Solorank.ai
Signal Lab · Lv3 study
Does the Lv3 sorting actually work?
Every historical downside list is rebuilt, then Lv1/Lv2/Lv3 are traded the next day at the open, costs deducted, against SPY. Judgement uses only the segment the threshold tuner had not seen. The Lv3 lists are downside forecasts traded long, exactly as the poster page presents them.
Back to Signal Lab · Track record monitor
Segments: unseen: 2026-01-20 → 2026-09-02 · tuned: 2026-09-08 → 2026-10-23 · after: 2026-10-26 → 2026-10-26 · unseen_backfill: 2026-01-20 → 2026-06-02 · unseen_live: 2026-07-24 → 2026-09-02 · ledger: 2026-10-08 → 2026-10-26
Membership correction (survivorship bias)
Membership per day is rebuilt from Wikipedia's change log. Stocks below were in the rebuilt lists but were not index members that day, so the corrected run drops them.
Dropped as non-members (symbol × list days)
FOXA ×13FOX ×13MRVL ×7FLEX ×6BLDR ×5VEEV ×3T ×3CAG ×2TTD ×1EQT ×1FDXF ×1SYF ×1
Members missing from the site table entirely
CPB · 2026-06-22 · 600 barsCTRA · 2026-05-07 · no barsDAY · 2026-02-09 · no barsEPAM · 2026-06-02 · 600 barsHOLX · 2026-04-09 · no barsILMN · 2024-06-24 · 600 barsLW · 2026-03-23 · 600 barsMOH · 2026-03-23 · 600 barsMTCH · 2026-03-23 · 600 barsPAYC · 2026-03-23 · 600 barsPOOL · 2026-06-22 · 600 bars
Delisted exits at last price (5D): Lv1 top 8: 0 · Lv2 top 8: 0 · Lv3 top 8 (poster): 0 · Lv3 (track table): 0
Missing-member bound: 183 eligible stock-days · 183 inserted picks · their 5D return -3.13% (t=-6.12)
Does each filter layer help?
Next-day open entry (tradable)
| List | Hold | Periods (unseen) | Excess vs SPY (unseen) | t-stat | Win rate (unseen) | Excess vs SPY (all) | Verdict | Excess vs SPY (live era) | t-stat | Live verdict |
|---|---|---|---|---|---|---|---|---|---|---|
| Lv1 頭 8 Lv1 top 8 |
1D | 103 | +0.08% | 0.54 | 54.4% | +0.06% | Unproven | +0.42% | 1.08 | — |
| Lv1 頭 8 Lv1 top 8 |
3D | 37 | +0.76% | 1.46 | 70.3% | +0.41% | Unproven | +1.53% | 1.07 | — |
| Lv1 頭 8 Lv1 top 8 |
5D | 23 | -0.42% | -0.46 | 56.5% | -0.53% | No edge | -1.90% | -0.74 | Not enough data |
| Lv1 頭 8 Lv1 top 8 |
10D | 13 | +0.73% | 0.39 | 53.9% | +0.56% | Unproven | -0.83% | -0.19 | — |
| Lv2 頭 8 Lv2 top 8 |
1D | 99 | +0.16% | 1.25 | 54.5% | +0.10% | Unproven | +0.58% | 1.67 | — |
| Lv2 頭 8 Lv2 top 8 |
3D | 36 | +0.82% | 1.65 | 72.2% | +0.06% | Unproven | +2.25% | 1.38 | — |
| Lv2 頭 8 Lv2 top 8 |
5D | 23 | -0.11% | -0.12 | 65.2% | -0.51% | No edge | -1.31% | -0.45 | Not enough data |
| Lv2 頭 8 Lv2 top 8 |
10D | 13 | -0.04% | -0.03 | 53.9% | -0.27% | No edge | -1.22% | -0.29 | — |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
1D | 84 | -0.01% | -0.07 | 52.4% | +0.05% | No edge | +0.47% | 1.21 | — |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
3D | 32 | +0.06% | 0.10 | 53.1% | +0.18% | Unproven | +1.70% | 1.42 | — |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
5D | 21 | +1.27% | 1.15 | 57.1% | +0.93% | Unproven | +4.44% | 2.50 | Not enough data |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
10D | 12 | +2.55% | 1.04 | 50.0% | +2.45% | Unproven | +8.82% | 1.50 | — |
| Lv3（追蹤表版） Lv3 (track table) |
1D | 53 | +0.19% | 0.69 | 52.8% | +0.19% | Unproven | +0.55% | 1.40 | — |
| Lv3（追蹤表版） Lv3 (track table) |
3D | 25 | -0.09% | -0.13 | 48.0% | +0.09% | No edge | +1.74% | 1.45 | — |
| Lv3（追蹤表版） Lv3 (track table) |
5D | 16 | +2.25% | 1.69 | 75.0% | +1.59% | Unproven | +4.23% | 1.95 | Not enough data |
| Lv3（追蹤表版） Lv3 (track table) |
10D | 9 | +0.94% | 0.34 | 55.6% | +1.22% | Not enough data | +8.06% | 1.35 | — |
Verdicts are pinned to the unseen segment at the 5-day horizon; rows show each horizon's unseen excess.
