We Tried to Trade Our Own GEX Levels. Here Is What Held Up.
Our level statistics page shows that some of our levels hold well above a random baseline. A hold rate is not a trade. So we asked the harder question, with our own levels, our own costs and a proper arbiter — and published the answer, including the parts that hurt.
The short version
Fading a level at contact does not convert into a profitable rule, not even the cells with the best hold rates. What survived, in sample, is the opposite structure: continuation after price accepts beyond a battle-zone edge, traded with a wide stop, and only in the hours where the tape trends rather than chops. And "in sample" is the operative phrase: the rule is frozen and being measured forward before anyone should believe it.
What we tested
84 trading sessions, May 26 to September 4, 2026. The levels are the ones the Indicator published each minute — Put Wall, Call Wall, Gamma Flip, Battle Zone edges, Focus 1 to 3 — logged every ~36 seconds and joined onto NQ 1-minute bars. The trade rules were deliberately unflattering:
- Entry at the open of the bar after the signal. No intra-bar fills.
- Stop on the wick, not on the close. Fixed target. Forced exit 15:55 ET, time stop 240 minutes.
- 1 point of cost per trade (commission and slippage on MNQ/NQ).
- One trade per level per "arming": the level has to be left alone by 30 points before it can trigger again.
The arbiter matters more than the rules. Every result is compared with the same rule run on permuted levels — the levels of day j applied to the prices of day k, 60 draws. If a rule cannot beat levels that have nothing to do with the day, it is not the level doing the work.
Battery one: the hold-rate cells, played as fades
| Cell (hold rate on /level-stats) | Trades | Win rate | Pts / trade | Beats permuted levels? |
|---|---|---|---|---|
| Put Wall, negative gamma (67%) | 67 | 42% | −0.4 | No (p = 0.57) |
| Battle-zone floor, positive gamma (58%) | 66 | 52% | +5.6 | No — permuted levels made +11 on the same rule |
| Focus 1, negative gamma (59%) | 67 | 45% | −0.1 | No (p = 0.55) |
| Put Wall, all conditions | 307 | 56% | +2.2 | No (p = 0.22) |
| Gamma Flip, all conditions | 343 | 45% | +0.7 | No (p = 0.40) |
Why the gap with the hold rates? A "hold" on the statistics page means a 25-point bounce before a 15-point close through the level. A trade needs the target hit before the wick takes the stop, one bar late, minus costs. Wicks take stops. And the battle-zone floor in positive gamma looked good only because calm days drift up: random levels drifted up with them.
Battery two: acceptance, then continuation
The signal flips: a 1-minute close at least 10 points beyond a level, coming from the other side, traded in the direction of the break. Stop on the far side of the level, fixed target.
| Rule | Trades | Win rate | Pts / trade | t-stat |
|---|---|---|---|---|
| Both battle-zone edges, stop 40 / target 80, all hours | 304 | 47% | +4.4 | 1.1 |
| Same, entries 09:35–10:30 only | 95 | — | 0.0 | — |
| Same, entries 12:00–14:00 only | 77 | — | −8.0 | — |
| Same, entries 10:30–12:00 and 14:00–15:55 | 132 | 56% | +15.7 | 2.7 |
Positive in May, June, July and August. Bootstrap confidence interval by day: +6 to +25 points per trade. It is the best thing in the whole study — and we found the hour filter on these same sessions, which is exactly how people fool themselves. Against permuted levels with the same hour filter, p = 0.22. Better than random levels most of the time, not a proof.
The part we would rather not publish
We then ran every level family the Indicator draws through four interaction types, four exits and seven conditioning variables — 58,112 simulated trades, 3,355 cells — and replayed the whole grid on 20 sets of randomly permuted levels. The best real cell had a t-statistic of 2.7. The best cell of a random grid beat it in 14 of 20 draws. On 84 sessions, nothing in that grid is distinguishable from chance. Some things lose clearly, though: buying the Put Wall at contact (−17 pts per trade), buying an upside break of Focus 1 (−14), fading a Focus level against the day's drift (−5 to −7).
What we did with it
We froze one rule — battle-zone acceptance, the two hour windows, stop 40, target 80 — wrote every parameter down, and started measuring it out of sample from September 5. The worker now samples the published map every 10 minutes for exactly this purpose. The decision is fixed in advance: at 200 trades, the rule stays only if it clears +6 points per trade with t ≥ 1.5. The running numbers are on the level statistics page, good or bad.
The levels behind these numbers
The Indicator draws the same walls, flip, Focus levels and battle zone on your own chart, locked at each session's open. Context, never signals.
Start the 7-day free trial — $9.99/moWhy publish a negative result
Because a level that does not convert into a trade is exactly what a trader needs to know, and because a vendor that only shows its good months is marketing. Our levels are context: they describe where dealer positioning sits, and they behave differently in different regimes. The Academy teaches that reading. Nothing here is a signal, and this article is not a recommendation to trade anything.