高股息研究:读取基座分红结论,保留缺口与固定来源(台账 081)

一、复用基座已发布研究
每日独立检测只读专项摘要,冻结引用与来源。缺失、更新中、失败和月度到期分开展示,旧档案缺键继续兼容。桥不触发模型。

二、验证契约而不扩大交易资格
研究与计划契约70项禁网回归通过,六组既有脚本通过。专项有支持不改变investment_approved=false,不改变研究池准入及排序。技术详情保留在展开区。

三、保留原系统职责
ST与股息率筛选仍在桥,研究事实与结论归基座。未改变每日计划、PMS参数或执行,也未部署或启用日程。台账登记事实边界,四仓文档另行同步。
This commit is contained in:
zlt 2026-10-21 17:17:13 +08:00
parent 802d340136
commit 238b5f7319
6 changed files with 166 additions and 9 deletions

View File

@ -228,10 +228,13 @@ docker exec akg_factor_bridge python dividend_watch.py health
输出在`data/dividend_watch/`,包含不可变输入、完整研究及可读报告。
没有分红公告和支付能力专项证据时,持续性保持待核验。
不调用模型,不改每日计划,不写交易股票池或PMS。
桥不调用模型,不改每日计划,不写交易股票池或PMS。
可选读取基座已发布的分红专项研究,冻结引用、结论与材料缺口。
基座缺失或复核失败时明确显示不可用,旧档案缺键继续兼容。
专项研究支持也不形成配置批准。新增接口当前尚未部署。
详见`docs/高股息阶段检测开发规格_2026-10-21.md`和同日交接。
```bash
# 开发机 · akg-factor-bridge;预期36项通过,完全离线
PYTHONDONTWRITEBYTECODE=1 python3 -m unittest test_dividend_watch test_dividend_jobs
# 开发机 · akg-factor-bridge;预期43项通过,完全离线
PYTHONDONTWRITEBYTECODE=1 python3 -m unittest test_dividend_watch test_dividend_jobs test_dividend_research_feed
```

55
dividend_research_feed.py Normal file
View File

@ -0,0 +1,55 @@
"""接收基座已发布的专项研究;不触发研究,不改变交易候选资格。"""
import json
import re
import urllib.parse
import urllib.request
def collect(codes, cutoff, base):
from dividend_watch import instant
result = {}
ordered = sorted(set(codes))
for start in range(0, len(ordered), 200):
batch = ordered[start:start+200]
try:
url = base.rstrip('/') + '/research/dividends?' + urllib.parse.urlencode(
{'codes': ','.join(batch), 'asof': str(cutoff)})
with urllib.request.urlopen(url, timeout=15) as response:
body = response.read(8_000_001)
if len(body) > 8_000_000:
raise ValueError('研究摘要过大')
raw = json.loads(body)
if raw.get('schema_version') != 1 or instant(raw['cutoff_at']) != instant(cutoff):
raise ValueError('研究契约或截止时刻不符')
for code in batch:
value = raw['items'][code]
research = value.get('research')
if research:
if (research.get('ts_code') != code or research.get('schema_version') != 1
or research.get('kind') != 'dividend_sustainability'
or not re.fullmatch('[0-9a-f]{64}', str(research.get('ref', '')))
or research.get('conclusion') not in ('有支持', '存在压力', '证据不足')
or instant(research['cutoff_at']) > instant(cutoff)
or instant(research['generated_at']) > instant(cutoff)):
raise ValueError('专项研究身份、时间或结论异常')
for code in batch:
value = raw['items'][code]
research = value.get('research')
last = value.get('last_check') or {}
status = ('pending' if last.get('state') == 'running' else 'unavailable' if last.get('state') == 'failed'
else 'due' if value.get('review_due') and research
else 'ok' if research else 'missing')
result[code] = {'status': status, **value,
'url': base.rstrip('/') + '/research/dividends/' + code + '?ref=' + research['ref'] if research else None}
except Exception as exc:
for code in batch:
result[code] = {'status': 'unavailable', 'error_type': type(exc).__name__, 'research': None}
return result
def display(value):
value = value or {}
status = value.get('status')
if status == 'ok' and value.get('research'):
return value['research']['conclusion']
return {'pending': '专项研究更新中', 'unavailable': '专项研究暂不可用', 'due': '待月度复核'}.get(status, '待核验')

View File

@ -87,6 +87,8 @@ def collect(expected, previous=None):
if not out['factors'] or not out['names']:
raise ValueError('预期数据日行情或名称为空')
want = candidates(out)
from dividend_research_feed import collect as collect_research
out['dividend_research'] = collect_research(want, cutoff.isoformat(), config.AKG_API_BASE)
with db.akg_pg_conn() as connection:
q = connection.execute("SELECT DISTINCT ON (ts_code) ts_code,period,review_version,grades,summary,ran_at,period_fi,period_rf,n_unknown,warnings,report_key,json_key FROM company_review_runs WHERE status='ok' AND ts_code=ANY(%s) AND ran_at<=%s ORDER BY ts_code,ran_at DESC,period DESC,review_version DESC", (want, cutoff))
out['reviews'] = rows(q)

