Research · Display Only研究 · 仅供展示

Technical Lab技术实验室

Signal screener + descriptive backtest profiles. Build confluence filters to find names where multiple signals align. Drill into any ticker for a per-stock signal breakdown. Descriptive framing — survivor mega-cap universe, not a verdict.信号筛选器 + 描述性回测档案。构建多因子共振筛选条件，聚焦多信号同向股票。点击任意标的查看个股信号分解。描述性研究框架 — 幸存大盘股宇宙，非结论性研究。

Technical Signals技术信号 2026-07-17 02:36 UTC 64 signals信号 225 universe覆盖标的

ⓘ

Descriptive · Survivor Universe · Not a Verdict.描述性 · 幸存者宇宙 · 非结论。

Survivor mega-caps only — no delisted stocks仅含幸存大盘股，不含退市标的 Pooled stats include up-market concentration bias汇总统计含牛市集中偏差 Pre/post-2010 era split shown — pooled WR is post-2010 when pre-2010 is absent显示2010前后分期，前期缺失时汇总=2010后 Signal fires are observation counts, not trade recommendations触发次数为观察计数，非交易建议

Composite Screener 综合筛选器 score = buy signals ÷ total signals 评分 = 买入信号 ÷ 信号总数

Ticker 标的	Price 价格	Score ↓ 评分 ↓	Band 评级	7D 7日	30D 30日	Active signals 活跃信号

Combos matching now 当前匹配的信号组合

Backtested signal groups with at least one stock matching today. Click one for its full record, or build your own in the Combos tab. 回测过的信号组合，今日至少有一只股票满足条件。点击查看完整记录，或在“信号组合”页自行搭配。

Loading combos… 加载组合中…

Signal combos — groups of 2–4 signals that historically won together more often than chance.信号组合 — 由2–4个信号构成、历史上共同出现时胜率高于随机的组合。 Found by automated search, so treat them as leads to watch, not proof. Win rates count each calendar month once (clustered fires don't inflate them) and are quoted for the recent era (2018+), with a "held up in both eras" check against the years before. 由自动搜索发现，应视为值得关注的线索而非定论。胜率按日历月去重统计（密集触发不会虚增），引用2018年以来的数据，并对照更早年份做“两期均稳”检验。 ?

Combo leaderboard 组合排行榜 ?

Combo组合	Wins per 10 (2018+)每10次胜次(2018后)	vs random对比随机	Both eras?两期均稳?	Entries/yr年均触发	Fires 3y近3年触发	Matching now当前匹配
Loading…加载中…

Build your own combo 自建信号组合 ?

Loading signals…加载信号中…

—

pick 2+ signals请选择至少2个信号

Loading signals… 加载信号中…

Signal Catalog 信号目录 click a row to drill in 点击行查看详情

Signal ↓ 信号 ↓	Family 族群	Dir 方向	WR vs Base 胜率/基准	Edge 优势	Era 时代	Up Tape 上涨市占	Fires 触发
Loading… 加载中…

—

↗ Open in Terminal 在终端中打开

—

30D

—

Active Buy 活跃买入

—

Score 评分

—

buy signals active 个买入信号活跃
out of — tracked

Select a ticker to load the chart 选择标的以加载图表

Illustrative sketch — real bars unavailable 示意图 — 真实行情不可用

All Signal States 全部信号状态

Select a ticker on the right to see its signal states. 在右侧选择标的查看信号状态。

Switch Ticker 切换标的

Era-split context 时代分割背景

pre-2010 No data for most signals 多数信号缺乏数据

post-2010 All pooled stats — QE regime, rates cycle 所有汇总统计 — QE周期，利率周期

How to Test — CLI & pytest reference 如何检验 — CLI 与 pytest 参考

For LLM sessions and human reviewers. All code runs against prod-shaped fixtures under tests/fixtures/tech_parity/. Exit 0 on honest nulls. 适用于 LLM 会话和人工审核者。所有代码运行于 tests/fixtures/tech_parity/ 下的生产形态测试夹具。零触发为诚实空值，退出码为0。

▶ Signal inventory信号列表

Print the full catalog as JSON: id, family, kind, direction, bilingual names. 以 JSON 格式打印完整目录：id、族群、类型、方向、双语名称。

