HK Pick Lab · Display-only · ACCRUINGHK选股实验室 · 仅展示 · 累积中

🧪 HK Candidate Pick Engines港股候选选股引擎

20 shadow books running on forward ledgers — measured against a random-control yardstick and the Hang Seng Index. Ruler: 21-session ^HSI excess (next HK session fill, close basis). All data is display-only; nothing here drives live selection or gating.20个影子书在前向账本上运行 — 与随机控制基准和恒生指数对比衡量。标尺:21节恒生超额(下一港股交易日入场,收盘价基准)。所有数据仅供展示;这里没有任何内容驱动实时选股或门控。

Back to HK Stock Desk返回港股研判台
🧪

First accrual tonight今晚首次累积

The HK lab runs in the asia-close lane nightly. The scoreboard will appear here after the first snapshot is taken and the first picks are fired.HK实验室在亚盘收市通道每晚运行。首次快照拍摄并触发第一批选择后,记分板将出现在这里。

First accrual tonight今晚首次累积

1D velocity picks will appear here after the first nightly build completes.首次夜间构建完成后,1日动能选择将出现在这里。

📚

First accrual tonight今晚首次累积

All 20 HK entry books will appear here after the first nightly build.首次夜间构建后,所有20本HK入场书将出现在这里。

What this HK lab measures本HK实验室测量什么

20 frozen HK candidate pick engines run in parallel on forward ledgers in the asia-close lane. Each book fires picks nightly; those picks are graded when the horizon elapses. Nothing here drives live selection. The purpose is evidence accrual: let the data tell us which HK constructions are worth a formal gauntlet.20个冻结HK候选选股引擎在亚盘通道的前向账本上并行运行。每本书每晚触发选择;当地平线经过时,这些选择被评级。这里没有任何内容驱动实时选股。目的是证据累积:让数据告诉我们哪些港股构造值得正式考验。

Ruler (HKPL-R3)标尺(HKPL-R3)

Primary ruler: 21-session ^HSI excess — win rate + median excess. 5/10/63-session descriptive ladder also recorded. MFE/MAE over 25 sessions.主标尺:21节恒生(^HSI)超额 — 胜率+中位超额。同时记录5/10/63节描述性阶梯。25节MFE/MAE。 Inverse books (Knife Avoid, Chase Avoid): scored as avoid-accuracy — expected negative 21-session excess.反向书(刀口规避、追涨规避):评分为规避准确性 — 预期21节负超额。

Execution law (HKPL-R4)执行法则(HKPL-R4)

Exec = next HK session close (fill_basis="close"). No price limits — HK has no daily limit. Board lots ignored at ledger tier (noted, not modeled).执行 = 下一港股交易节收盘价(fill_basis="close")。无涨跌停 — 港股无每日限制。账本层忽略手数(已注明,未建模)。 Halt law: if the exec session has no print, fill at the first traded session within 5 sessions, else the fire is voided (halt_voided counter). A name halted >5 consecutive sessions inside a grade window grades to last trade with halted=true — never silent ffill.停牌法则:若执行节无成交,在5节内首个有成交节入场,否则触发撤销(halt_voided计数)。评级窗口内连续停牌>5节的个股以最后成交价评级并标记halted=true — 绝不静默填充。

Organ staleness (HKPL-R7)器官陈旧度(HKPL-R7)

A book input staler than 2 HK sessions disables that book for the night (fail-closed). The disabled_stale_nights counter on the scoreboard tracks how many nights each organ-dependent book was suppressed.书的输入数据超过2个港股交易节将使该书当夜停用(故障关闭)。记分板上的disabled_stale_nights计数器跟踪每本器官依存书被抑制的夜数。

HK kill laws (spec §8)HK禁止法则(规格§8)

NO residual/selection momentum ranks. NO southbound delta-ranking (SB_ACCUM organ signal = legal). NO COILED books for HK. NO Connect-inclusion event books.无残差/选择动量排名。无南向资金增量排名(SB_ACCUM器官信号 = 合法)。无HK盘绕书。无互联互通纳入事件书。 Inverse books (Knife Avoid, Chase Avoid) are AVOID books — expected negative excess. The 20 books use beta-role, organ-state, and leverage-mechanics constructions — not momentum factors.反向书(刀口规避、追涨规避)是规避书 — 预期负超额。20本书使用贝塔角色、器官状态和杠杆机制构造 — 而非动量因子。