View File

@ -21,7 +21,7 @@ import unicodedata
from zoneinfo import ZoneInfo
BJ = ZoneInfo('Asia/Shanghai')
VERSION = '1'
VERSION = '2'
DEFAULT_ROOT = 'data/dividend_watch'
LABELS = {'priority': '质地较好,优先核验', 'mixed': '观点有分歧',
'risk': '已有明确风险', 'missing_review': '待补个股评析',
@ -41,7 +41,7 @@ def digest(value):
def implementation_fingerprint():
base = Path(__file__).parent
return digest({name: hashlib.sha256((base/name).read_bytes()).hexdigest()
for name in ('dividend_watch.py', 'dividend_sources.py')})
for name in ('dividend_watch.py', 'dividend_sources.py', 'dividend_research_feed.py')})
def instant(value):
@ -198,6 +198,11 @@ def assemble(source):
reason.insert(0, '需补个股多视角评析及反方观点')
if status == 'identity_unknown':
reason.append('无法确认最新名称与上市状态,暂不纳入研究池')
from dividend_research_feed import display as research_display
specialized = (source.get('dividend_research') or {}).get(code) or {}
if specialized.get('status') == 'ok' and specialized.get('research'):
reason = [x for x in reason if x not in ('普通分红与特别分红需按公告区分', '分红记录完整性及支付能力待核验')]
reason.extend(specialized['research'].get('unknowns') or [])
rows.append({'code': code, 'name': name, 'status': status,
'yield_ttm_pct': y, 'yield_last_year_pct': number(f.get('dv_ratio')),
'close': number(f.get('close')), 'quality': summary.get('overall'),
@ -209,7 +214,8 @@ def assemble(source):
'review_at': rv.get('ran_at'), 'review_ref': summary.get('review_ref'),
'review_url': rv.get('report_url'), 'report_capture': rv.get('report_capture'),
'fundamentals': facts, 'needs': reason,
'sustainability': '待核验', 'investment_approved': False})
'dividend_research': specialized,
'sustainability': research_display(specialized), 'investment_approved': False})
counts = dict(collections.Counter(r['status'] for r in rows))
high = [r for r in rows if (r['yield_ttm_pct'] or 0) >= 5]
screen = {'market_rows': len(rows), 'high_yield': len(high),
@ -246,6 +252,12 @@ def next_checks(row):
parts.append('补齐个股评析与反方观点')
if any('材料期' in n or re.search(r'需补\d{4}Q', n) or '财务指标尚未覆盖' in n for n in row['needs']):
parts.append('补齐最新财报与现金流资料')
special = row.get('dividend_research') or {}
if special.get('status') == 'ok':
parts.extend((special.get('research') or {}).get('unknowns', [])[:2])
if not parts:
parts.append('复核专项研究的支持与反方,尚未形成配置批准')
else:
parts.append('核实普通分红及其支付能力')
return ';'.join(parts) + '。'
@ -263,12 +275,12 @@ def report(result):
continue
perspectives = r['perspectives_text'] or '视角资料待补齐'
detail = esc(json.dumps({'各视角': r['perspectives'], '材料期': r['material_periods'],
'完整核查事项': r['needs'],
'完整核查事项': r['needs'], '分红专项研究': r.get('dividend_research'),
'来源': r['review_url'], '财务与分红事实': fact_summary(r['fundamentals']),
'原报告留存': {k: v for k, v in (r['report_capture'] or {}).items() if k != 'markdown'}}, ensure_ascii=False, indent=2))
content.append(f'<article data-status="{esc(r["status"])}"><h2>{esc(r["name"] or "名称待核")} '
f'<small>{esc(r["code"])}</small><b>{r["yield_ttm_pct"]:.2f}%</b></h2>'
f'<p class="tag">{esc(LABELS[r["status"]])} · 持续性待核验</p>'
f'<p class="tag">{esc(LABELS[r["status"]])} · 分红研究:{esc(r["sustainability"])}</p>'
f'<p><strong>依据</strong> {esc(r["thesis"] or "尚缺个股评析")}</p>'
f'<p><strong>反方</strong> {esc(r["antithesis"] or "尚未形成反方研究,不能视为无风险")}</p>'
f'<p><strong>待查</strong> {esc(next_checks(r))}</p>'

