.SH",
"603459.SH",
"603468.SH",
"688635.SH",
"688712.SH",
"688727.SH",
"688781.SH",
"688785.SH",
"688790.SH",
"688795.SH",
"688796.SH",
"688797.SH",
"688801.SH",
"688802.SH",
"688805.SH",
"688806.SH",
"688807.SH",
"688808.SH",
"688809.SH",
"688811.SH",
"688813.SH",
"688816.SH",
"688818.SH",
"688820.SH",
"688825.SH",
"688826.SH",
"688828.SH",
"688835.SH",
"688836.SH",
"688837.SH"
],
"missing_historical_name_stock_days": 3087,
"legacy_repairs": [
{
"date": "2025-11-18",
"code": "002856.SZ"
},
{
"date": "2025-11-18",
"code": "002883.SZ"
},
{
"date": "2025-11-18",
"code": "002910.SZ"
},
{
"date": "2026-03-02",
"code": "002316.SZ"
},
{
"date": "2026-03-02",
"code": "002360.SZ"
},
{
"date": "2026-03-02",
"code": "600561.SH"
},
{
"date": "2026-03-02",
"code": "600697.SH"
},
{
"date": "2026-06-26",
"code": "300417.SZ"
},
{
"date": "2026-06-26",
"code": "300500.SZ"
},
{
"date": "2026-06-26",
"code": "300535.SZ"
},
{
"date": "2026-06-26",
"code": "300549.SZ"
},
{
"date": "2026-06-26",
"code": "300605.SZ"
}
],
"calendar_extension": "2026-10-26至29取全市场原始日线共同日期,后续与trade_cal交叉核验",
"survivorship_limit": "已退市证券和名称历史覆盖未完全认证;不能解释为全体历史A股无偏样本。"
}

View File

@ -0,0 +1,143 @@
{
"version": "1.0",
"frozen_at": "2026-10-30T09:12:35.387801+08:00",
"purpose": "两类筹码断档、套牢压力及下跌中吸筹次数对未来60日胜率的增量预测价值",
"data_cutoff": "2026-10-29",
"price_source_start": "2025-06-01",
"phase_start": "2025-09-18",
"phase_generation": "phase-v1-20261012",
"research_nature": "现存历史分布与吸筹阶段回放。首次可用时间未经完整认证;不是当年真实可交易信号或独立样本外验证。",
"universe": "当时沪深普通A股,上市满120市场交易日,有历史名称且非ST非退市整理;不使用今日存续名单。北交所单列不纳入。",
"sampling": {
"frequency": "市场每周最后一个交易日",
"assignment": "SHA256(点式股票代码)前8位整数取模14,对应顺序周序号模14",
"minimum_market_days_between_samples": 60,
"first_week": "2025-12-15",
"eligibility_before_outcomes": true,
"missing_data": "不得视为零,不因未来数据缺失取消既有入样"
},
"history_windows": [
60,
20
],
"exclude_current_day_from_counts": true,
"accumulation": {
"episodes": "已有episode_id首次recognized_date;同事件去重,包含未结束事件;phase_quality连续完整OK才形成完整计数",
"days": "details.state含明确吸筹的有效历史日数;不混用phase_score和旧score",
"groups": [
0,
1,
"2+"
],
"high_count": 2,
"recency": "距最近已识别事件的市场交易日数;窗口外单列",
"risk_tags": "当日invalidated与idling分别保留;风险排除仅作预定敏感性"
},
"downtrend": {
"return60_max": -0.1,
"log_price_slope61_max": 0,
"price_position_control": true
},
"chip": {
"valid_mass_range_percent": [
98,
102
],
"upper_interval": [
1.02,
1.12
],
"upper_mass_max": 0.05,
"relative_bins_width": 0.01,
"peak_search_range": [
0.7,
1.3
],
"smoothing": "1:2:1对称三点仅用于峰检测",
"minimum_peak_neighborhood_mass": 0.05,
"peak_neighborhood_bins": 5,
"valley_relative_density_max": 0.25,
"minimum_valley_width": 0.05,
"valley_selection": "每种位置取距现价最近的合格低密度段;主检验只取下沿>=1.02的上方段",
"positions": [
"above",
"crossing",
"below"
],
"resolution_rule": "价格档间隔中位数大于现价2%时双峰断档记不可判;不把过粗离散档之间的零值当断档",
"pressure": "以完整筹码分布计算现价上方比例和上方15%内比例;与gp_stock_trap_pressure的现存值对拍,不冒充历史tsp"
},
"execution": {
"entry": "信号次交易日开盘的历史假设;当日无有效价量或触及涨跌停不成交、不顺延追买",
"exit": "入场索引+60开盘,跌停或停牌顺延到下一可卖开盘;截止仍未卖出单列",
"holding_days": 60,
"cost_per_side": [
0.003,
0.006
],
"path": "入场日到入场后第59日收盘;+15%拉升、先+15%还是先-10%、最大收盘浮亏;不以未来最高价卖出",
"not_simulated": [
"整手",
"最低佣金",
"市场容量",
"实际盘中撮合"
]
},
"matching": {
"exact": [
"signal_date",
"historical_industry"
],
"ratio": 1,
"replacement": false,
"algorithm": "按股票代码排序的处理组,逐个最小标准化欧氏距离,平局取代码较小者",
"calipers": {
"return60": 0.05,
"log_market_cap": 0.69314718056,
"log_volatility60": 0.69314718056,
"log_turnover20": 0.69314718056,
"price_position60": 0.2,
"overhead_mass_for_chip_tests": 0.1
},
"no_match": "保留计数但不纳入配对效果,不按收益放宽门槛",
"industry_unavailable": "主配对不运行;不拿当前行业替代历史行业"
},
"primary_tests": [
"上方稀疏相对无稀疏",
"上方双峰断档相对无此断档",
"下跌组中吸筹2次及以上相对0或1次"
],
"primary_outcome": "每边0.3%成本后60日持有策略盈利率及相对匹配对照的百分点差",
"secondary": [
"期内上涨15%比例",
"先上涨15%而非先下跌10%比例",
"平均与中位净收益",
"最大收盘浮亏",
"20日计数敏感性",
"每边0.6%成本敏感性",
"结构完好子组",
"吸筹天数独立信息",
"两类断档重合",
"套牢压力相关性"
],
"statistics": {
"business_min_lift": 0.05,
"bootstrap_repetitions": 1000,
"seed": 20261030,
"dependence": "60市场交易日的信号时间块整块重抽,另做股票簇重抽;分别列区间而不把样本行视为独立",
"minimum_time_blocks_for_formal_inference": 8,
"multiple_testing": "正式三个主检验用Holm控制5%;时间块不足不出正式p值、不宣称显著",
"validation": "现有全部历史仅探索;不因时间短强切极小留出集,规则冻结后新满60日样本才为前瞻验证"
},
"current_observed_limits": {
"trap_days": 26,
"daily_accumulation_days": 33,
"phase_days": 266,
"chip_2025_boundaries": [
"2025-11-10",
"2025-11-20"
],
"live_60_day_matured": 0
},
"failure_policy": "缺表、质量不合格、没有完整成熟样本时报告证据不足;不合成真实筹码、不回写业务表、不上线。"
}

View File

@ -0,0 +1 @@
afd62f6ab1b0cf74182df468c3a98bf244f135906be348460b912cd99d2aedaf protocol.json