# list all signals
python -m scripts.tech_lab_cli list

# filter to one family
python -m scripts.tech_lab_cli list --family ichimoku

📋 Per-ticker state个股信号状态

Emit latest per-signal state for one ticker. Output: JSON dict keyed by signal_id with {state, direction, kind, display_en}. 输出单只股票的最新信号状态。输出：以 signal_id 为键的 JSON，含 {state, direction, kind, display_en}。

python -m scripts.tech_lab_cli state NVDA
python -m scripts.tech_lab_cli state AAPL | python -m json.tool

🔎 Descriptive fire profile描述性触发档案

Horizons: {10, 21, 42, 63} days only (TLT-R6). Era split (pre/post 2010) always printed. --sample N for quick runs. 窗口期：仅限 {10, 21, 42, 63} 日（TLT-R6）。始终打印时代分割（2010前/后）。--sample N 用于快速运行。

# single signal, all tickers, horizon 21d
python -m scripts.tech_lab_cli profile rsi14_oversold --horizon 21

# specific tickers, multiple horizons
python -m scripts.tech_lab_cli profile golden_star_st_7_35 \
  --tickers NVDA,AAPL,MSFT --horizon 10 --horizon 21

# quick sample of universe
python -m scripts.tech_lab_cli profile ichimoku_tk_cross_up \
  --sample 20 --horizon 21

📈 Per-bar series tail逐K信号序列

Emit the last N bars of the raw signal series for a ticker (for chart debugging and indicator parity checks). 输出指定标的的最近 N 根K线信号序列（用于图表调试和指标一致性验证）。

python -m scripts.tech_lab_cli series rsi14_oversold NVDA --tail 30
python -m scripts.tech_lab_cli series ichimoku_above_cloud AAPL --tail 50

🧩 Combo miner (multi-signal backtest)组合挖掘器（多信号回测）

Mines 2–4-signal combos across daily + weekly legs (rising-edge fires, month-collapsed win rates, 2018 train/test split). Writes site/factordata/tech_confluence.json. Full run ≈ 8 min; nightly off-render. 在日线+周线信号上挖掘2–4个信号的组合（上升沿触发、按月去重胜率、2018年训练/检验分割）。输出 site/factordata/tech_confluence.json。全量运行约8分钟，夜间离线执行。

# smoke run (12 tickers)
python scripts/build_tech_confluence.py --sample 12 \
  --output /tmp/tcf_smoke.json

# full universe (what the nightly runs)
python scripts/build_tech_confluence.py

# unit tests (synthetic fixtures, no data store needed)
python -m pytest tests/test_tech_confluence.py -q

✅ Run the test suite运行测试套件

Correctness, era-split logic, CLI tier stamps, fixture determinism, parity with Terminal TypeScript. 正确性、时代分割逻辑、CLI 层级标签、夹具确定性、与终端 TypeScript 的一致性。

# full tech lab test suite
python -m pytest tests/test_tech_stars_metrics.py \
               tests/test_tech_events_export.py \
               tests/test_tech_lab_cli.py \
               tests/test_tech_parity_fixtures.py \
               tests/test_tech_catalog.py -q

# smoke-test the export pipeline (5 tickers)
python scripts/build_tech_lab_data.py --sample 5 \
  --output-dir /tmp/te_check

⚠ Gauntlet protocol reminder检验协议提醒

All data here is display-tier. The gauntlet applies only at promotion to authority (rank/size/gate). 所有数据为展示级别。检验协议仅在晋升为权威时适用（排名/规模/门控）。

Null as standalone ≠ worthless — retain as confluence input空值单独成立 ≠ 无价值 — 保留为共振输入
Survivor bias: mega-caps only; n_months varies by signal幸存偏差：仅大盘股；覆盖月数因信号而异
Pre-2010 absent for most signals — pooled = post-2010多数信号缺乏2010年前数据 — 汇总即为2010年后
Era split required before promotion: use wr_pre2010 / wr_post2010晋升前需进行时代分割：使用 wr_pre2010 / wr_post2010
The CI guard blocks affirmative authority claims ("this signal has a gauntleted edge") without a backing artifact — do not claim gauntlet results you do not haveCI保护阻止未经认证的权威声明 — 未有检验结果时不得声明信号已通过检验