生成时间：2026-03-19 09:34 UTC ｜ 最小 clean replication：固定复用 BTC/ETH/SOL 120d 15m 本地 cache；只测试 shared admission layer，不扩写新框架；统一执行 signal 当根及之前数据 + next-bar open + no-overlap + hold 8 bars。
先核对 desk 状态：全 desk 仍无 due-now / overdue；最近 due 点仍是 美股 1d+1wk（SPY/QQQ/AAPL） -> 2026-03-19 20:00 UTC。 manual narrow-paper 最新 refresh @ 2026-03-19T09:06:27Z，new_closed_trades_appended=0。
VPIS 口径冻结：VPIS = thrust(close-open vs ATR) + close_efficiency(price location × rvol_z) - absorption_penalty(wick × rvol_z)。single_volume_gate 只保留 rvol_z > 0；interaction_admission 用 trailing q60 做 through-pass；interaction_sizing 再把 q60~q80 分成 half/full，明确不把它偷渡成独立 alpha。
6bps/side desk 级结果：
baseline：mean total return ≈ -1.97%，mean avg net ret ≈ -0.065%，3-bar flip-to-fail ≈ 43.53%single_volume_gate：mean total return ≈ -1.97%，avg net ret ≈ -0.065%，retention ≈ 100.00%interaction_admission：mean total return ≈ -1.40%，avg net ret ≈ -0.052%，retention ≈ 93.55%interaction_sizing：mean total return ≈ -1.35%，avg net ret ≈ -0.045%，retention ≈ 93.55%，mean size ≈ 0.97x，flip-to-fail ≈ 41.13%Hard verdict：park / evidence_pool。这轮最小 clean replication 没证明 VPIS interaction 真能稳定优于单纯 volume gate 或 baseline：要么收益不更好，要么只是靠砍交易数换来。当前更诚实的结论是 park。
| variant | mean_total_return | positive_cell_ratio | mean_trades | mean_trade_count_retention | mean_avg_net_ret | mean_flip_to_fail_3bars_rate | mean_mae_3bars | mean_size |
|---|---|---|---|---|---|---|---|---|
| baseline | -1.97% | 33.33% | 22.0 | 100.00% | -0.06% | 43.53% | -0.64% | 1.00 |
| interaction_admission | -1.40% | 44.44% | 20.8 | 93.55% | -0.05% | 41.13% | -0.65% | 1.00 |
| interaction_sizing | -1.35% | 55.56% | 20.8 | 93.55% | -0.05% | 41.13% | -0.62% | 0.97 |
| single_volume_gate | -1.97% | 33.33% | 22.0 | 100.00% | -0.06% | 43.53% | -0.64% | 1.00 |
| setup | variant | mean_total_return | positive_asset_ratio | mean_trades | mean_trade_count_retention | mean_avg_net_ret | mean_flip_to_fail_3bars_rate | mean_size |
|---|---|---|---|---|---|---|---|---|
| breakout_short | baseline | -3.55% | 0.00% | 20.3 | 100.00% | -0.20% | 44.57% | 1.00 |
| breakout_short | interaction_admission | -3.12% | 0.00% | 17.7 | 83.55% | -0.20% | 39.25% | 1.00 |
| breakout_short | interaction_sizing | -2.77% | 0.00% | 17.7 | 83.55% | -0.18% | 39.25% | 0.92 |
| breakout_short | single_volume_gate | -3.55% | 0.00% | 20.3 | 100.00% | -0.20% | 44.57% | 1.00 |
| ema_psar_long | baseline | -3.55% | 33.33% | 34.7 | 100.00% | -0.10% | 37.54% | 1.00 |
| ema_psar_long | interaction_admission | -2.27% | 66.67% | 33.7 | 97.09% | -0.06% | 35.67% | 1.00 |
| ema_psar_long | interaction_sizing | -2.54% | 66.67% | 33.7 | 97.09% | -0.07% | 35.67% | 0.99 |
| ema_psar_long | single_volume_gate | -3.55% | 33.33% | 34.7 | 100.00% | -0.10% | 37.54% | 1.00 |
| fib_retest_long | baseline | 1.18% | 66.67% | 11.0 | 100.00% | 0.11% | 48.48% | 1.00 |
| fib_retest_long | interaction_admission | 1.18% | 66.67% | 11.0 | 100.00% | 0.11% | 48.48% | 1.00 |
| fib_retest_long | interaction_sizing | 1.25% | 100.00% | 11.0 | 100.00% | 0.12% | 48.48% | 0.98 |
| fib_retest_long | single_volume_gate | 1.18% | 66.67% | 11.0 | 100.00% | 0.11% | 48.48% | 1.00 |