View File

@ -0,0 +1,166 @@
"""stdin只读任务。允许表清单和单表查询,禁止业务写入和交易连接。"""
import datetime as dt
import decimal
import gzip
import hashlib
import json
import re
import signal
import sys
import time
import pymysql
from sqlalchemy.engine import make_url
from config.settings import settings
from app.db.session import assert_single_table
SECRET_KEYS = ('PROXY_DB_URL', 'DB_MYSQL_URL')
TABLES = {'gp_stock_trap_pressure','gp_chip_data','accumulation_daily',
'accumulation_phase_daily','accumulation_phase_batch',
'gp_chip_distribution_2025','gp_chip_distribution_2026',
'gp_category_industry','ths_industry_member','gp_stock_basic',
'gp_stock_category','gp_stock_factor_pro_202610','gp_stock_st',
'gp_stk_limit','gp_stock_namechange','gp_day_basic','trade_cal','gp_trade_cal'}
CONNECTIONS = {}
signal.alarm(550)
def encode(x):
if isinstance(x,(dt.date,dt.datetime)): return x.isoformat()
if isinstance(x,decimal.Decimal): return float(x)
raise TypeError(type(x).__name__)
def connection(source):
if source not in CONNECTIONS:
u = make_url(settings.PROXY_DB_URL if source=='proxy' else settings.DB_MYSQL_URL)
if u.host not in ('192.168.16.153','192.168.16.150'):
raise ValueError('数据源主机超出153代理及150只读源范围')
CONNECTIONS[source] = pymysql.connect(host=u.host,port=u.port or 3306,
user=u.username,password=u.password,database=u.database,charset='utf8mb4',
connect_timeout=8,read_timeout=90,write_timeout=8,autocommit=True,
cursorclass=pymysql.cursors.DictCursor)
return CONNECTIONS[source]
def read(source,sql,params=None,cap=10000):
if not re.match(r'^(SELECT|SHOW)\s',sql,re.I) or ';' in sql:
raise ValueError('仅允许单条SELECT或SHOW')
if re.search(r'\b(INTO|OUTFILE|DUMPFILE|FOR\s+UPDATE|LOCK|SLEEP|BENCHMARK|JOIN)\b',sql,re.I):
raise ValueError('禁止写入、锁定、联表及副作用')
assert_single_table(sql)
found = re.findall(r'\bFROM\s+`?([A-Za-z_0-9]+)',sql,re.I)
if found and (len(found)!=1 or found[0] not in TABLES):
raise ValueError('表超出白名单')
begin=time.monotonic()
try:
with connection(source).cursor() as c:
c.execute(sql,params); rows=c.fetchmany(cap+1)
if len(rows)>cap: raise ValueError('读取超限')
return {'ok':True,'rows':rows,'seconds':round(time.monotonic()-begin,3)}
except Exception as e:
args=getattr(e,'args',())
return {'ok':False,'error':{'type':type(e).__name__,
'code':args[0] if args and isinstance(args[0],int) else None},
'seconds':round(time.monotonic()-begin,3)}
def main():
mode=sys.argv[1]
data={}
if mode=='probe':
for source in ['proxy','index']:
data[source]={}
catalog=read(source,'SHOW TABLES',cap=5000)
if catalog['ok']:
data[source]['catalog']=[str(next(iter(r.values()))) for r in catalog['rows']
if re.search('trap|chip|cyq|accum|industry',str(next(iter(r.values()))),re.I)]
else: data[source]['catalog']=catalog
for table in sorted(TABLES):
cols=read(source,'SHOW COLUMNS FROM '+table)
data[source][table]={'columns':cols}
if cols['ok']:
data[source][table]['indexes']=read(source,'SHOW INDEX FROM '+table)
data[source][table]['sample']=read(source,'SELECT * FROM '+table+' LIMIT 1',cap=1)
elif mode=='coverage':
for table,datecol in [('gp_stock_trap_pressure','trade_date'),('accumulation_daily','data_date')]:
data[table]=read('proxy',f'SELECT {datecol},COUNT(*) AS n FROM {table} GROUP BY {datecol}',cap=3000)
data['batches']=read('proxy','SELECT * FROM accumulation_phase_batch',cap=3000)
for table in ['gp_chip_distribution_2025','gp_chip_distribution_2026','gp_category_industry','ths_industry_member']:
data[table]={'columns':read('proxy','SHOW COLUMNS FROM '+table),
'indexes':read('proxy','SHOW INDEX FROM '+table),
'sample':read('proxy','SELECT * FROM '+table+' LIMIT 1',cap=1)}
elif mode=='history_probe':
for y in ['2025','2026']:
table='gp_chip_distribution_'+y
data[table]={}
for order in ['ASC','DESC']:
data[table][order]=read('proxy',f'SELECT trade_date FROM {table} ORDER BY trade_date {order} LIMIT 1',cap=1)
for day in (['2025-09-19','2025-12-19'] if y=='2025' else ['2026-03-20','2026-06-19','2026-10-23']):
data[table][day]=read('proxy',f'SELECT COUNT(*) AS n,COUNT(DISTINCT symbol) AS stocks,MIN(create_time) AS first_created,MAX(create_time) AS last_created FROM {table} WHERE trade_date=%s',(day,),cap=1)
for table in ['gp_stock_basic','gp_stock_category','gp_stock_factor_pro_202610','gp_day_basic']:
data[table]={'columns':read('proxy','SHOW COLUMNS FROM '+table),
'sample':read('proxy','SELECT * FROM '+table+' LIMIT 1',cap=1)}
elif mode=='chip_dates':
for y in ['2025','2026']:
table='gp_chip_distribution_'+y
data[table]=read('proxy',f'SELECT trade_date,COUNT(*) AS n FROM {table} GROUP BY trade_date',cap=500)
elif mode=='recent_prices':
data={
'bars':read('proxy','SELECT ts_code,trade_date,open,high,low,close,pre_close,pct_chg,vol,amount,total_mv,circ_mv,turnover_rate,adj_factor FROM gp_stock_factor_pro_202610 WHERE trade_date>=%s AND trade_date<=%s',('2026-10-26','2026-10-29'),cap=30000),
'st':read('proxy','SELECT ts_code,trade_date,name,type,type_name FROM gp_stock_st WHERE trade_date>=%s AND trade_date<=%s',('2026-10-26','2026-10-29'),cap=10000),
'limits':read('proxy','SELECT ts_code,trade_date,up_limit,down_limit FROM gp_stk_limit WHERE trade_date>=%s AND trade_date<=%s',('2026-10-26','2026-10-29'),cap=40000)}
if not all(x['ok'] for x in data.values()): raise RuntimeError('补充行情读取失败')
elif mode=='industry':
data=read('proxy','SELECT ts_code,symbol,trade_date,industry,is_st,created_at,updated_at FROM gp_stock_category WHERE trade_date>=%s AND trade_date<=%s',('2025-09-01','2026-10-29'),cap=2000000)
if not data['ok']: raise RuntimeError('行业读取失败')
elif mode=='industry_history':
data=read('proxy','SELECT ts_code,symbol,trade_date,industry,is_st,created_at,updated_at FROM gp_stock_category WHERE trade_date<=%s',('2026-10-29',),cap=100000)
if not data['ok']: raise RuntimeError('历史行业读取失败')
elif mode=='calendar_probe':
for table in ['trade_cal','gp_trade_cal']:
data[table]={'columns':read('proxy','SHOW COLUMNS FROM '+table),
'sample':read('proxy','SELECT * FROM '+table+' LIMIT 3',cap=3)}
elif mode=='calendar':
data=read('proxy','SELECT exchange,cal_date,is_open FROM trade_cal WHERE exchange=%s AND cal_date>=%s AND cal_date<=%s',('SSE','2025-06-01','2026-11-06'),cap=1000)
if not data['ok']: raise RuntimeError('交易日历读取失败')
elif re.fullmatch(r'chips_20\d{6}',mode):
ds=mode.split('_')[1];day=ds[:4]+'-'+ds[4:6]+'-'+ds[6:]
if ds[:4] not in ('2025','2026') or day>'2026-10-29': raise ValueError('筹码日期越界')
codes=REQUEST
if not codes or len(codes)>1000 or not all(re.fullmatch(r'(SH|SZ)\d{6}',c) for c in codes): raise ValueError('股票清单越界')
marks=','.join(['%s']*len(codes));params=[day]+codes
data={}
data['distribution']=read('proxy','SELECT symbol,ts_code,trade_date,price,percent,create_time FROM gp_chip_distribution_'+ds[:4]+' WHERE trade_date=%s AND symbol IN ('+marks+')',params,cap=200000)
data['summary']=read('proxy','SELECT symbol,ts_code,trade_date,weight_avg,winner_rate,cost_5pct,cost_50pct,cost_95pct,create_time FROM gp_chip_data WHERE trade_date=%s AND symbol IN ('+marks+')',params,cap=1200)
data['requested_symbols']=codes
if not all(data[k]['ok'] for k in ('distribution','summary')): raise RuntimeError('筹码导出失败')
elif mode=='trap':
data=read('proxy','SELECT * FROM gp_stock_trap_pressure WHERE trade_date<=%s',('2026-10-29',),cap=200000)
if not data['ok']: raise RuntimeError('套牢盘导出失败')
elif mode=='daily':
data=read('proxy','SELECT * FROM accumulation_daily WHERE data_date<=%s',(20261029,),cap=250000)
if not data['ok']: raise RuntimeError('吸筹日分导出失败')
elif re.fullmatch(r'phase_20\d{4}',mode):
ym=mode.split('_')[1]
batches=read('proxy','SELECT * FROM accumulation_phase_batch WHERE generation_id=%s',('phase-v1-20261012',),cap=1000)
if not batches['ok']: raise RuntimeError('批次读取失败')
chosen=[x for x in batches['rows'] if x['status']=='COMPLETE' and str(x['data_date']).startswith(ym) and x['data_date']<=20261029]
rows=[]
for batch in sorted(chosen,key=lambda x:x['data_date']):
got=read('proxy','SELECT batch_id,stock_code,data_date,phase,phase_quality,episode_id,end_confirmed_date,phase_score,score,structure,is_st,is_delisting,row_hash,payload FROM accumulation_phase_daily WHERE batch_id=%s',(batch['batch_id'],),cap=6500)
if not got['ok'] or len(got['rows'])!=batch['expected_rows']: raise RuntimeError('批次不完整')
actual=sorted((r['stock_code'],hashlib.sha256(r['payload'].encode()).hexdigest()) for r in got['rows'])
if any(hashlib.sha256(r['payload'].encode()).hexdigest()!=r['row_hash'] for r in got['rows']): raise RuntimeError('逐行摘要不符')
content_hash=hashlib.sha256(json.dumps(actual,ensure_ascii=False,sort_keys=True,separators=(',',':'),allow_nan=False).encode()).hexdigest()
if content_hash!=batch['content_hash']: raise RuntimeError('批次摘要不符')
for row in got['rows']:
payload=json.loads(row.pop('payload')); item=payload['item']
row['details']={k:item.get(k) for k in ['recognized_date','state','pos_pct','phase_pos_pct','idling','bars_lag','bars_used','mf_rows','last_bar_date','close_basis','subscores','acc_start_date','acc_end_date','reference_low_px','reference_high_px','terminal_date','left_truncated','data_issue']}
replaced=payload.get('replaced_episode')
row['replaced_episode']=replaced
rows.extend(got['rows'])
data={'ok':True,'batches':chosen,'rows':rows}
else:
raise ValueError('未定义模式')
out={'collected_at_beijing':dt.datetime.now(dt.timezone(dt.timedelta(hours=8))).isoformat(),
'host':'155','mode':mode,'read_only':True,'data':data}
sys.stdout.buffer.write(gzip.compress(json.dumps(out,ensure_ascii=False,default=encode,separators=(',',':')).encode(),compresslevel=1))
for c in CONNECTIONS.values(): c.close()
if __name__=='__main__': main()

