Rank 213 long history review(frozen spec)

对象:Rank 213 / large-cap XS momentum × short-leg jump veto

页面目标:严格冻结当前 spec,不改 universe / formation / hold / topN / bottomN / veto / cost,只回答长历史同口径验证。

同口径锚点:reports/artifacts/optimization_loop/rank213_p2_admission_20260328/summary.json;当前 frozen spec:f64_h12_floor150_mult2p0 / 15m / 64 / 12 / top3-bottom3 / max(1.5%, 2.0×median) / 4bps×turnover_x

当前 runner 页面 · 最小可复现 spec · 短样本 performance review · honesty_audit · ffill_impact_audit · readiness_note · asof_universe_long_history_review · regime_review

1) 回测区间与数据可用性

common start candidate(latest onboard)
2025-10-24T08:30:00Z
actual common start
2025-10-24T08:30:00Z
actual common end
2026-04-10T23:45:00Z
bars
16190
calendar days
168.64d
rebalances
1343
这页先回答最重要的 honesty 问题:在不改当前 30 币 universe 的前提下,公共共同历史被最新上市的成员截断;因此这次 long-history review 的上限由 universe 数据可用性决定,不是我自由挑窗口。

限制共同历史的最新上市成员(前 5)

symbolonboard_date_utcsource
ONUSDT2025-10-24T08:30:00Zcached_exchangeInfo_snapshot_2026-04-11
RIVERUSDT2025-10-17T14:45:00Zcached_exchangeInfo_snapshot_2026-04-11
QUSDT2025-09-02T07:30:00Zcached_exchangeInfo_snapshot_2026-04-11
CUSDT2025-07-15T09:00:00Zcached_exchangeInfo_snapshot_2026-04-11
HYPEUSDT2025-05-30T10:30:00Zcached_exchangeInfo_snapshot_2026-04-11

2) plain baseline 与 veto 版:full available history

plain net mean
1.33 bps
plain net cumulative
-22.11%
plain win rate
49.74%
plain max drawdown
-79.94%
plain avg turnover
1.000x
veto net mean
8.96 bps
veto net cumulative
166.72%
veto win rate
49.74%
veto max drawdown
-36.82%
veto avg turnover
1.282x

直接回答“长历史下有多强”:看的是这里这组 full available history 指标;如果 1Y/2Y/3Y 不可用,就不能把下面的 full-history 数字偷换成 1Y/2Y/3Y 结论。

3) 1Y / 2Y / 3Y 冻结窗口验证

windowavailablerequired_start_utcavailable_start_utcplain_net_mean_bpsplain_net_cum_pctplain_win_rateplain_max_dd_pctplain_avg_turnover_xveto_net_mean_bpsveto_net_cum_pctveto_win_rateveto_max_dd_pctveto_avg_turnover_xdelta_net_mean_bpsdelta_net_cum_pctnote
1Yno2025-04-10T23:45:00Z2025-10-24T08:30:00Zcommon-history shorter than requested window under frozen universe
2Yno2024-04-10T23:45:00Z2025-10-24T08:30:00Zcommon-history shorter than requested window under frozen universe
3Yno2023-04-11T23:45:00Z2025-10-24T08:30:00Zcommon-history shorter than requested window under frozen universe
这一节严格回答:在当前 frozen universe/spec 下,1Y / 2Y / 3Y 到底能不能测。不可用就直接记 no,不为了凑长度改 universe 或重调参数。

4) baseline vs veto 的增量是否稳定

Δ net mean(full history)
7.64 bps
Δ net cumulative
188.83%
Δ win rate
0.00%
Δ avg turnover
0.282x
MDD reduction
43.12%

half-split stability

halfstart_utcend_utcrebalancesplain_net_mean_bpsplain_net_cum_pctveto_net_mean_bpsveto_net_cum_pctdelta_net_mean_bpsdelta_net_cum_pct
12025-10-25T00:30:00Z2026-01-16T21:30:00Z672-2.98 bps-27.81%3.33 bps14.25%6.31 bps42.06%
22026-01-17T00:30:00Z2026-04-10T18:30:00Z6715.64 bps7.90%14.61 bps133.45%8.97 bps125.55%
half-split 只用于回答“全可用历史内部前后半是否一致方向”,不是新调参入口。