Listing-close entry (the track pages' own convention, for comparison)
| List | Hold | Periods (unseen) | Excess vs SPY (unseen) | t-stat | Win rate (unseen) | Excess vs SPY (all) | Verdict | Excess vs SPY (live era) | t-stat | Live verdict |
|---|---|---|---|---|---|---|---|---|---|---|
| Lv1 頭 8 Lv1 top 8 |
1D | 103 | +0.00% | 0.02 | 54.4% | -0.01% | Unproven | +0.20% | 0.37 | — |
| Lv1 頭 8 Lv1 top 8 |
3D | 37 | +0.46% | 0.72 | 67.6% | +0.25% | Unproven | +0.80% | 0.41 | — |
| Lv1 頭 8 Lv1 top 8 |
5D | 23 | -0.65% | -0.71 | 47.8% | -0.81% | No edge | -1.72% | -0.72 | — |
| Lv1 頭 8 Lv1 top 8 |
10D | 13 | +0.53% | 0.27 | 46.1% | +0.35% | Unproven | -0.72% | -0.19 | — |
| Lv2 頭 8 Lv2 top 8 |
1D | 99 | +0.10% | 0.58 | 57.6% | +0.07% | Unproven | +0.30% | 0.56 | — |
| Lv2 頭 8 Lv2 top 8 |
3D | 36 | +0.60% | 0.97 | 63.9% | -0.03% | Unproven | +1.46% | 0.66 | — |
| Lv2 頭 8 Lv2 top 8 |
5D | 23 | -0.33% | -0.37 | 60.9% | -0.69% | No edge | -1.24% | -0.46 | — |
| Lv2 頭 8 Lv2 top 8 |
10D | 13 | -0.23% | -0.14 | 53.9% | -0.49% | No edge | -1.04% | -0.28 | — |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
1D | 84 | -0.16% | -0.64 | 50.0% | +0.00% | No edge | +0.04% | 0.07 | — |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
3D | 32 | +0.01% | 0.02 | 50.0% | +0.18% | Unproven | +1.97% | 1.37 | — |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
5D | 21 | +1.20% | 1.04 | 66.7% | +0.90% | Unproven | +4.30% | 2.09 | — |
| Lv3 頭 8（海報版） Lv3 top 8 (poster) |
10D | 12 | +2.59% | 0.99 | 58.3% | +2.42% | Unproven | +9.22% | 1.27 | — |
| Lv3（追蹤表版） Lv3 (track table) |
1D | 53 | -0.03% | -0.08 | 54.7% | +0.07% | No edge | +0.06% | 0.11 | — |
| Lv3（追蹤表版） Lv3 (track table) |
3D | 25 | -0.14% | -0.17 | 48.0% | +0.10% | No edge | +2.00% | 1.39 | — |
| Lv3（追蹤表版） Lv3 (track table) |
5D | 16 | +1.93% | 1.36 | 75.0% | +1.39% | Unproven | +3.85% | 1.60 | — |
| Lv3（追蹤表版） Lv3 (track table) |
10D | 9 | +0.49% | 0.15 | 44.4% | +0.80% | Not enough data | +8.49% | 1.14 | — |
This entry uses the listing day's close, which is only knowable after the list is final; it is shown to explain the existing pages' numbers, not as a tradable result.