View File

@ -0,0 +1,254 @@
"""生成中文研究报告和可导出的静态研究图。"""
import datetime as dt
from pathlib import Path
import numpy as np
import pandas as pd
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from common import ROOT,read,dump,digest,protocol,install_offline,dot
def pct(x,sign=False):
if x is None or pd.isna(x):return '样本不足'
return f'{x*100:+.2f}%' if sign else f'{x*100:.2f}%'
def points(x):return '样本不足' if x is None or pd.isna(x) else f'{x*100:+.2f}个百分点'
def table(headers,rows):
return '\n'.join(['| '+' | '.join(headers)+' |','| '+' | '.join(['---']*len(headers))+' |']+
['| '+' | '.join(str(v) for v in r)+' |' for r in rows])
def figures(result,groups,features):
out=ROOT/'figures';out.mkdir(exist_ok=True)
plt.rcParams.update({'font.family':'Arial Unicode MS','axes.unicode_minus':False,'font.size':11})
keys=['upper_gap','valley_above','frequency_down'];labels=['现价上方筹码稀疏','现价上方双峰断档','下跌中吸筹至少2次']
fig,ax=plt.subplots(figsize=(10,4.8))
for y,(key,label) in enumerate(zip(keys,labels)):
r=result['tests'][key];u=r.get('uncertainty',{})
for offset,field,color in [(-.10,'time_interval','#167d9a'),(.10,'stock_interval','#df8a2e')]:
interval=u.get(field)
if interval:
lo,hi=np.array(interval)*100;ax.hlines(y+offset,lo,hi,color=color,lw=4)
if r['win_lift'] is not None:
ax.scatter(r['win_lift']*100,y,s=55,color='#253b53',zorder=3)
if u.get('degenerate_observed_differences'):ax.annotate('重抽退化,区间不可估',xy=(r['win_lift']*100,y),xytext=(10,12),textcoords='offset points',fontsize=9)
else:ax.text(0,y,'无可用配对',va='center',ha='left')
ax.axvline(0,color='#777',lw=1,ls='--')
ax.set_yticks(range(3),[labels[i]+f"\n完成配对 {result['tests'][k]['completed_pairs']} 组" for i,k in enumerate(keys)])
ax.invert_yaxis();ax.set_ylim(2.55,-.55);ax.set_xlabel('60日净盈利率相对匹配对照的差异(百分点)')
ax.set_title('三项主检验:点估计与描述性95%区间',loc='left',fontsize=15,pad=16)
from matplotlib.lines import Line2D
ax.legend(handles=[Line2D([0],[0],color='#167d9a',lw=4,label='按60交易日时间块重抽'),Line2D([0],[0],color='#df8a2e',lw=4,label='按股票重抽')],loc='upper center',bbox_to_anchor=(.5,-.18),ncol=2,frameon=False)
ax.spines[['right','top']].set_visible(False);ax.grid(axis='x',alpha=.2)
fig.text(.01,.01,'历史回放;时间块不足,不作显著性或未来胜率保证。费用每边0.3%。',fontsize=10,color='#666')
fig.tight_layout(rect=(0,.08,1,1));fig.savefig(out/'匹配后的胜率差.png',dpi=180,bbox_inches='tight');plt.close(fig)
fig,axes=plt.subplots(1,2,figsize=(11,4.5),sharey=True)
for ax,name,title in zip(axes,['frequency_down','frequency_down_intact'],['全部下跌样本','当日结构未失效且非空耗']):
g=groups[groups.analysis==name].set_index('group')
for x,k in enumerate(['0','1','2+']):
r=g.loc[k];n=int(r.completed)
if n:
ax.bar(x,r.win_rate*100,color=['#9caab7','#2b8cbe','#045a8d'][x],width=.58)
ax.text(x,r.win_rate*100+2,f'{r.win_rate*100:.1f}%\n{n}笔',ha='center',va='bottom',fontsize=10)
else:ax.text(x,5,'无完成样本',ha='center',fontsize=10)
ax.set_xticks([0,1,2],['0次','1次','至少2次']);ax.set_ylim(0,110);ax.set_title(title);ax.set_xlabel('此前60日独立吸筹次数')
ax.spines[['top','right']].set_visible(False)
axes[0].set_ylabel('完成交易的60日净盈利率(%)')
fig.suptitle('原始分组只展示关系,需结合匹配结果判断',fontsize=15)
fig.tight_layout();fig.savefig(out/'下跌中的吸筹次数.png',dpi=180,bbox_inches='tight');plt.close(fig)
cov=read(ROOT/'raw/coverage.json.gz')['data'];chip=read(ROOT/'raw/chip_dates.json.gz')['data']
sets=[('筹码价格分档',[r['trade_date'] for x in chip.values() for r in x['rows']]),
('吸筹阶段快照',[str(r['data_date']) for r in cov['batches']['rows'] if r['status']=='COMPLETE']),
('吸筹每日保存',[str(r['data_date']) for r in cov['accumulation_daily']['rows']]),
('套牢盘压力表',[r['trade_date'] for r in cov['gp_stock_trap_pressure']['rows']])]
fig,ax=plt.subplots(figsize=(11,3.6))
for y,(label,dates) in enumerate(sets):
d=pd.to_datetime(dates,format='mixed');ax.vlines(d,y-.18,y+.18,lw=1,color=['#087e8b','#337ab7','#87a4bd','#d48736'][y])
ax.axvline(pd.Timestamp(result['coverage']['latest_mature_signal']),ls='--',color='#666')
ax.set_yticks(range(4),[x[0] for x in sets]);ax.invert_yaxis();ax.xaxis.set_major_locator(mdates.MonthLocator(interval=2));ax.xaxis.set_major_formatter(mdates.DateFormatter('%Y-%m'))
ax.set_title('数据覆盖决定能回答的问题',loc='left',fontsize=15);ax.set_xlabel('数据日期;虚线为本轮最后一个成熟观察信号日')
ax.spines[['top','right']].set_visible(False);fig.tight_layout();fig.savefig(out/'数据覆盖.png',dpi=180,bbox_inches='tight');plt.close(fig)
# 示例仅按日期、代码排序,选例过程不读取收益。
selected=[]
for field,label in [('upper_gap','现价上方稀疏'),('valley_above','双峰之间断档')]:
g=features[(features.chip_valid)&(features[field]==True)].sort_values(['date','code'])
if len(g):selected.append((g.iloc[0],label))
if selected:
fig,axes=plt.subplots(1,len(selected),figsize=(6*len(selected),4),squeeze=False)
for ax,(r,label) in zip(axes[0],selected):
data=read(ROOT/f"raw/chips_{r['date'].replace('-','')}.json.gz")['data']['distribution']['rows']
rr=sorted([x for x in data if dot(x['symbol'])==r['code']],key=lambda x:x['price'])
prices=np.array([x['price'] for x in rr])/r.raw_close;weights=np.array([x['percent'] for x in rr])
ax.bar(prices,weights,width=np.median(np.diff(prices))*.9,color='#4285a8')
ax.axvline(1,color='#d36539',lw=1.5,label='当日收盘价')
lo,hi=(1.02,1.12) if label=='现价上方稀疏' else (r.valley_low,r.valley_high)
ax.axvspan(lo,hi,color='#e3b95f',alpha=.35,label='检测区间')
ax.set_xlim(.65,1.4);ax.set_xlabel('筹码成本 / 当日收盘价');ax.set_ylabel('该价格档筹码占比(%)')
ax.set_title(f"{label}\n{r['code']} {r['date']}");ax.legend(frameon=False,fontsize=9);ax.spines[['top','right']].set_visible(False)
fig.tight_layout();fig.savefig(out/'两种断档示例.png',dpi=180,bbox_inches='tight');plt.close(fig)
def main():
install_offline();p=protocol();res=read(ROOT/'results.json');aud=read(ROOT/'audit.json')
phase=read(ROOT/'phase_coverage.json');chip=read(ROOT/'chip_coverage.json');prices=read(ROOT/'price_coverage.json');diag=read(ROOT/'data_diagnostics.json')
g=pd.read_csv(ROOT/'group_results.csv',dtype={'group':str});features=pd.read_csv(ROOT/'cache/features_before_outcomes.csv.gz')
figures(res,g,features)
labels={'upper_gap':'上方稀疏','valley_above':'上方双峰断档','frequency_down':'下跌中至少2次吸筹',
'frequency_down_same_days':'再控制吸筹天数','frequency_down_intact':'只看结构未失效','frequency20_down':'改看此前20日次数'}
primary=[]
for key in ['upper_gap','valley_above','frequency_down']:
r=res['tests'][key];primary.append([labels[key],r['completed_pairs'],pct(r.get('treat_win')),pct(r.get('control_win')),points(r['win_lift']),points(r['mean_return_lift'])])
sensitivity=[]
for key,r in res['tests'].items():
sensitivity.append([labels[key],r['completed_pairs'],points(r['win_lift']),points(r.get('win_lift_cost60')),points(r.get('mean_return_lift_cost60'))])
freq=[]
for _,r in g[g.analysis=='frequency_down'].iterrows():
freq.append([r['group'],int(r['signals']),int(r['completed']),pct(r['win_rate']),pct(r['mean_net30'],True),pct(r['median_net30'],True),pct(r['hit15']),pct(r['up_before_down']),pct(r['max_close_loss'],True)])
paths=[]
for key in ['upper_gap','valley_above','frequency_down']:
r=res['tests'][key];v=r.get('path_comparison',{})
paths.append([labels[key],r.get('path_pairs',0),pct(v.get('treat_hit15')),pct(v.get('control_hit15')),points(v.get('hit15_lift')),points(v.get('up_first_lift'))])
joint=[]
for key in ['upper_gap','valley_above']:
for _,r in g[g.analysis==key+'_joint_down'].iterrows():
gap='有' if 'gap=True' in r['group'] else '无';high='至少2次' if 'high_freq=True' in r['group'] else '0或1次'
