J3P Backtest — START HERE
The one-page answer: what worked, what didn't, which numbers to trust, and how to look at any single trade. Updated 2026-06-28 ET.
⚠️ Bottom line: as currently built, none of the 3 plays have a tradeable edge.
On
804 trades across 52 liquid names, with an out-of-sample check, every play loses money. The
only positive signal is
selectivity — A-grade setups are slightly positive; the low-grade firehose is the bleed.
Don't automate any of it yet — the job now is to refine, then re-test.
✅ What worked / ❌ what didn't
| Play | Trades | Win% | Avg R | Verdict |
| VWAP Pull-In (long) | 563 | 44.8% | −0.161 | ✗ loses — over-fires in ROTATION chop |
| Trend Trade (long) | 220 | 44.1% | −0.282 | ✗ loses — worst play; no continuation edge |
| Backside Scalp (short) | 21 | 42.9% | +0.018 | ~ breakeven but barely fires (21 trades) |
| By grade (across all plays) | Trades | Win% | Avg R | Verdict |
| A-grade (most selective) | 13 | 69% | +0.174 | ✓ positive — but tiny sample |
| B-grade | 67 | 45% | −0.159 | negative |
| C-grade (loosest) | 182 | 41% | −0.336 | ✗ the bleed — 86% get stopped out |
The one real lesson: selectivity wins. A-grade is positive; C-grade is a bloodbath. The fix is fewer, higher-grade trades — plus the +4σ target is a phantom (hit ≤5%) and the short side barely fires.
📂 Which numbers should I trust? (wide vs market vs SIP)
| Dataset | Names | Trades | Use it? |
| WIDE — broad liquid universe + walk-forward | 52 | 804 | ✓ THIS ONE — biggest sample, out-of-sample tested = the answer |
| MARKET — first liquid cut | 18 | 171 | superseded by WIDE (same conclusion, smaller) |
| SIP — original day-only names | 11 | 157 | ⚠️ biased (selection) — only kept to show the trap |
Wide vs Market in one line: same test, Wide is just bigger and more rigorous (52 names & a walk-forward split vs 18 names). Trust WIDE. Market & SIP were earlier, smaller steps.
🔎 Look at any individual trade
Open the
interactive viewer → pick a symbol (or filter by play/grade/regime) → click any trade. You'll see the chart with entry/stop/target/exit marked (price + shares), plus a panel that explains
why it triggered and
why it exited. Hover any marker for a quick tooltip.
About "scale-out" trades: a long can show entry > the runner's exit and still be a win — because 80% was already sold at the +2σ target (above entry) and only the 20% runner came back. The viewer now shows the blended exit so wins/losses read correctly, and the trade panel spells out both legs.
What's next
- Refine to v2.3: kill the C-grade firehose, fix the phantom +4σ target, wake up the short side.
- Re-run this exact loop and compare — only promote a play to hotkeys once it's positive on the WIDE set + out-of-sample.