Before vs after the membership correction (5D)
| Variant | List | Periods (unseen) | Excess vs SPY (unseen) | t-stat | Excess (backfill era) | Excess (live era) | Verdict |
|---|---|---|---|---|---|---|---|
| 修正前（用 6-21 成分表） Before (Jun-21 table) |
Lv1 頭 8 Lv1 top 8 |
23 | -0.36% | -0.39 | +0.11% | -1.66% | No edge |
| 修正前（用 6-21 成分表） Before (Jun-21 table) |
Lv3 頭 8（海報版） Lv3 top 8 (poster) |
21 | +1.30% | 1.18 | +0.04% | +4.44% | Unproven |
| 修正後（當日成分） Corrected (point-in-time) |
Lv1 頭 8 Lv1 top 8 |
23 | -0.42% | -0.46 | +0.11% | -1.90% | No edge |
| 修正後（當日成分） Corrected (point-in-time) |
Lv3 頭 8（海報版） Lv3 top 8 (poster) |
21 | +1.27% | 1.15 | +0.00% | +4.44% | Unproven |
| 計埋缺失股票（上限） With missing members (upper bound) |
Lv3 頭 8（計埋缺失股票） Lv3 top 8 (with missing members) |
23 | +0.06% | 0.07 | -1.48% | +4.44% | Unproven |
The bound variant assumes members missing from the site table passed every gate we cannot rebuild (projection, score); it is an upper bound. The backfill era was computed later with look-ahead inputs; the live era was computed on the day.
Does the Lv3 rank (#1 vs #8) mean anything?
Rank buckets, next-day open
| List | Hold | Periods (unseen) | Excess vs SPY (unseen) | t-stat | Win rate (unseen) | Excess vs SPY (all) | Verdict | Excess vs SPY (live era) | t-stat | Live verdict |
|---|---|---|---|---|---|---|---|---|---|---|
| Lv3 #1–2 Lv3 #1-2 |
1D | 84 | -0.01% | -0.03 | 45.2% | +0.02% | No edge | +0.63% | 1.36 | — |
| Lv3 #1–2 Lv3 #1-2 |
3D | 32 | +0.10% | 0.13 | 43.8% | +0.17% | Unproven | +1.83% | 1.03 | — |
| Lv3 #1–2 Lv3 #1-2 |
5D | 21 | +0.54% | 0.42 | 57.1% | +0.16% | Unproven | +2.84% | 1.32 | — |
| Lv3 #1–2 Lv3 #1-2 |
10D | 12 | +1.63% | 0.64 | 50.0% | +1.72% | Unproven | +5.86% | 0.78 | — |
| Lv3 #3–4 Lv3 #3-4 |
1D | 53 | +0.06% | 0.24 | 56.6% | +0.21% | Unproven | +0.78% | 1.53 | — |
| Lv3 #3–4 Lv3 #3-4 |
3D | 23 | -0.32% | -0.63 | 52.2% | +0.51% | No edge | -0.44% | -0.45 | — |
| Lv3 #3–4 Lv3 #3-4 |
5D | 16 | -0.14% | -0.12 | 56.3% | +1.42% | No edge | -0.09% | -0.03 | — |
| Lv3 #3–4 Lv3 #3-4 |
10D | 10 | +0.77% | 0.40 | 70.0% | +2.14% | Unproven | -0.64% | -0.10 | — |
| Lv3 #5–8 Lv3 #5-8 |
1D | 27 | +0.17% | 0.72 | 63.0% | +0.27% | Unproven | +0.59% | 0.85 | — |
| Lv3 #5–8 Lv3 #5-8 |
3D | 12 | -0.56% | -0.66 | 58.3% | -0.17% | No edge | -1.06% | -0.51 | — |
| Lv3 #5–8 Lv3 #5-8 |
5D | 9 | -1.22% | -0.93 | 44.4% | -0.01% | Not enough data | -2.28% | -0.56 | — |
| Lv3 #5–8 Lv3 #5-8 |
10D | 6 | -1.34% | -0.62 | 50.0% | +1.00% | Not enough data | -5.16% | -1.73 | — |
If the rank works, #1-2 should beat #5-8. The IC table below tests the same question day by day.
| Hold | Days (unseen) | Rank IC (unseen) | t-stat | Rank IC (all) | t-stat |
|---|---|---|---|---|---|
| 1D | 42 | -0.001 | -0.02 | -0.006 | -0.09 |
| 3D | 42 | -0.116 | -1.71 | -0.063 | -1.00 |
| 5D | 42 | -0.125 | -1.80 | -0.117 | -1.85 |
| 10D | 42 | -0.187 | -2.76 | -0.185 | -2.87 |
IC is the daily Spearman correlation between Lv3 rank (better rank = higher quality) and the forward return. Positive IC means #1 really does beat #8.