joint.append([labels[key],gap,high,int(r.signals),int(r.completed),pct(r.win_rate)])
cover=res['coverage'];q=aud['source_quality'];now=res['observed_at'];proto_time=p['frozen_at']
text=f'''# 筹码断档与吸筹次数的首轮实验
核验时间:{now}。协议冻结时间:{proto_time}。行情截止2026年10月29日。
本轮完成数据审计、两类断档计算、吸筹轮次去重和60日历史对照。结果属于现存历史数据的探索,尚不足以确认稳定的买入胜率优势。套牢盘压力表只有26个数据日;可用的较长历史来自筹码价格分档和吸筹阶段回放。回填数据、重复市场环境和不完整的历史资格覆盖限制了结论。
目前只有上方稀疏出现了较小样本的正向差异。下跌中多次吸筹的原始胜率较高,但在可比股票之间没有保留胜率优势;双峰断档的成熟配对更少。这里的“没有看到优势”不能解释为已证明指标无效,继续收集冻结口径后的新样本才有机会区分偶然性和稳定关系。
## 三项主要比较
每笔在信号次交易日开盘尝试买入,满60个交易日后开盘卖出。卖出受阻顺延。费用每边0.3%。配对只使用信号日的行业、市值、跌幅、波动率、换手率及价格位置;两项筹码比较还控制上方筹码总量。
{table(['问题','完成配对','处理组净胜率','对照组净胜率','胜率差','平均净收益差'],primary)}
“平均净收益差”是两组单笔平均收益之差,不是资金账户的累计收益。完整匹配、未匹配、不能买入和未完成退出的数量见[结果明细](results.json),逐对记录见[配对表](matched_pairs.csv.gz)。
上方稀疏组虽然较对照高14.29个百分点,但绝对胜率只有33.33%,平均净收益仍为负6.33%,并不代表买入有较高胜率。其未匹配的原始分组胜率为27.75%,无稀疏组为33.59%,方向与这21组局部配对不同。全部1488个稀疏观察只有40个找到对照,其中21个成熟;结果只代表这些特征重叠的样本,不能推广到全部稀疏股票。双峰135个观察仅7个匹配、6个完成;下跌多次吸筹105个观察仅27个匹配、14个成熟。
配对门槛也没有消除所有差异:双峰组的波动率标准化差仍为负0.58,样本之间尚有明显不平衡。本轮没有根据收益放宽门槛,也没有把匹配解释成随机试验。
![匹配后的胜率差](figures/匹配后的胜率差.png)
本轮成熟观察来自{cover['mature_sample_dates']}个信号日,只有{cover['mature_time_blocks']}个按60交易日划分的时间块。图中两个区间分别按时间块和股票重抽,均只作描述。依照事先约定,不出正式p值,不执行缺乏前提的显著性判定,也不把较高原始胜率当作验证通过。
双峰的6组配对中,各组两边的盈利与亏损状态都相同,经验重抽会机械地产生零宽区间。这不表示真实差异没有不确定性,所以不展示该区间。相邻时间块的持有期仍会交叠,3个时间块也不等于3段完全独立的市场环境。
## 60日内是否出现拉升
为了对应“之后会不会拉升”,另看完整60日收盘路径中的15%上涨,而不把最终盈利和曾经上涨混为一谈。
{table(['比较','路径完整配对','处理组曾涨15%','对照组曾涨15%','拉升比例差','先涨15%比例差'],paths)}
上方稀疏在这18组完整路径中没有提高15%拉升比例;下跌中多次吸筹的13组完整路径也没有显示优势。这些是预定的辅助指标,样本依然很少,不能用来反向选择阈值。
## 下跌中的吸筹次数
下跌定义为此前60日跌幅至少10%,且61日对数价格趋势向下。次数来自已有episode_id,按当时recognized_date计数;观察窗口不含信号当天。同一次吸筹持续多日仍计一次,未结束事件保留。
{table(['此前60日次数','观察样本','完成交易','净胜率','平均净收益','中位净收益','曾涨15%','先涨15%','平均最大收盘浮亏'],freq)}
![下跌中的吸筹次数](figures/下跌中的吸筹次数.png)
“曾涨15%”只说明观察期收盘价达到过该幅度;“先涨15%”要求达到15%之前没有先跌到负10%。两者都不是在最高价卖出的交易收益。最大浮亏按收盘价计算,没有冒充盘中最大回撤。路径缺价的样本不计算这些路径指标,退出结果仍单独保留。
至少2次组只有54笔完成交易,原始胜率46.30%;0次组为38.57%,1次组为37.80%。对照后两边均为57.14%,且多次组平均净收益2.57%,对照为17.51%。这说明原始组间差异还不能归因于吸筹次数。当前完整历史最多观察到3次,20日内没有至少2次的合格下跌样本,短窗口检验暂时没有数据。
## 费用与持续天数的敏感性
{table(['比较','完成配对','每边0.3%胜率差','每边0.6%胜率差','每边0.6%平均收益差'],sensitivity)}
“再控制吸筹天数”要求两只股票的明确吸筹天数相差不超过5天。它用于区分独立轮次与持续时间。如果配对不足,不能解释为次数没有作用。“结构未失效”只使用信号日已经知道的失效与空耗标记,不按未来是否破位筛选。
再控制持续天数后的66.67个百分点差异只来自3组配对,不作为正面证据。每边成本提高至0.6%后,下跌多次吸筹主比较的胜率差由0变为负7.14个百分点。
## 两类筹码断档
现价上方稀疏使用现价上方2%至12%的固定区间,区间筹码不超过全部筹码的5%。这也会包含全部筹码都在现价下方的情况,因此匹配时额外控制上方总量。
双峰断档使用现价70%至130%的搜索范围,按现价1%的宽度分档。三点平滑只用于定位峰,两侧峰附近各需至少5%的筹码。峰间连续低密度区至少宽5%,原始分档密度不高于较弱峰的四分之一。价格档过粗则不判定双峰。主检验使用下沿至少高于现价2%的上方断档;跨越现价、位于现价下方以及两种形态重合的结果另列在[全部分组表](group_results.csv)。
总筹码质量应处在98%至102%之间,再归一为1。缺失、负数、重复价格档和质量不合格均保留原因,不填成“没有断档”。{chip['valid_chip_samples']}个观察样本具有合格分布,其中上方稀疏{chip['upper_gap_samples']}个,上方双峰断档{chip['above_valley_samples']}个。
其中{diag['coarse_valid_samples']}个样本的档位较粗,不能判断双峰;上方稀疏组仍有{diag['coarse_upper_gap_samples']}个粗档样本,固定区间占比也存在离散误差。分档估算的获利比例与gp_chip_data摘要相差超过5个百分点的有{diag['winner_gap_over5pp_samples']}个,不能混用两种口径。5月22日请求327只股票仅有2只股票返回分档,10月16日请求307只仅返回265只。缺失记录按预定规则排除筹码比较,保留吸筹比较,不根据其收益决定去留。
![两类断档的实际样本](figures/两种断档示例.png)
示例按日期和股票代码选择,选例过程不读取收益。筹码字段的价格和百分比含义可参照[Tushare筹码分布说明](https://tushare.pro/document/2?doc_id=294)。该类分布使用交易价量与衰减模型估计,研究中的“筹码移动”不等于观测到了真实账户持仓迁移。
## 同时出现断档和多次吸筹
以下只展示下跌样本中的联合分组,未进行配对调整。
{table(['断档类型','断档','此前60日吸筹','观察数','完成交易','净胜率'],joint)}
上方稀疏叠加至少2次吸筹只有2笔完成交易;上方双峰叠加至少2次只有1个观察且尚未成熟。当前无法检验组合是否比单独指标更强。跨越现价的双峰断档有3笔完成交易,位于现价下方的有71笔,结果分别列在全部分组表,不合并成上方阻力断档。
## 数据覆盖与真实性
| 数据 | 实际覆盖 | 本轮用途 |
| --- | --- | --- |
| gp_stock_trap_pressure | {q['trap']['start']}至{q['trap']['end']},{q['trap']['days']}日,{q['trap']['rows']:,}行 | 核对现存套牢指标;没有满60日的直接检验样本 |
| accumulation_daily | {q['daily']['start']}至{q['daily']['end']},{q['daily']['days']}日,{q['daily']['rows']:,}行 | 核对版本与数据质量;不足以独立形成完整60日历史计数 |
| accumulation_phase_daily | {phase['dates'][0]}至{phase['dates'][1]},{phase['batches']}批,{phase['rows']:,}行 | 轮次及明确吸筹天数;{phase['replay_days']}个回放日、{phase['live_days']}个实际运行日 |
| 2025年筹码分档 | 2025年11月10日至20日,仅9日 | 不能把年度表名当全年连续覆盖 |
| 2026年筹码分档 | 2026年1月5日至10月29日,197个数据日 | 从实际存在的日期读取所需股票,计算两类断档 |
![数据覆盖](figures/数据覆盖.png)
抽样在查看收益之前固定。股票代码经SHA256分成14组,轮流进入每周最后一个交易日的观察;同股间隔至少60个市场交易日。共有{cover['all_samples']:,}个合格观察、{cover['mature_samples']:,}个成熟观察,{cover['pending_60_days']:,}个尚未走完60日。最后一个成熟信号日为{cover['latest_mature_signal']}。
吸筹阶段的历史回放于后来生成,筹码历史也存在回填。本轮合格筹码样本中,记录创建日期晚于数据日期的有{chip['chip_created_later_than_data_day']:,}个。创建日期相同也不能证明收盘前已可用。历史行业记录按数据日期向后接续,但有{phase['industry_backfill_stock_samples']:,}个观察的行业记录在事后写入。以上结果因此采用次日开盘的历史研究假设,不冒充当年实际可交易的完整信息集。
accumulation_daily同时存在fund_flow_v1的50171行和fund_flow_v2的134152行。本轮没有混合这两个版本的分数阈值,而是固定使用phase-v1-20261012代次的事件识别日期和状态。吸筹评分本身包含价格低位信息,因此匹配另控制过去跌幅和价格位置;不能把评分与后续反弹的关系全部解释为资金吸筹的作用。
价格面板覆盖{prices['stocks']}只沪深股票。缺上市日期的代码有{len(prices['missing_listing_codes'])}只;历史名称缺失的股票日有{prices['missing_historical_name_stock_days']:,}个,均按协议排除。退市覆盖和历史修订未完全认证。北交所未纳入,整手、最低佣金、容量与真实撮合未模拟。
## 套牢盘与吸筹的关系
历史套牢压力使用筹码分档直接计算的上方占比及上方15%内占比。没有把它伪装成gp_stock_trap_pressure的历史tsp评分,也没有回写该表。与现存表重合的{chip['trap_reconciliation_rows']}个股票日对照见[套牢指标核对](trap_reconciliation.csv)。近端与远端定义分别比较,不能仅凭ocr_115的名字猜测含义。
进一步对{diag['trap_comparison_n']}条原始分档核对,ocr_above等于价格严格高于压力表收盘价的原始筹码百分比之和;ocr_115等于价格大于或等于该收盘价115%的百分比之和。两者最大绝对误差小于0.000000001个百分点。因此ocr_115对应远端套牢盘,并非上方15%以内。研究特征按总质量归一,边界也另有定义,不能直接冒充表内字段。两处收盘价格有{diag['trap_close_mismatch_rows']}条不一致,最大差{diag['trap_close_max_difference']:.2f}元;这些差异见[数据口径诊断](data_diagnostics.json)。
在{chip['fixed_price_change_pairs']}个可比较样本中,还将20日前筹码价格调整到今天的同一复权坐标,以今天收盘价作为两期共同门槛,比较上方占比变化。解释变量来自更早20日的吸筹次数,控制本段涨跌幅、市值、波动和换手。描述性残差相关为{res['mechanism']['partial_correlation']},可用样本为{res['mechanism']['n']};它不能解释为因果作用。当前筹码形态与吸筹次数的关系详见[结果文件](results.json)中的feature_relationships。
固定价格门槛下的相关约为负0.016;当期次数与上方筹码总量的控制后相关约为负0.006。两者都接近零,本轮未看到“吸筹越多,套牢盘就明显越少”的清楚线性关系。这不排除其他非线性关系,但本轮不会继续试阈值寻找正面结果。
## 验证与复现
开发机通过隔离配置和断网测试。{aud['checks']['independent_samples_recomputed']}条真实观察独立复算历史次数、买卖索引和费用;{aud['checks']['matching_calipers_verified']}组配对逐项复核门槛。分组数量守恒,特征摘要在收益计算后保持不变。trade_cal为空,日期轴已与上证、创业板、沪深300指数交叉核对;最后4个日期由全市场日线核对,尚缺独立交易所日历验证。吸筹导出在155的一次性只读研究容器逐行校验payload摘要,并校验每个批次的汇总摘要;所有业务SQL严格单表。
10项专项测试通过,覆盖计数排除当日、缺失不填零、历史截断不变、断档尺度不变、粗档拒绝、交易时序、未成熟样本、路径先后、匹配不读收益及退化区间。另核对98项源摘要;用原始缓存重新生成筹码特征,与首次计算收益前的摘要一致。四幅研究图已检查中文、单位、样本量和注释。155一次性研究容器已回收,当前未残留本任务容器。
主入口为[协议](protocol.json)、[实现细化](implementation_notes.json)、[数据审计](audit.json)、[结果](results.json)。原始数据在raw/,中间面板在cache/,均保留在开发机且不入Git。源摘要和最终文件摘要见[清单](manifest.json)。
开发机复现使用本轮记录的Python依赖。以下命令不会访问运行数据库,要求原始缓存已存在:
```bash
python research/chip_accumulation_20261030/prepare_prices.py
python research/chip_accumulation_20261030/prepare_phase.py
python research/chip_accumulation_20261030/prepare_chips.py
python research/chip_accumulation_20261030/analyze.py
python scripts/unit_test_guard.py research/chip_accumulation_20261030/test_research.py
python scripts/unit_test_guard.py research/chip_accumulation_20261030/audit.py
python research/chip_accumulation_20261030/diagnose_data.py
python research/chip_accumulation_20261030/report.py
python research/chip_accumulation_20261030/finalize.py
```
## 未完成的验证
冻结口径后的新样本尚未走完60日,因此前瞻验证未完成。现有历史没有足够独立市场阶段,不能用大量股票行数补足。后续应继续保存不可覆盖的数据快照,在新样本成熟后按同一规则复算;若改变断档阈值或吸筹窗口,应另建实验版本。
本轮仅有源数据读取和开发机研究文件,没有修改PMS运行参数、持仓、委托或策略规则,没有部署业务服务。
'''
(ROOT/'筹码断档与吸筹次数首轮实验.md').write_text(text)
print('报告和图形已生成',flush=True)
if __name__=='__main__':main()

