J3P Multi-Day Audit — EBAY · Jul 1–6, 2026
Does the J3P engine (VWAP Pull-In · Trend Trade · Backside Scalp) work inside a higher-timeframe TREND environment? · card built 2026-07-06
PARTIAL — ENGINE RUN PENDING
The J3P engine (
j3p_backtest.py) depends on
vcet_engine, which lives on the HOME machine
(
C:\Users\jtdom\Downloads\vcet_v1_1\python) — it is not installed on this workstation and no copy exists on Z: (verified).
Cells marked
pending fill from one command on the home box:
py j3p_backtest.py --symbols EBAY --end-date 2026-07-06 --days 4 --out runs/j3p_trades_EBAY_jul.csv
— then re-run
make_report.py / the viewer export for the 7/2 and 7/6 replay pages. Only the 7/1 replay
(
EBAY_2026-07-01.html) exists today; its embedded signal data is used below verbatim.
Environment (verified from daily bars)
Jul 1: base-high break, close 111.28 (red close, low-vol). Jul 2: trend day +2.95% close 114.84 on a 0.64 Daily-RVOL tape (dollar-breakout qualified). Jul 6: 0.63-RVOL narrow pause, close 113.605. Daily regime: TREND (up) — a fresh base breakout holding its gains, price above a rising 5-DMA the entire window. This is exactly the regime where the exit-experiment finding applies: trend plays need trend exits.
Per-session summary
| Session | Play | Entry | Stop | T1/T2 (+2σ/+4σ) | Outcome (60/20/20) | MA-trail alternative |
| 7/1 | vwap_pull_in (long) — signal #1 | 111.52 | 111.13 | 111.89 / 112.40 | T1 and T2 both traded through same session (high 113.57); replay page marks W +1.4R ladder* | pending engine run — intraday MA-trail needs the 2-min replay; the DAILY-scale answer is already measured: TMT full-size hold = +0.56R on 100% exposure vs actual 60/20/20-style scaling = +0.56R on 5% remaining (see the EBAY playbook Management Tracking section) — same R banked, trend exposure kept only by the trail |
| 7/1 | long — signal #2 | 113.415 | 113.00 | 113.80 / 115.29 | T1 hit 7/2; T2 hit 7/2 (high 115.98) — multi-session ladder* |
| 7/2 | vwap_pull_in continuation (long) | per Bella Vol MA trigger-window: same-bar RVOL 1.96 @10:08, 2.27 @10:28 — re-graded qualified-A | engine ladder pending | pending | pending |
| 7/2 | backside_scalp (short) — 2 signals flagged on the live overlay | screenshot-reported: "BACKSIDE short entry · R:R 4.27" markers ~13:00–13:30 | engine outcome pending | FLAG: counter-trend family inside a trend day — see verdict below |
| 7/6 | vwap_pull_in (long) — last signal on overlay | screenshot-reported: R:R 0.39 (vwap_pull_in) — sub-1 R:R late in a pause day | engine outcome pending | pending | pending |
* 7/1 rows quote the signal ladder embedded in the existing replay page's data (entries/stops/targets verbatim; target-touch dates verified against daily H/L: 7/1 H 113.57, 7/2 H 115.98). Rows marked screenshot-reported are point-in-time overlay readings from the trader's charts, not engine output — they will be replaced by the engine run.
Verdict so far (what is already answerable)
- The long families worked with the trend: both 7/1 long ladders resolved through T1 and T2 — in a trending window the 60/20/20 σ-ladder banks reliably.
- But the σ-ladder caps the trend: the measured comparison on this exact move (playbook Management Tracking) shows scaling-out and full-size MA-trailing arrived at the SAME +0.56R by 7/6 — with the trail keeping 20× the exposure to continuation. In trend regimes the 20% runner is doing all the long-tail work; a per-play MA-trail variant is the experiment the engine run should score.
- Family to flag: backside_scalp. Counter-trend shorts fired on 7/2 — a +2.95% dollar-breakout trend day. Per the exit-experiment finding (trend plays need trend exits; counter-trend plays need trend FILTERS), backside signals inside a confirmed daily TREND regime should be regime-gated. The engine's
--market-regime / --shannon-dir flags exist for exactly this test — include them in the pending run.
- Gate note: 7/2's Daily RVOL (0.64) would have skipped the day entirely under the old gate; the trigger-window gate (same-bar 1.96–2.27) correctly qualified it. The corrected RVOL gate spec is in
tmt-tracker/TMT_SPEC.md.
LAB material — partial audit; engine-faithful per-play outcomes and the 7/2 / 7/6 replay pages require the home-machine run (command above). Nothing here models slippage/commissions. Sources: EBAY_2026-07-01.html embedded signal data, kti daily bars (karb + Liberator 6/9 patch), trader-reported Bella Vol MA readings, TMT specimen #1 tracking.