Would sorting the Lv3 pool by another field do better?
| Field | Hold | Field IC (unseen) | t-stat | Top-3 minus pool (unseen) | t-stat | Top-3 excess vs SPY (unseen) | Significant? |
|---|---|---|---|---|---|---|---|
| 投影中位數（現行） Median (current) |
1D | — | — | — | — | — | — |
| 投影中位數（現行） Median (current) |
3D | — | — | — | — | — | — |
| 投影中位數（現行） Median (current) |
5D | — | — | — | — | — | — |
| 投影中位數（現行） Median (current) |
10D | — | — | — | — | — | — |
| P25 回報 P25 |
1D | — | — | — | — | — | — |
| P25 回報 P25 |
3D | — | — | — | — | — | — |
| P25 回報 P25 |
5D | — | — | — | — | — | — |
| P25 回報 P25 |
10D | — | — | — | — | — | — |
| Dn−Up Dn-Up |
1D | — | — | — | — | — | — |
| Dn−Up Dn-Up |
3D | — | — | — | — | — | — |
| Dn−Up Dn-Up |
5D | — | — | — | — | — | — |
| Dn−Up Dn-Up |
10D | — | — | — | — | — | — |
| 下跌比 Down rate |
1D | — | — | — | — | — | — |
| 下跌比 Down rate |
3D | — | — | — | — | — | — |
| 下跌比 Down rate |
5D | — | — | — | — | — | — |
| 下跌比 Down rate |
10D | — | — | — | — | — | — |
| 總分 Total score |
1D | — | — | — | — | — | — |
| 總分 Total score |
3D | — | — | — | — | — | — |
| 總分 Total score |
5D | — | — | — | — | — | — |
| 總分 Total score |
10D | — | — | — | — | — | — |
| 動能分 Momentum score |
1D | — | — | — | — | — | — |
| 動能分 Momentum score |
3D | — | — | — | — | — | — |
| 動能分 Momentum score |
5D | — | — | — | — | — | — |
| 動能分 Momentum score |
10D | — | — | — | — | — | — |
| 60日回撤 60D drawdown |
1D | — | — | — | — | — | — |
| 60日回撤 60D drawdown |
3D | — | — | — | — | — | — |
| 60日回撤 60D drawdown |
5D | — | — | — | — | — | — |
| 60日回撤 60D drawdown |
10D | — | — | — | — | — | — |
| 120日回報 120D return |
1D | — | — | — | — | — | — |
| 120日回報 120D return |
3D | — | — | — | — | — | — |
| 120日回報 120D return |
5D | — | — | — | — | — | — |
| 120日回報 120D return |
10D | — | — | — | — | — | — |
Eight fields are tested at once, so significance requires |t| > 2.7 on the unseen segment. Field directions were fixed before running this study.
Latest Lv3 list (for reference only)
2026-10-26
Frozen published lists (ledger)
Days frozen: 13 (2026-10-08 → 2026-10-26)
Each day's published list is frozen the first time it is captured, so future runs stop depending on recomputed history. A first clean 5-day verdict needs about 40 frozen trading days.
Caveats
- Point-in-time membership is rebuilt from Wikipedia's constituent change log; ticker renames without an add/remove row (ECHO/SATS, SKYD/PSKY) are mapped by hand and the log itself may contain errors.
- Ten members missing from the site table cannot have their projection or score rebuilt; the bound variant assumes they passed every gate we cannot check, so it is an upper bound, not an estimate.
- Scores before the live start date were backfilled afterwards with the latest fundamentals, catalysts, sentiment and market regime, so the backfill segment carries look-ahead bias that membership correction cannot remove. Read the live segment on its own.
- Historical lists are still recomputed by the site until the daily ledger has frozen them; the ledger segment is the only part guaranteed to match what was published.
2026-10-27T03:03:38.479Z