View File

@ -0,0 +1,4 @@
numpy==2.3.5
pandas==2.2.3
scipy==1.16.3
matplotlib==3.10.7

View File

@ -0,0 +1,655 @@
{
"observed_at": "2026-10-30T09:38:59.829275+08:00",
"coverage": {
"all_samples": 14595,
"mature_samples": 10231,
"mature_stocks": 4960,
"pending_60_days": 4364,
"latest_mature_signal": "2026-07-24",
"mature_sample_dates": 31,
"mature_time_blocks": 3,
"chip_mature_samples": 8941,
"downtrend_mature_samples": 3200,
"historical_only": true,
"formal_inference_performed": false,
"reason": "已有历史包含回填与回放,且独立60日时间块不足8个。"
},
"tests": {
"upper_gap": {
"matching": {
"treated_with_matching_fields": 1488,
"unmatched": 1448,
"matched": 40,
"calipers": {
"ret60": 0.05,
"log_cap": 0.6931471805599453,
"log_vol": 0.6931471805599453,
"log_turn": 0.6931471805599453,
"position60": 0.2,
"overhead": 0.1
}
},
"all_pairs": 40,
"mature_pairs": 21,
"completed_pairs": 21,
"win_lift": 0.14285714285714285,
"mean_return_lift": 0.0391437340801432,
"treat_entry_failures": 0,
"control_entry_failures": 0,
"unmatched_or_unfinished_pairs": 0,
"path_pairs": 18,
"path_comparison": {
"treat_hit15": 0.2777777777777778,
"control_hit15": 0.2777777777777778,
"hit15_lift": 0.0,
"treat_up_first": 0.2222222222222222,
"control_up_first": 0.2222222222222222,
"up_first_lift": 0.0
},
"treat_win": 0.3333333333333333,
"control_win": 0.19047619047619047,
"mean_treat_return": -0.06329803651482903,
"mean_control_return": -0.10244177059497221,
"median_treat_return": -0.08287778174909222,
"median_control_return": -0.16283949250580887,
"win_lift_cost60": 0.09523809523809523,
"mean_return_lift_cost60": 0.038909569953089244,
"mean_clear_days_difference": -2.8095238095238093,
"uncertainty": {
"n": 21,
"mean": 0.14285714285714285,
"time_blocks": 3,
"unique_stocks": 42,
"time_interval": [
0.0,
0.2222222222222222
],
"stock_interval": [
0.0,
0.43767361111111097
],
"formal_p_value": null,
"degenerate_observed_differences": false,
"interval_note": "少量簇的重抽区间只作描述。",
"interpretation": "固定配对的描述性区间;少于8个时间块不进行正式显著性判断。"
},
"matched_balance": {
"ret60": {
"mean_difference": 0.0024576208933672896,
"standardized_difference": 0.008890655714009722
},
"log_cap": {
"mean_difference": -0.010400010771270814,
"standardized_difference": -0.009478408097229035
},
"log_vol": {
"mean_difference": 0.013761083691130752,
"standardized_difference": 0.033450770861500584
},
"log_turn": {
"mean_difference": 0.06329404683059812,
"standardized_difference": 0.07429212307972964
},
"position60": {
"mean_difference": 0.035714285714285726,
"standardized_difference": 0.10419813289109982
},
"overhead": {
"mean_difference": -0.03776617662850414,
"standardized_difference": -0.12915796461441825
}
},
"signal_level_lift_bounds_unfilled_cash_zero": [
0.14285714285714285,
0.14285714285714285
],
"monthly": [
{
"month": "2026-01",
"pairs": 7,
"win_lift": 0.0,
"return_lift": -0.04466513737372998
},
{
"month": "2026-02",
"pairs": 4,
"win_lift": 0.25,
"return_lift": 0.032092917801094106
},
{
"month": "2026-03",
"pairs": 1,
"win_lift": 1.0,
"return_lift": 0.32885380806653985
},
{
"month": "2026-04",
"pairs": 3,
"win_lift": 0.3333333333333333,
"return_lift": 0.3394255669037222
},
{
"month": "2026-05",
"pairs": 5,
"win_lift": 0.0,
"return_lift": -0.0627324258689081
},
{
"month": "2026-06",
"pairs": 1,
"win_lift": 0.0,
"return_lift": -0.027165673338425456
}
]
},
"valley_above": {
"matching": {
"treated_with_matching_fields": 135,
"unmatched": 128,
"matched": 7,
"calipers": {
"ret60": 0.05,
"log_cap": 0.6931471805599453,
"log_vol": 0.6931471805599453,
"log_turn": 0.6931471805599453,
"position60": 0.2,
"overhead": 0.1
}
},
"all_pairs": 7,
"mature_pairs": 7,
"completed_pairs": 6,
"win_lift": 0.0,
"mean_return_lift": -0.10672806437857897,
"treat_entry_failures": 1,
"control_entry_failures": 0,
"unmatched_or_unfinished_pairs": 1,
"path_pairs": 6,
"path_comparison": {
"treat_hit15": 0.6666666666666666,
"control_hit15": 0.6666666666666666,
"hit15_lift": 0.0,
"treat_up_first": 0.6666666666666666,
"control_up_first": 0.5,
"up_first_lift": 0.16666666666666666
},
"treat_win": 0.5,
"control_win": 0.5,
"mean_treat_return": 0.14572239301213918,
"mean_control_return": 0.2524504573907182,
"median_treat_return": 0.06027336333245453,
"median_control_return": -0.05032117510581924,
"win_lift_cost60": 0.0,
"mean_return_lift_cost60": -0.10608959989340226,
"mean_clear_days_difference": -1.0,
"uncertainty": {
"n": 6,
"mean": 0.0,
"time_blocks": 3,
"unique_stocks": 12,
"time_interval": null,
"stock_interval": null,
"formal_p_value": null,
"degenerate_observed_differences": true,
"interval_note": "各配对观察差完全相同,经验重抽退化,不能给出有效不确定性区间。",
"interpretation": "固定配对的描述性区间;少于8个时间块不进行正式显著性判断。"
},
"matched_balance": {
"ret60": {
"mean_difference": -0.006852313215853763,
"standardized_difference": -0.022695636941937308
},
"log_cap": {
"mean_difference": -0.3021595875656609,
"standardized_difference": -0.26576012086759343
},
"log_vol": {
"mean_difference": -0.23817027337643046,
"standardized_difference": -0.5773300296254709
},
"log_turn": {
"mean_difference": 0.04232134044551903,
"standardized_difference": 0.04915080048574294
},
"position60": {
"mean_difference": -0.06388888888888891,
"standardized_difference": -0.1848538644232215
},
"overhead": {
"mean_difference": -0.003321951729226169,
"standardized_difference": -0.011097702732125547
}
},
"signal_level_lift_bounds_unfilled_cash_zero": [
0.0,
0.0
],
"monthly": [
{
"month": "2026-01",
"pairs": 1,
"win_lift": 0.0,
"return_lift": 0.15600403223893322
},
{
"month": "2026-03",
"pairs": 2,
"win_lift": 0.0,
"return_lift": -0.2513629900597268
},
{
"month": "2026-04",
"pairs": 1,
"win_lift": 0.0,
"return_lift": -0.4190438577091009
},
{
"month": "2026-07",
"pairs": 2,
"win_lift": 0.0,
"return_lift": 0.0626987096590737
}
]
},
"frequency_down": {
"matching": {
"treated_with_matching_fields": 105,
"unmatched": 78,
"matched": 27,
"calipers": {
"ret60": 0.05,
"log_cap": 0.6931471805599453,
"log_vol": 0.6931471805599453,
"log_turn": 0.6931471805599453,
"position60": 0.2
}
},
"all_pairs": 27,
"mature_pairs": 14,
"completed_pairs": 14,
"win_lift": 0.0,
"mean_return_lift": -0.14942164392396867,
"treat_entry_failures": 0,
"control_entry_failures": 0,
"unmatched_or_unfinished_pairs": 0,
"path_pairs": 13,
"path_comparison": {
"treat_hit15": 0.38461538461538464,
"control_hit15": 0.5384615384615384,
"hit15_lift": -0.15384615384615385,
"treat_up_first": 0.38461538461538464,
"control_up_first": 0.5384615384615384,
"up_first_lift": -0.15384615384615385
},
"treat_win": 0.5714285714285714,
"control_win": 0.5714285714285714,
"mean_treat_return": 0.02571136894001849,
"mean_control_return": 0.17513301286398716,
"median_treat_return": 0.00661533078738008,
"median_control_return": 0.04092541761524171,
"win_lift_cost60": -0.07142857142857142,
"mean_return_lift_cost60": -0.1485277795639464,
"mean_clear_days_difference": 12.928571428571429,
"uncertainty": {
"n": 14,
"mean": 0.0,
"time_blocks": 3,
"unique_stocks": 28,
"time_interval": [
-0.5,
0.1
],
"stock_interval": [
-0.5833333333333334,
0.562723214285714
],
"formal_p_value": null,
"degenerate_observed_differences": false,
"interval_note": "少量簇的重抽区间只作描述。",