Controls (HKPL-R5)控制变量(HKPL-R5)

Every book is measured against two yardsticks: (1) the HK Random Control book (8 deterministic-random liquid names, seed = sha256(engine_id+asof)), visible in the scoreboard; and (2) the universe buy-anytime base rate. Every headline number is lift over both.每本书均与两个基准进行比较:(1)HK随机控制书(8个确定性随机流动名称,种子=sha256(engine_id+asof)),在记分板中可见;以及(2)全市场随时买入基准率。每个标题数字是相对于两者的提升。

ACCRUING floor (HKPL-R6)累积下限(HKPL-R6)

No verdict language before n≥25 fires spanning ≥3 calendar months and ≥6 distinct fire dates. Refire lockout: 21 sessions (HKPL-R6). Max 8 picks/day (small HK universe ~160 names).在n≥25次触发跨越≥3个日历月和≥6个不同触发日期之前,无结论语言。重触锁定:21节(HKPL-R6)。每日最多8个选择(港股小宇宙约160只)。 Until then, the status badge reads ACCRUING.在此之前,状态徽章显示ACCRUING。

No Long-Hold grid (HKPL-R9)无长持网格(HKPL-R9)

HK long-hold grids are deliberately deferred: the H3/H2a edges are already accruing under their own programs. Revisit only with dedicated pre-registration.HK长持网格被故意推迟:H3/H2a优势已在各自程序下累积。仅凭专用预注册方可重新审视。

Candidate families候选族

  • A — 1D Velocity:1日速度: 5 books — 1D MACD/StochRSI cross-up entry timing; cross×washout organ; ADR overnight confirmation; blastoff (3D gate miss); regime (Risk-on + peg ok).5本书 — 1日MACD/StochRSI交叉上穿入场时机;交叉×洗盘器官;ADR隔夜确认;起飞(3日门遗漏);状态(风险偏好开启+联系汇率正常)。
  • B — Washout/Ignition (organ):洗盘/点火(器官): 5 books — ignition_watch strongest state; washout×SB_ACCUM confluence; washout×buyback; pullback second-chance entry; Knife Avoid INVERSE (knife_risk=True).5本书 — ignition_watch最强状态;洗盘×南向积累汇流;洗盘×回购;回调第二次机会入场;刀口规避反向(knife_risk=True)。
  • C — HK-unique structure:HK特有结构: 4 books — CBBC bear-skew forced-buy fuel; A/H discount value (H3 near-GO); short squeeze (SFC pressure+RSI reclaim); catalyst+narrative attention shock.4本书 — CBBC空头偏斜强制买入燃料;A/H折价价值(H3接近通过);轧空(SFC压力+RSI收复);催化剂+叙事关注冲击。
  • D — Beta/Regime:贝塔/状态: 3 books — amplifiers on Risk-on; cushions on Risk-off; HIBOR EASY liquidity + washout_2w rebound.3本书 — 风险偏好开启时的放大器;风险回避时的缓冲;HIBOR宽松流动性+2周洗盘反弹。
  • E — Ablations + Controls:消融+控制: 3 books — flagship no-gate ablation (edge_z, no signal-gate grouping); Chase Avoid INVERSE (chase_risk state, RSI≥70 gapped); HK Random Control yardstick.3本书 — 旗舰无门消融(edge_z,无信号门分组);追涨规避反向(chase_risk状态,RSI≥70跳空);HK随机控制基准。

Clocks时钟

  • 2026-08-20 — first operator read; 1D-vs-stale-cross diagnostic first look; prune degenerate books.2026-08-20 — 第一次操作员读取;1日vs迟滞交叉诊断首次查看;修剪退化书本。
  • 2026-10-09 — floor-eligible verdict window (n≥25/≥3mo/≥6 dates) for high-frequency books.2026-10-09 — 高频书的达标评判窗口(n≥25/≥3月/≥6日期)。
  • 2027-01 — H3 come-back clock; hklab_ah_value re-baselines if H3 promotes/changes.2027-01 — H3回归时钟;H3晋升/变更时hklab_ah_value重新基准。

Authority: display_only. All rows carry authority: "display_only" per HKPL-R1.授权:仅展示。所有行按照HKPL-R1携带授权:"display_only"。 · Built生成时间 2026-07-09 14:35 UTC