insufficient_data_premkt31 of 205 in-scope trades (15.1%) entered before 09:30 ET. These trades fail the v1.0-inherited session_active gate in the VCET engine and currently produce empty trajectory snaps. The simplest, least-risky Stage B path is to exclude them with a clear residual label: insufficient_data_premkt. Including pre-market in VWAP/SD computation has known data-quality and behavior issues (thin volume, wide spreads, gappy bars) and would require engine work that's out of scope for Stage A.
Pre-market net P&L is +$113.29 across 31 trades, but +$199.07 of that comes from a single trade (NVDA 2026-06-02 05:58 LONG). Excluding that outlier, pre-market trades net -$85.78 across 30 trades. There is no statistical case for including pre-market in the mechanism work today.
| Time bucket (ET) | n | Net P&L | Notes |
|---|---|---|---|
| < 09:30 (pre-market) | 31 | +$113.29 | +$199 from NVDA 6/2; without it -$85.78 over 30 trades |
| 09:30 - 09:35 (open burst) | 27 | +$121.03 | Open hour zone |
| 09:35 - 10:00 | 65 | -$267.50 | Main morning leak |
| 10:00 - 12:00 | 59 | -$231.09 | Mid-morning leak |
| 12:00 - 14:00 | 19 | +$196.66 | Mid-day reset |
| 14:00 - 16:00 | 6 | +$13.51 | Afternoon |
| after 16:00 | 3 | +$4.83 | Edge case; very few |
| Slug | Pre-mkt trades | Earliest | Latest |
|---|---|---|---|
vwap-pull-in | 14 | 04:24:24 | 09:19:59 |
chase-extended | 17 | 05:58:09 | 09:26:40 |
vwap-pull-in -> trend | 0 | — | — |
Notably: no PI → Trend trades occur pre-market. The mechanism question (true conversion vs rotation scalp) for that bucket is unaffected by pre-market policy.
In the Stage A pilot, pre-market entries produced empty trajectory snaps. The cause is the session_active gate in compute_features():
typical = df.groupby("session")["volume"].transform("sum").shift(1).bfill()
df["session_active"] = df["cum_v"] >= (typical * cfg.min_vol_fraction) # min_vol_fraction = 0.10
This gate requires cumulative session volume to exceed 10% of the prior session's TOTAL volume before any state classification (other than ROTATION with reason SESSION_INACTIVE) is allowed. Pre-market typically takes hours to accumulate 10% of an RTH session's volume, so the gate stays closed until well after most pre-market entries.
vcet_state = "ROTATION" reason_code = "SESSION_INACTIVE" direction = "" vcet_score = (low; computed from rotation formula with thin data) band_width_pct, slope_deg, etc = computed but unreliable on thin volume
The engine emits something, but the mechanism plan's downstream classifier would have no signal to work with.
session_active gate, which has downstream effects on TREND/EXPANSION detectioninsufficient_data_premktinsufficient_data as the residual; a sub-label insufficient_data_premkt fits the existing frameworkchase-extended bucket (17 of 31 pre-mkt) that's a meaningful gap| Question | Under Option B (selected) |
|---|---|
| PI → Trend mechanism count | 15 trades (0 pre-mkt; no change) |
| Plain Pull-In mechanism count | 118 trades (14 pre-mkt excluded; 89% retained) |
| Chase-extended mechanism count | 41 trades (17 pre-mkt excluded; 71% retained) |
| Total mechanism-eligible population | 174 trades (15% reduction) |
| Impact on the 6-mechanism class study | NVDA 6/2 +$199 outlier excluded (cleaner statistics); BIRD 4/15 with gap=333.7% (likely corp-action artifact) and other pre-mkt noise excluded |
| What replaces the excluded trades | Residual label insufficient_data_premkt; trade still appears in the population summary but does not contribute to mechanism classification stats |
| Date | Symbol | Time (ET) | Side | P&L | Slug |
|---|---|---|---|---|---|
| 2026-05-29 | IBIT | 04:24:24 | LONG | -$13.92 | vwap-pull-in |
| 2026-06-02 | NVDA | 05:58:09 | LONG | +$199.07 | chase-extended |
| 2026-04-24 | IBIT | 07:32:47 | LONG | -$65.50 | vwap-pull-in |
| 2026-04-29 | IBIT | 07:36:31 | LONG | +$1.08 | chase-extended |
| 2026-05-07 | AAPL | 07:56:45 | LONG | +$6.57 | vwap-pull-in |
| 2026-05-05 | NVDA | 08:06:38 | LONG | +$10.68 | vwap-pull-in |
| 2026-05-28 | META | 08:12:06 | LONG | -$47.73 | vwap-pull-in |
| 2026-05-27 | NVDA | 08:21:49 | LONG | -$100.51 | chase-extended |
| 2026-05-27 | CRWV | 08:21:56 | LONG | -$13.61 | chase-extended |
| 2026-05-05 | WULF | 08:22:44 | LONG | +$2.29 | chase-extended |
| 2026-05-28 | NVDA | 08:29:38 | LONG | +$7.70 | chase-extended |
| 2026-05-13 | SOXL | 08:30:31 | SHORT | +$57.18 | chase-extended |
| 2026-05-13 | AMZN | 08:36:54 | SHORT | -$9.11 | vwap-pull-in |
| 2026-05-12 | NVDA | 08:37:15 | LONG | +$35.87 | vwap-pull-in |
| 2026-05-08 | CRWV | 08:39:04 | LONG | -$79.53 | chase-extended |
| 2026-04-13 | IBIT | 08:54:30 | LONG | +$18.39 | chase-extended |
| 2026-04-24 | SLV | 08:58:35 | LONG | -$2.50 | vwap-pull-in |
| 2026-06-02 | QQQ | 08:59:43 | SHORT | +$10.42 | vwap-pull-in |
| 2026-05-28 | ETHU | 09:01:46 | LONG | -$5.00 | vwap-pull-in |
| 2026-05-05 | COIN | 09:05:50 | LONG | -$12.21 | chase-extended |
| 2026-05-06 | QQQ | 09:06:47 | SHORT | -$24.88 | vwap-pull-in |
| 2026-05-14 | ETHU | 09:09:01 | LONG | +$35.84 | chase-extended |
| 2026-05-06 | ETHU | 09:09:10 | LONG | -$96.01 | chase-extended |
| 2026-05-06 | MU | 09:09:21 | SHORT | -$1.43 | vwap-pull-in |
| 2026-05-07 | MSFT | 09:17:11 | LONG | +$83.95 | vwap-pull-in |
| 2026-05-28 | GOOGL | 09:19:59 | LONG | +$32.73 | vwap-pull-in |
| 2026-05-05 | INTC | 09:20:25 | LONG | +$27.61 | chase-extended |
| 2026-05-04 | IBIT | 09:22:24 | LONG | +$22.45 | chase-extended |
| 2026-05-29 | RGTI | 09:25:30 | LONG | -$23.20 | chase-extended |
| 2026-05-14 | CRWV | 09:26:29 | LONG | +$44.69 | chase-extended |
| 2026-05-29 | AVGO | 09:26:40 | LONG | +$11.91 | chase-extended |
chase-extended. A separate behavioral study on pre-market chasing could be valuable; today's tooling can't classify them mechanism-wise.insufficient_data_premkt sub-label to any DB column (Stage B / classifier work)