"interpretation": "固定配对的描述性区间;少于8个时间块不进行正式显著性判断。"
},
"matched_balance": {
"ret60": {
"mean_difference": 0.007931698456118845,
"standardized_difference": 0.092436727114394
},
"log_cap": {
"mean_difference": 0.10546707994687678,
"standardized_difference": 0.1019322193115162
},
"log_vol": {
"mean_difference": -0.06936235028233871,
"standardized_difference": -0.19601871280330435
},
"log_turn": {
"mean_difference": 0.029180362996087783,
"standardized_difference": 0.04093975096794802
},
"position60": {
"mean_difference": 0.005952380952380953,
"standardized_difference": 0.03011176849935174
}
},
"signal_level_lift_bounds_unfilled_cash_zero": [
0.0,
0.0
],
"monthly": [
{
"month": "2025-12",
"pairs": 2,
"win_lift": -0.5,
"return_lift": -0.09916598488933204
},
{
"month": "2026-04",
"pairs": 1,
"win_lift": 0.0,
"return_lift": 0.028627892089139295
},
{
"month": "2026-05",
"pairs": 1,
"win_lift": 0.0,
"return_lift": 0.005613835809489931
},
{
"month": "2026-06",
"pairs": 6,
"win_lift": -0.16666666666666666,
"return_lift": -0.3897326456579389
},
{
"month": "2026-07",
"pairs": 4,
"win_lift": 0.5,
"return_lift": 0.10264577522302665
}
]
},
"frequency_down_same_days": {
"matching": {
"treated_with_matching_fields": 105,
"unmatched": 98,
"matched": 7,
"calipers": {
"ret60": 0.05,
"log_cap": 0.6931471805599453,
"log_vol": 0.6931471805599453,
"log_turn": 0.6931471805599453,
"position60": 0.2,
"clear_days60": 5.0
}
},
"all_pairs": 7,
"mature_pairs": 3,
"completed_pairs": 3,
"win_lift": 0.6666666666666666,
"mean_return_lift": 0.21108052394194884,
"treat_entry_failures": 0,
"control_entry_failures": 0,
"unmatched_or_unfinished_pairs": 0,
"path_pairs": 3,
"path_comparison": {
"treat_hit15": 0.3333333333333333,
"control_hit15": 0.3333333333333333,
"hit15_lift": 0.0,
"treat_up_first": 0.3333333333333333,
"control_up_first": 0.0,
"up_first_lift": 0.3333333333333333
},
"treat_win": 0.6666666666666666,
"control_win": 0.0,
"mean_treat_return": 0.06580637978843092,
"mean_control_return": -0.14527414415351791,
"median_treat_return": 0.020474697580263213,
"median_control_return": -0.18725822378103563,
"win_lift_cost60": 0.6666666666666666,
"mean_return_lift_cost60": 0.2098178062225364,
"mean_clear_days_difference": 0.3333333333333333,
"uncertainty": {
"n": 3,
"mean": 0.6666666666666666,
"time_blocks": 2,
"unique_stocks": 6,
"time_interval": [
0.0,
1.0
],
"stock_interval": [
0.0,
1.0
],
"formal_p_value": null,
"degenerate_observed_differences": false,
"interval_note": "少量簇的重抽区间只作描述。",
"interpretation": "固定配对的描述性区间;少于8个时间块不进行正式显著性判断。"
},
"matched_balance": {
"ret60": {
"mean_difference": -0.00870272744095837,
"standardized_difference": -0.10142236824323918
},
"log_cap": {
"mean_difference": 0.01847513382084524,
"standardized_difference": 0.017855916683998184
},
"log_vol": {
"mean_difference": 0.23179164946646802,
"standardized_difference": 0.6550455770605677
},
"log_turn": {
"mean_difference": 0.006608162185797813,
"standardized_difference": 0.009271183990365178
},
"position60": {
"mean_difference": -0.027777777777777766,
"standardized_difference": -0.14052158633030803
}
},
"signal_level_lift_bounds_unfilled_cash_zero": [
0.6666666666666666,
0.6666666666666666
],
"monthly": [
{
"month": "2026-04",
"pairs": 1,
"win_lift": 0.0,
"return_lift": 0.028627892089139295
},
{
"month": "2026-06",
"pairs": 1,
"win_lift": 1.0,
"return_lift": 0.23371507096689592
},
{
"month": "2026-07",
"pairs": 1,
"win_lift": 1.0,
"return_lift": 0.3708986087698113
}
]
},
"frequency_down_intact": {
"matching": {
"treated_with_matching_fields": 61,
"unmatched": 45,
"matched": 16,
"calipers": {
"ret60": 0.05,
"log_cap": 0.6931471805599453,
"log_vol": 0.6931471805599453,
"log_turn": 0.6931471805599453,
"position60": 0.2
}
},
"all_pairs": 16,
"mature_pairs": 7,
"completed_pairs": 7,
"win_lift": 0.0,
"mean_return_lift": -0.0492534195519005,
"treat_entry_failures": 0,
"control_entry_failures": 0,
"unmatched_or_unfinished_pairs": 0,
"path_pairs": 7,
"path_comparison": {
"treat_hit15": 0.14285714285714285,
"control_hit15": 0.42857142857142855,
"hit15_lift": -0.2857142857142857,
"treat_up_first": 0.14285714285714285,
"control_up_first": 0.42857142857142855,
"up_first_lift": -0.2857142857142857
},
"treat_win": 0.5714285714285714,
"control_win": 0.5714285714285714,
"mean_treat_return": -0.03875414576911199,
"mean_control_return": 0.010499273782788508,
"median_treat_return": 0.0017354767994157694,
"median_control_return": 0.05822384386415691,
"win_lift_cost60": -0.14285714285714285,
"mean_return_lift_cost60": -0.048958777656720534,
"mean_clear_days_difference": 5.142857142857143,
"uncertainty": {
"n": 7,
"mean": 0.0,
"time_blocks": 3,
"unique_stocks": 14,
"time_interval": [
-0.5,
0.25
],
"stock_interval": [
-0.6666666666666666,
0.7537499999999966
],
"formal_p_value": null,
"degenerate_observed_differences": false,
"interval_note": "少量簇的重抽区间只作描述。",
"interpretation": "固定配对的描述性区间;少于8个时间块不进行正式显著性判断。"
},
"matched_balance": {
"ret60": {
"mean_difference": 0.00506925103889423,
"standardized_difference": 0.058117754638148264
},
"log_cap": {
"mean_difference": 0.06263189891111891,
"standardized_difference": 0.057614788306884275
},
"log_vol": {
"mean_difference": -0.03503354542763493,
"standardized_difference": -0.09802477196676858
},
"log_turn": {
"mean_difference": -0.06478062483012119,
"standardized_difference": -0.08952562165362918
},
"position60": {
"mean_difference": 0.02380952380952383,
"standardized_difference": 0.11879408049232834
}
},
"signal_level_lift_bounds_unfilled_cash_zero": [
0.0,
0.0
],
"monthly": [
{
"month": "2025-12",
"pairs": 2,
"win_lift": -0.5,
"return_lift": -0.09916598488933204
},
{
"month": "2026-04",
"pairs": 1,
"win_lift": 0.0,
"return_lift": -0.10736549037905385
},
{
"month": "2026-06",
"pairs": 3,
"win_lift": 0.3333333333333333,
"return_lift": 0.011387364475456518
},
{
"month": "2026-07",
"pairs": 1,
"win_lift": 0.0,
"return_lift": -0.07323857013195512
}
]
},
"frequency20_down": {
"matching": {
"treated_with_matching_fields": 0,
"unmatched": 0,
"matched": 0,
"calipers": {
"ret60": 0.05,
"log_cap": 0.6931471805599453,
"log_vol": 0.6931471805599453,
"log_turn": 0.6931471805599453,
"position60": 0.2
}
},
"all_pairs": 0,
"mature_pairs": 0,
"completed_pairs": 0,
"win_lift": null,
"mean_return_lift": null
}
},
"mechanism": {
"n": 11908,
"partial_correlation": -0.01613456492755954,
"controls": [
"return_last20",
"log_cap",
"log_vol",
"log_turn"
],
"interpretation": "描述性线性残差相关,非因果效应,未作显著性检验"
},
"feature_relationships": {
"overhead": {
"n": 13247,
"partial_correlation": -0.005776517255221698,
"controls": [
"ret60",
"position60",
"log_cap",
"log_vol",
"log_turn"
],
"interpretation": "描述性线性残差相关,非因果效应,未作显著性检验"
},
"overhead_near15": {
"n": 13247,
"partial_correlation": -0.011865461107237112,
"controls": [
"ret60",
"position60",
"log_cap",
"log_vol",
"log_turn"
],
"interpretation": "描述性线性残差相关,非因果效应,未作显著性检验"
},
"upper_mass": {
"n": 13247,
"partial_correlation": -0.01140018925409177,
"controls": [
"ret60",
"position60",
"log_cap",
"log_vol",
"log_turn"
],
"interpretation": "描述性线性残差相关,非因果效应,未作显著性检验"
}
},
"verdict": "仅提供历史探索和反例;现有历史不能确认稳定的60日胜率提升。"
}