View File

@ -826,3 +826,12 @@
- **依据。** 10-20月表5568行中,171只滚动股息不低于5%。排除ST八菱1只,6只名称缺失,余164只仅9只有评析,其中好2、中3、差4。旧财务材料不能因新报告时间变成新材料。
- **预期。** 高股息入口、公司研究与分红持续性分开。所有持续性暂待专项核验。原211只及16只样本记录保留,079中ST仅作旧样本的安排不延续到新池。现有排序、门槛、交易参数不动。
- **复核日期。** 本地36项新增测试和63项联合禁网回归通过,原XXL13项回归通过。真实同月复查2.438秒,164家财务缓存和9份报告复用。未部署、未启用日程,季度研究和自动补齐评析未完成。
## 081 · 2026-10-21 · 分红研究沉淀基座并保留事实边界
- **日期。** 2026-10-21,北京时间。
- **改动。** 基座新增分红持续性研究数据产品。冻结材料、六项判断、支持反方、模型配置与固定引用。月度复查默认关闭,每日最多五次模型尝试。桥只读取已发布研究,ST和股息率初筛仍在桥。
- **依据。** 用户要求此项沉淀为基座数据,供其他流程复用。丽珠与海尔的真实材料仍有一季度指标、缺分红公告与冲突记录,不能用新报告时间或模型判断补齐事实。
- **预期。** 未知、失败与过期待复核明确区分。同输入复用,旧版本不改。未来核对时间不倒灌历史。专项结论不替代公司质地,也不改变候选资格、排序、计划、PMS参数或交易。
- **复核日期。** 本地基座229项、桥70项联合禁网回归及六组既有脚本通过。真实取材已冻结,尚未部署或调用真实模型。上线后另验模型语义、幂等落库、固定查询与页面。

View File

@ -0,0 +1,76 @@
"""本地:python3 -m unittest test_dividend_research_feed;HTTP全部替身。"""
import copy
import json
import unittest
from unittest.mock import patch
import dividend_research_feed as f
import dividend_watch as w
from test_dividend_watch import reviewed
CUTOFF = '2026-10-21T08:00:00+08:00'
def payload():
return {'schema_version': 1, 'cutoff_at': CUTOFF, 'items': {'600001.SH': {
'research': {'ts_code': '600001.SH', 'schema_version': 1, 'kind': 'dividend_sustainability',
'ref': 'a'*64, 'conclusion': '有支持', 'cutoff_at': '2026-10-20T20:00:00+08:00',
'generated_at': '2026-10-20T20:03:00+08:00', 'unknowns': []},
'last_check': {'state': 'success'}}}}
class Response:
def __init__(self, body): self.body = json.dumps(body).encode()
def __enter__(self): return self
def __exit__(self, *args): pass
def read(self, n): return self.body[:n]
class FeedTests(unittest.TestCase):
def collect(self, p):
with patch('urllib.request.urlopen', return_value=Response(p)) as get:
result = f.collect(['600001.SH'], CUTOFF, 'http://base')
self.assertEqual(get.call_count, 1)
self.assertIn('/research/dividends?', get.call_args.args[0])
return result
def test_read_only_fixed_research_reference(self):
x = self.collect(payload())['600001.SH']
self.assertEqual(x['status'], 'ok')
self.assertIn('?ref='+'a'*64, x['url'])
def test_unavailable_never_becomes_approved(self):
with patch('urllib.request.urlopen', side_effect=TimeoutError):
x = f.collect(['600001.SH'], CUTOFF, 'http://base')
self.assertEqual(x['600001.SH']['status'], 'unavailable')
def test_future_or_wrong_identity_rejected(self):
for k, v in [('generated_at', '2026-10-22T08:00:00+08:00'), ('ts_code', '000001.SZ'), ('ref', '../file')]:
p = payload(); p['items']['600001.SH']['research'][k] = v
self.assertEqual(self.collect(p)['600001.SH']['status'], 'unavailable')
def test_failed_refresh_does_not_claim_fresh_success(self):
p = payload(); p['items']['600001.SH']['last_check']['state'] = 'failed'
self.assertEqual(self.collect(p)['600001.SH']['status'], 'unavailable')
def test_monthly_due_does_not_claim_current_support(self):
p = payload(); p['items']['600001.SH']['review_due'] = True
row = self.collect(p)['600001.SH']
self.assertEqual(row['status'], 'due'); self.assertEqual(f.display(row), '待月度复核')
def test_missing_fields_old_archive_compatible(self):
self.assertEqual(w.assemble(reviewed())['rows'][0]['sustainability'], '待核验')
def test_research_never_changes_eligibility_or_rank(self):
raw = reviewed(); before = w.assemble(raw)
raw['dividend_research'] = self.collect(payload())
after = w.assemble(raw)
self.assertEqual(before['counts'], after['counts'])
self.assertEqual(before['rows'][0]['status'], after['rows'][0]['status'])
self.assertEqual(after['rows'][0]['sustainability'], '有支持')
self.assertFalse(after['rows'][0]['investment_approved'])
self.assertIn('分红研究:有支持', w.report(after))
if __name__ == '__main__': unittest.main()