5) 按月 / 按季度 / 按年表现

按月

monthrebalancesplain_net_mean_bpsplain_net_cum_pctplain_win_rateplain_max_dd_pctplain_avg_turnover_xveto_net_mean_bpsveto_net_cum_pctveto_win_rateveto_max_dd_pctveto_avg_turnover_x
2025-10565.58 bps2.00%55.36%-15.37%1.000x3.26 bps0.96%51.79%-14.16%1.310x
2025-11240-4.07 bps-14.68%48.33%-34.10%1.000x3.78 bps4.68%49.17%-22.02%1.212x
2025-122486.72 bps14.21%48.39%-25.44%1.000x9.02 bps21.92%46.37%-20.85%1.276x
2026-01248-10.76 bps-28.08%50.81%-30.92%1.000x4.92 bps9.86%52.02%-15.16%1.272x
2026-02224-35.01 bps-60.24%48.21%-64.36%1.000x-8.05 bps-19.61%47.32%-26.12%1.214x
2026-0324830.03 bps94.61%51.21%-25.95%1.000x21.91 bps63.01%52.42%-22.32%1.352x
2026-047948.68 bps40.81%50.63%-18.27%1.000x48.89 bps43.78%51.90%-7.62%1.498x

按季度

quarterrebalancesplain_net_mean_bpsplain_net_cum_pctplain_win_rateplain_max_dd_pctplain_avg_turnover_xveto_net_mean_bpsveto_net_cum_pctveto_win_rateveto_max_dd_pctveto_avg_turnover_x
2025Q45441.84 bps-0.61%49.08%-36.33%1.000x6.12 bps28.85%48.16%-22.02%1.251x
2026Q1720-4.25 bps-44.35%50.14%-74.66%1.000x6.74 bps43.97%50.69%-36.82%1.281x
2026Q27948.68 bps40.81%50.63%-18.27%1.000x48.89 bps43.78%51.90%-7.62%1.498x

按年

yearrebalancesplain_net_mean_bpsplain_net_cum_pctplain_win_rateplain_max_dd_pctplain_avg_turnover_xveto_net_mean_bpsveto_net_cum_pctveto_win_rateveto_max_dd_pctveto_avg_turnover_x
20255441.84 bps-0.61%49.08%-36.33%1.000x6.12 bps28.85%48.16%-22.02%1.251x
20267990.98 bps-21.63%50.19%-74.66%1.000x10.90 bps107.00%50.81%-36.82%1.303x

6) 不同年份里 veto 是改善 short leg 还是只是降频

定义:按每次换仓 gross return 线性分解,long leg = 0.5 × mean(long future return),short leg = 0.5 × mean(-future return of selected shorts)。这一页不讨论新想法,只看同一 frozen spec 下,不同年份里 veto 是否把 short leg contribution 往正方向推,以及它是否只是通过降低换手“看起来更好”。
yearrebalancesplain_short_leg_mean_gross_bpsveto_short_leg_mean_gross_bpsdelta_short_leg_bpsplain_avg_turnover_xveto_avg_turnover_xdelta_turnover_xpct_rebalances_with_any_veto
2025544-3.30 bps1.98 bps5.28 bps1.000x1.251x0.251x55.88%
2026799-2.98 bps8.16 bps11.13 bps1.000x1.303x0.303x62.83%
如果某年 delta_short_leg_bps > 0delta_turnover_x 并没有显著下降,说明 veto 更像是在改善 short leg,而不只是靠“少动”拿表面分数。

7) veto 命中率与误杀率

name-level 命中:被 veto 的 plain-short 候选,在该持有窗里如果原本做空它会亏钱(short PnL < 0),记为 hit; name-level 误杀:被 veto 的 plain-short 候选,在该持有窗里如果原本做空它会赚钱(short PnL > 0),记为 false kill; rebalance-level outperform / underperform:仅在发生过 veto 的换仓里,看 veto gross 是否高于 / 低于 plain gross。

8) 当前 keep/live 是否还能成立

在严格冻结当前 universe/spec 的前提下,公共共同历史仅能回到 2025-10-24T08:30:00Z,不足 1Y;因此本轮不能把 Rank213 宣称为已通过 1Y/2Y/3Y 长历史验证。当前 keep/live 只能继续依赖原 admission seed 证据;这次 long-history check 的作用是确认在最大可用共同历史上同口径表现如何,而不是升级为更长历史已证实。

这句 verdict 只基于这次 frozen-spec long-history validation:不新增研究、不改参数、不扩样本后重调。