View File

@ -0,0 +1,69 @@
"""关键计数、断档、交易时序及匹配隔离测试。使用unit_test_guard运行。"""
import unittest
import numpy as np
import pandas as pd
from common import previous_sum
from chip_features import chip_features
from analysis_core import outcome,match_pairs,cluster_intervals
class ResearchTests(unittest.TestCase):
def test_history_excludes_today_and_preserves_missing(self):
a=np.array([1,0,1,0,1],float)
np.testing.assert_allclose(previous_sum(a,2),[np.nan,np.nan,1,1,1],equal_nan=True)
a[1]=np.nan
np.testing.assert_allclose(previous_sum(a,2),[np.nan,np.nan,np.nan,np.nan,1],equal_nan=True)
def test_prefix_does_not_change_history(self):
a=np.arange(20,dtype=float)%3
np.testing.assert_allclose(previous_sum(a,5)[:12],previous_sum(a[:12],5),equal_nan=True)
def test_upper_gap_and_scale_invariance(self):
p=np.linspace(5,15,101);w=np.exp(-((p-8)/.4)**2);w=w/w.sum()*100
a=chip_features(p,w,10);b=chip_features(p*2,w,20)
self.assertTrue(a['upper_gap']);self.assertAlmostEqual(a['upper_mass'],b['upper_mass'])
self.assertEqual(a['valley_above'],b['valley_above'])
def test_above_valley_detected(self):
p=np.linspace(5,15,501);w=np.exp(-((p-10.4)/.12)**2)+np.exp(-((p-12.4)/.12)**2);w*=100/w.sum()
a=chip_features(p,w,10)
self.assertTrue(a['valley_evaluable']);self.assertTrue(a['valley_above']);self.assertGreaterEqual(a['valley_width'],.05)
def test_bad_mass_and_coarse_grid_not_valid_valley(self):
p=np.linspace(5,15,20);w=np.ones(20)*5
a=chip_features(p,w,10);self.assertTrue(a['chip_valid']);self.assertFalse(a['valley_evaluable'])
self.assertFalse(chip_features(p,w*.5,10)['chip_valid'])
self.assertFalse(chip_features(np.r_[p[:-1],p[0]],w,10)['chip_valid'])
def make_panel(self):
return {'dates':np.array([str(i) for i in range(70)]),'open':np.ones((70,1))*10,
'close':np.ones((70,1))*10,'can_buy':np.ones((70,1),bool),'can_sell':np.ones((70,1),bool)}
def test_entry_and_exit_delay_and_cost(self):
a=self.make_panel();a['open'][1,0]=10;a['open'][62,0]=12;a['can_sell'][61,0]=False
r=outcome(a,0,0);self.assertEqual(r['exit_date'],'62');self.assertEqual(r['exit_delay'],1)
self.assertAlmostEqual(r['net30'],1.2*.997/1.003-1)
a['can_buy'][1,0]=False;self.assertFalse(outcome(a,0,0)['entry_filled'])
def test_right_censoring_and_unresolved_exit(self):
a=self.make_panel();r=outcome(a,20,0);self.assertFalse(r['calendar_matured']);self.assertFalse(r['completed'])
a['can_sell'][61:,0]=False;r=outcome(a,0,0)
self.assertTrue(r['calendar_matured']);self.assertTrue(r['entry_filled']);self.assertFalse(r['completed'])
def test_path_order_and_missing(self):
a=self.make_panel();a['close'][3,0]=8;a['close'][5,0]=12
r=outcome(a,0,0);self.assertEqual(r['hit15'],1);self.assertEqual(r['up_before_down'],0)
a['close'][2,0]=np.nan;r=outcome(a,0,0);self.assertFalse(r['path_complete']);self.assertTrue(np.isnan(r['hit15']))
def test_matching_never_uses_outcome(self):
f=pd.DataFrame({'date':['a']*3,'industry':['银行']*3,'code':['a','b','c'],'ret60':[-.2,-.19,-.18],
'log_cap':[1.]*3,'log_vol':[1.]*3,'log_turn':[1.]*3,'position60':[.1]*3,'net30':[.1,-.8,.9]})
pairs,_=match_pairs(f,pd.Series([True,False,False]));self.assertEqual(pairs[0]['control_id'],1)
f.net30=-f.net30
other,_=match_pairs(f,pd.Series([True,False,False]));self.assertEqual(pairs,other)
def test_degenerate_resampling_is_not_zero_uncertainty(self):
r=cluster_intervals([0,0,0],[1,61,121],['a','b','c'],['d','e','f'],reps=100)
self.assertTrue(r['degenerate_observed_differences'])
self.assertIsNone(r['time_interval']);self.assertIsNone(r['stock_interval'])
if __name__=='__main__':unittest.main()

File diff suppressed because it is too large Load Diff

View File

@ -0,0 +1,22 @@
{
"checked_at": "2026-10-30T09:45:25.096209+08:00",
"machine": "开发机",
"configuration_isolated": true,
"network_blocked": true,
"runs": [
{
"entry": "python scripts/unit_test_guard.py research/chip_accumulation_20261030/test_research.py",
"exit_code": 0,
"stdout": "",
"stderr": "..........\n----------------------------------------------------------------------\nRan 10 tests in 0.055s\n\nOK\n"
},
{
"entry": "python scripts/unit_test_guard.py research/chip_accumulation_20261030/audit.py",
"exit_code": 0,
"stdout": "审计通过 {'source_digests_verified': 98, 'exchange_calendar_matches': None, 'index_calendars_match': True, 'recent_dates_marketwide': True, 'independent_samples_recomputed': 500, 'duplicate_stock_dates': 0, 'minimum_sample_spacing': 60, 'features_unchanged_after_outcomes': True, 'group_totals_reconcile': True, 'matching_calipers_verified': 97}\n",
"stderr": ""
}
],
"visual_review": "四幅PNG已逐张检查中文、单位、标注和样本量;均可读。",
"remote_cleanup": "2026-10-30T09:43:01+08:00,155 docker ps -a名称过滤pms-chip-research-,无残留。"
}

View File

@ -0,0 +1,161 @@
# 筹码断档与吸筹次数的首轮实验
核验时间:2026-10-30T09:38:59.829275+08:00。协议冻结时间:2026-10-30T09:12:35.387801+08:00。行情截止2026年10月29日。
本轮完成数据审计、两类断档计算、吸筹轮次去重和60日历史对照。结果属于现存历史数据的探索,尚不足以确认稳定的买入胜率优势。套牢盘压力表只有26个数据日;可用的较长历史来自筹码价格分档和吸筹阶段回放。回填数据、重复市场环境和不完整的历史资格覆盖限制了结论。
目前只有上方稀疏出现了较小样本的正向差异。下跌中多次吸筹的原始胜率较高,但在可比股票之间没有保留胜率优势;双峰断档的成熟配对更少。这里的“没有看到优势”不能解释为已证明指标无效,继续收集冻结口径后的新样本才有机会区分偶然性和稳定关系。
## 三项主要比较
每笔在信号次交易日开盘尝试买入,满60个交易日后开盘卖出。卖出受阻顺延。费用每边0.3%。配对只使用信号日的行业、市值、跌幅、波动率、换手率及价格位置;两项筹码比较还控制上方筹码总量。
| 问题 | 完成配对 | 处理组净胜率 | 对照组净胜率 | 胜率差 | 平均净收益差 |
| --- | --- | --- | --- | --- | --- |
| 上方稀疏 | 21 | 33.33% | 19.05% | +14.29个百分点 | +3.91个百分点 |
| 上方双峰断档 | 6 | 50.00% | 50.00% | +0.00个百分点 | -10.67个百分点 |
| 下跌中至少2次吸筹 | 14 | 57.14% | 57.14% | +0.00个百分点 | -14.94个百分点 |
“平均净收益差”是两组单笔平均收益之差,不是资金账户的累计收益。完整匹配、未匹配、不能买入和未完成退出的数量见[结果明细](results.json),逐对记录见[配对表](matched_pairs.csv.gz)。
上方稀疏组虽然较对照高14.29个百分点,但绝对胜率只有33.33%,平均净收益仍为负6.33%,并不代表买入有较高胜率。其未匹配的原始分组胜率为27.75%,无稀疏组为33.59%,方向与这21组局部配对不同。全部1488个稀疏观察只有40个找到对照,其中21个成熟;结果只代表这些特征重叠的样本,不能推广到全部稀疏股票。双峰135个观察仅7个匹配、6个完成;下跌多次吸筹105个观察仅27个匹配、14个成熟。
配对门槛也没有消除所有差异:双峰组的波动率标准化差仍为负0.58,样本之间尚有明显不平衡。本轮没有根据收益放宽门槛,也没有把匹配解释成随机试验。
![匹配后的胜率差](figures/匹配后的胜率差.png)
本轮成熟观察来自31个信号日,只有3个按60交易日划分的时间块。图中两个区间分别按时间块和股票重抽,均只作描述。依照事先约定,不出正式p值,不执行缺乏前提的显著性判定,也不把较高原始胜率当作验证通过。
双峰的6组配对中,各组两边的盈利与亏损状态都相同,经验重抽会机械地产生零宽区间。这不表示真实差异没有不确定性,所以不展示该区间。相邻时间块的持有期仍会交叠,3个时间块也不等于3段完全独立的市场环境。
## 60日内是否出现拉升
为了对应“之后会不会拉升”,另看完整60日收盘路径中的15%上涨,而不把最终盈利和曾经上涨混为一谈。
| 比较 | 路径完整配对 | 处理组曾涨15% | 对照组曾涨15% | 拉升比例差 | 先涨15%比例差 |
| --- | --- | --- | --- | --- | --- |
| 上方稀疏 | 18 | 27.78% | 27.78% | +0.00个百分点 | +0.00个百分点 |
| 上方双峰断档 | 6 | 66.67% | 66.67% | +0.00个百分点 | +16.67个百分点 |
| 下跌中至少2次吸筹 | 13 | 38.46% | 53.85% | -15.38个百分点 | -15.38个百分点 |
上方稀疏在这18组完整路径中没有提高15%拉升比例;下跌中多次吸筹的13组完整路径也没有显示优势。这些是预定的辅助指标,样本依然很少,不能用来反向选择阈值。
## 下跌中的吸筹次数
下跌定义为此前60日跌幅至少10%,且61日对数价格趋势向下。次数来自已有episode_id,按当时recognized_date计数;观察窗口不含信号当天。同一次吸筹持续多日仍计一次,未结束事件保留。
| 此前60日次数 | 观察样本 | 完成交易 | 净胜率 | 平均净收益 | 中位净收益 | 曾涨15% | 先涨15% | 平均最大收盘浮亏 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 0 | 3287 | 2209 | 38.57% | -3.04% | -4.41% | 39.22% | 32.16% | -14.78% |
| 1 | 1398 | 926 | 37.80% | -3.03% | -5.14% | 40.80% | 35.06% | -14.15% |
| 2+ | 105 | 54 | 46.30% | -1.87% | -2.23% | 39.62% | 35.85% | -12.85% |
![下跌中的吸筹次数](figures/下跌中的吸筹次数.png)
“曾涨15%”只说明观察期收盘价达到过该幅度;“先涨15%”要求达到15%之前没有先跌到负10%。两者都不是在最高价卖出的交易收益。最大浮亏按收盘价计算,没有冒充盘中最大回撤。路径缺价的样本不计算这些路径指标,退出结果仍单独保留。
至少2次组只有54笔完成交易,原始胜率46.30%;0次组为38.57%,1次组为37.80%。对照后两边均为57.14%,且多次组平均净收益2.57%,对照为17.51%。这说明原始组间差异还不能归因于吸筹次数。当前完整历史最多观察到3次,20日内没有至少2次的合格下跌样本,短窗口检验暂时没有数据。
## 费用与持续天数的敏感性
| 比较 | 完成配对 | 每边0.3%胜率差 | 每边0.6%胜率差 | 每边0.6%平均收益差 |
| --- | --- | --- | --- | --- |
| 上方稀疏 | 21 | +14.29个百分点 | +9.52个百分点 | +3.89个百分点 |
| 上方双峰断档 | 6 | +0.00个百分点 | +0.00个百分点 | -10.61个百分点 |
| 下跌中至少2次吸筹 | 14 | +0.00个百分点 | -7.14个百分点 | -14.85个百分点 |
| 再控制吸筹天数 | 3 | +66.67个百分点 | +66.67个百分点 | +20.98个百分点 |
| 只看结构未失效 | 7 | +0.00个百分点 | -14.29个百分点 | -4.90个百分点 |
| 改看此前20日次数 | 0 | 样本不足 | 样本不足 | 样本不足 |
“再控制吸筹天数”要求两只股票的明确吸筹天数相差不超过5天。它用于区分独立轮次与持续时间。如果配对不足,不能解释为次数没有作用。“结构未失效”只使用信号日已经知道的失效与空耗标记,不按未来是否破位筛选。
再控制持续天数后的66.67个百分点差异只来自3组配对,不作为正面证据。每边成本提高至0.6%后,下跌多次吸筹主比较的胜率差由0变为负7.14个百分点。
## 两类筹码断档
现价上方稀疏使用现价上方2%至12%的固定区间,区间筹码不超过全部筹码的5%。这也会包含全部筹码都在现价下方的情况,因此匹配时额外控制上方总量。
双峰断档使用现价70%至130%的搜索范围,按现价1%的宽度分档。三点平滑只用于定位峰,两侧峰附近各需至少5%的筹码。峰间连续低密度区至少宽5%,原始分档密度不高于较弱峰的四分之一。价格档过粗则不判定双峰。主检验使用下沿至少高于现价2%的上方断档;跨越现价、位于现价下方以及两种形态重合的结果另列在[全部分组表](group_results.csv)。
总筹码质量应处在98%至102%之间,再归一为1。缺失、负数、重复价格档和质量不合格均保留原因,不填成“没有断档”。13247个观察样本具有合格分布,其中上方稀疏1488个,上方双峰断档135个。
其中3485个样本的档位较粗,不能判断双峰;上方稀疏组仍有209个粗档样本,固定区间占比也存在离散误差。分档估算的获利比例与gp_chip_data摘要相差超过5个百分点的有579个,不能混用两种口径。5月22日请求327只股票仅有2只股票返回分档,10月16日请求307只仅返回265只。缺失记录按预定规则排除筹码比较,保留吸筹比较,不根据其收益决定去留。
![两类断档的实际样本](figures/两种断档示例.png)
示例按日期和股票代码选择,选例过程不读取收益。筹码字段的价格和百分比含义可参照[Tushare筹码分布说明](https://tushare.pro/document/2?doc_id=294)。该类分布使用交易价量与衰减模型估计,研究中的“筹码移动”不等于观测到了真实账户持仓迁移。
## 同时出现断档和多次吸筹
以下只展示下跌样本中的联合分组,未进行配对调整。
| 断档类型 | 断档 | 此前60日吸筹 | 观察数 | 完成交易 | 净胜率 |
| --- | --- | --- | --- | --- | --- |
| 上方稀疏 | 无 | 0或1次 | 4110 | 2736 | 39.66% |
| 上方稀疏 | 无 | 至少2次 | 92 | 43 | 55.81% |
| 上方稀疏 | 有 | 0或1次 | 208 | 50 | 26.00% |
| 上方稀疏 | 有 | 至少2次 | 4 | 2 | 0.00% |
| 上方双峰断档 | 无 | 0或1次 | 2806 | 1769 | 36.74% |
| 上方双峰断档 | 无 | 至少2次 | 57 | 29 | 44.83% |
| 上方双峰断档 | 有 | 0或1次 | 86 | 62 | 25.81% |
| 上方双峰断档 | 有 | 至少2次 | 1 | 0 | 样本不足 |
上方稀疏叠加至少2次吸筹只有2笔完成交易;上方双峰叠加至少2次只有1个观察且尚未成熟。当前无法检验组合是否比单独指标更强。跨越现价的双峰断档有3笔完成交易,位于现价下方的有71笔,结果分别列在全部分组表,不合并成上方阻力断档。
## 数据覆盖与真实性
| 数据 | 实际覆盖 | 本轮用途 |
| --- | --- | --- |
| gp_stock_trap_pressure | 2026-09-10至2026-10-29,26日,135,532行 | 核对现存套牢指标;没有满60日的直接检验样本 |
| accumulation_daily | 2026-09-04至2026-10-29,33日,184,323行 | 核对版本与数据质量;不足以独立形成完整60日历史计数 |
| accumulation_phase_daily | 20250918至20261029,266批,1,466,056行 | 轮次及明确吸筹天数;252个回放日、14个实际运行日 |
| 2025年筹码分档 | 2025年11月10日至20日,仅9日 | 不能把年度表名当全年连续覆盖 |
| 2026年筹码分档 | 2026年1月5日至10月29日,197个数据日 | 从实际存在的日期读取所需股票,计算两类断档 |
![数据覆盖](figures/数据覆盖.png)
抽样在查看收益之前固定。股票代码经SHA256分成14组,轮流进入每周最后一个交易日的观察;同股间隔至少60个市场交易日。共有14,595个合格观察、10,231个成熟观察,4,364个尚未走完60日。最后一个成熟信号日为2026-07-24。
吸筹阶段的历史回放于后来生成,筹码历史也存在回填。本轮合格筹码样本中,记录创建日期晚于数据日期的有1,305个。创建日期相同也不能证明收盘前已可用。历史行业记录按数据日期向后接续,但有302个观察的行业记录在事后写入。以上结果因此采用次日开盘的历史研究假设,不冒充当年实际可交易的完整信息集。
accumulation_daily同时存在fund_flow_v1的50171行和fund_flow_v2的134152行。本轮没有混合这两个版本的分数阈值,而是固定使用phase-v1-20261012代次的事件识别日期和状态。吸筹评分本身包含价格低位信息,因此匹配另控制过去跌幅和价格位置;不能把评分与后续反弹的关系全部解释为资金吸筹的作用。
价格面板覆盖5249只沪深股票。缺上市日期的代码有84只;历史名称缺失的股票日有3,087个,均按协议排除。退市覆盖和历史修订未完全认证。北交所未纳入,整手、最低佣金、容量与真实撮合未模拟。
## 套牢盘与吸筹的关系
历史套牢压力使用筹码分档直接计算的上方占比及上方15%内占比。没有把它伪装成gp_stock_trap_pressure的历史tsp评分,也没有回写该表。与现存表重合的2276个股票日对照见[套牢指标核对](trap_reconciliation.csv)。近端与远端定义分别比较,不能仅凭ocr_115的名字猜测含义。
进一步对2286条原始分档核对,ocr_above等于价格严格高于压力表收盘价的原始筹码百分比之和;ocr_115等于价格大于或等于该收盘价115%的百分比之和。两者最大绝对误差小于0.000000001个百分点。因此ocr_115对应远端套牢盘,并非上方15%以内。研究特征按总质量归一,边界也另有定义,不能直接冒充表内字段。两处收盘价格有2条不一致,最大差0.21元;这些差异见[数据口径诊断](data_diagnostics.json)。
在11908个可比较样本中,还将20日前筹码价格调整到今天的同一复权坐标,以今天收盘价作为两期共同门槛,比较上方占比变化。解释变量来自更早20日的吸筹次数,控制本段涨跌幅、市值、波动和换手。描述性残差相关为-0.01613456492755954,可用样本为11908;它不能解释为因果作用。当前筹码形态与吸筹次数的关系详见[结果文件](results.json)中的feature_relationships。
固定价格门槛下的相关约为负0.016;当期次数与上方筹码总量的控制后相关约为负0.006。两者都接近零,本轮未看到“吸筹越多,套牢盘就明显越少”的清楚线性关系。这不排除其他非线性关系,但本轮不会继续试阈值寻找正面结果。
## 验证与复现
开发机通过隔离配置和断网测试。500条真实观察独立复算历史次数、买卖索引和费用;97组配对逐项复核门槛。分组数量守恒,特征摘要在收益计算后保持不变。trade_cal为空,日期轴已与上证、创业板、沪深300指数交叉核对;最后4个日期由全市场日线核对,尚缺独立交易所日历验证。吸筹导出在155的一次性只读研究容器逐行校验payload摘要,并校验每个批次的汇总摘要;所有业务SQL严格单表。
10项专项测试通过,覆盖计数排除当日、缺失不填零、历史截断不变、断档尺度不变、粗档拒绝、交易时序、未成熟样本、路径先后、匹配不读收益及退化区间。另核对98项源摘要;用原始缓存重新生成筹码特征,与首次计算收益前的摘要一致。四幅研究图已检查中文、单位、样本量和注释。155一次性研究容器已回收,当前未残留本任务容器。
主入口为[协议](protocol.json)、[实现细化](implementation_notes.json)、[数据审计](audit.json)、[结果](results.json)。原始数据在raw/,中间面板在cache/,均保留在开发机且不入Git。源摘要和最终文件摘要见[清单](manifest.json)。
开发机复现使用本轮记录的Python依赖。以下命令不会访问运行数据库,要求原始缓存已存在:
```bash
python research/chip_accumulation_20261030/prepare_prices.py
python research/chip_accumulation_20261030/prepare_phase.py
python research/chip_accumulation_20261030/prepare_chips.py
python research/chip_accumulation_20261030/analyze.py
python scripts/unit_test_guard.py research/chip_accumulation_20261030/test_research.py
python scripts/unit_test_guard.py research/chip_accumulation_20261030/audit.py
python research/chip_accumulation_20261030/diagnose_data.py
python research/chip_accumulation_20261030/report.py
python research/chip_accumulation_20261030/finalize.py
```
## 未完成的验证
冻结口径后的新样本尚未走完60日,因此前瞻验证未完成。现有历史没有足够独立市场阶段,不能用大量股票行数补足。后续应继续保存不可覆盖的数据快照,在新样本成熟后按同一规则复算;若改变断档阈值或吸筹窗口,应另建实验版本。
本轮仅有源数据读取和开发机研究文件,没有修改PMS运行参数、持仓、委托或策略规则,没有部署业务服务。