Measured, not marketed
Do GEX levels actually hold? We measured.
Everyone selling gamma levels claims they matter. Nobody shows you the numbers. Here are ours — 74 sessions, 17,999 intraday snapshots on NQ futures (May 26 – Sep 1, 2026), scored against a proper random-walk baseline. The misses are published with the hits.
About these numbers. They measure how price behaved around our levels over a specific past period, on one instrument, under the measurement rules described below. Past behaviour does not indicate future results, these are not trading results and imply no profitability, and nothing here is financial advice or a recommendation. Trading futures and options involves a substantial risk of loss.
Read this first: the baseline is 38%, not 50%
A "hold" here means price came within 15 NQ points of a level, then bounced at least 25 points away before closing more than 15 points through it. Because those thresholds are asymmetric, a coin-flip market with no levels at all would still "hold" about 38% of the time (the gambler's-ruin ratio, break / (react + break)). So 50% is not neutral — 38% is. A level is only informative to the extent its hold rate sits away from 38%. Below 38% is information too: the level breaks more than chance — it behaves like a magnet, not a barrier.
The four findings that matter
The strongest reaction measured (n=63, baseline 38%). When dealers are short gamma and price trades down into the Put Wall, it has bounced roughly two times in three. This nuances the common doctrine "never fade a wall in negative gamma" — the Put Wall, specifically, has been where downside moves stall. Still 54% under stricter thresholds — the signal survives.
The lower edge of the neutral zone in a long-gamma tape (n=197 — the largest reliable sample in the set): +20 points over baseline. Calm regimes have respected their floor far more often than chance.
The top-ranked confluence level (n=70) held clearly above baseline in short-gamma conditions — evidence that stacked independent readings beat any single input.
Below baseline in both parameter sets (n=36): in long-gamma conditions the third focus level has broken more often than chance — a magnet, not a barrier. We publish this because a level that doesn't hold is exactly what you need to know.
Full table — every combination with n ≥ 30
Default parameters: touch ≤ 15 pts · hold ≥ 25 pts · break > 15 pts (baseline 38%). Stricter set: touch ≤ 10 · hold ≥ 40 · break > 20 (baseline 33%). "All conditions" rows include older sessions where the gamma regime was not recorded.
| Level | Gamma regime | Touches | Hold rate | Stricter set |
|---|---|---|---|---|
| Put Wall | Negative | 63 | 67% | 54% |
| Focus 1 | Negative | 70 | 59% | 46% |
| Battle Zone low | Positive | 197 | 58% | 46% |
| Focus 2 | All conditions | 301 | 57% | 51% |
| Focus 2 | Negative | 57 | 56% | 51% |
| Focus 3 | Negative | 94 | 55% | 48% |
| Battle Zone high | All conditions | 1,500 | 54% | 48% |
| Focus 1 | All conditions | 240 | 54% | 46% |
| Battle Zone high | Negative | 310 | 53% | 45% |
| Battle Zone low | Negative | 283 | 53% | 46% |
| Focus 3 | All conditions | 258 | 53% | 48% |
| Call Wall | All conditions | 72 | 53% | 46% |
| Put Wall | All conditions | 148 | 51% | 49% |
| Gamma Flip | All conditions | 324 | 51% | 42% |
| Battle Zone low | All conditions | 1,549 | 47% | 39% |
| Focus 3 | Positive | 36 | 36% | 24% |
The pattern across the whole table: individual levels alone sit modestly above baseline (51–57%). What separates is the combination of level × regime — which is exactly how the methodology teaches these levels should be read: as context inside a regime, never as standalone triggers.
Then we asked the harder question: can these levels be traded?
Hold rates describe behaviour. They are not trades. So we ran the same levels through a trade simulator — entry at the open of the bar after the signal, stop on the wick, fixed target, forced exit 15:55 ET, 1 point of cost — across 84 sessions (May 26 – Sep 4, 2026), and judged every result against permuted levels: the levels of day j applied to the prices of day k. Four studies, one conclusion, published in full.
None of the strong hold-rate cells above — Put Wall in negative gamma, battle-zone floor, Focus 1 — beats permuted levels once you have to survive the wick, enter a bar late and pay the cost. A bounce is not a filled target.
A 1-minute close ≥ 10 pts beyond a battle-zone edge, traded in the break direction with a 40-pt stop and an 80-pt target, made +4 pts per trade over all hours — and +15.7 pts over 132 trades (56% win rate, t = 2.7) between 10:30–12:00 and after 14:00 ET. The first hour is false starts, midday is chop. The hour filter was found on the same sessions: it is measured forward, below.
Price clears a value-zone band, holds beyond it, and pulls back to a Focus level: +5.8 pts over 532 trades (t = 2.0). The width of the IV-sized zones carried no information — bands of another day's width did as well — the sequence did. Fragile to parameters, so it is measured forward too.
Buying the Put Wall at contact (−17 pts per trade, t = −2.7). Buying an upside break of Focus 1 (−14, t = −2.4). Fading a Focus level against the day's drift from the overnight anchor (−5 to −7, t = −2). Rejections worked at the top of the range, not at the bottom.
The honest verdict. We then ran every level family the Indicator draws — walls, flip, vol trigger, battle zone, clusters, focus levels, value zones, the overnight anchor — through four interaction types and four exits, 58,112 simulated trades, and replayed the entire grid on 20 sets of randomly permuted levels. The best real cell (t = 2.7) was beaten by the best cell of a random grid in 14 of 20 draws. On 84 sessions, nothing in the grid is distinguishable from chance. That is why the two rules below are frozen and measured forward instead of being sold as an edge.
Live paper track record — two frozen rules, judged at 200 trades
Both rules were written down on September 6, 2026 with every parameter fixed. From that day the worker samples the published map every 10 minutes and the rules are evaluated on NQ 1-minute bars, out of sample, with 1 point of cost per trade. The decision is fixed in advance: at 200 trades each, keep a rule only if it clears +6 points per trade with t ≥ 1.5; otherwise it is dropped and this page says so. No parameter changes along the way. No live capital.
No out-of-sample trade yet. The measurement started on September 5, 2026; the worker samples the map every 10 minutes and the two rules are evaluated on NQ 1-minute bars as sessions accumulate.
Pre-registered reads (registered 2026-09-06)
A read is kept only if its pre-registered test passes on out-of-sample data. Nothing in this file is a signal; it is the list of what we promised to measure and how. The full list with the exact rules lives in the research repository; this is the public version.
-
0DTE pinning score
Definition. 0-100 from gamma concentration within 0.5 % of spot, distance to the peak strike and to max pain, session time left, long/short gamma regime, event-risk factor (worker edge-reads.pinningScore).
Outcome. hit = close within 0.25 % of the magnet strike on the scored expiry day Test. Buckets 70+, 45-69, <45 must separate: hit rate of the top bucket at least 15 points above the bottom bucket, each bucket n >= 10.
-
Level acceptance score
Definition. Per level over the last 30 one-minute bars: 55 % time within 0.1 % of the level, 30 % volume there vs the session average, 15 % share of closes on one side (terminal acceptance widget).
Outcome. continuation of at least 40 NQ points on the accepted side within 60 minutes after the score crosses 60 Test. Expectancy of the continuation rule above the permuted-levels baseline (same arbiter as the level-stats studies), t >= 1.5 at 200 events.
-
Dealer hedging direction (own model)
Definition. Sign of the net gamma exposure across all expirations from the own model (dampening vs pro-cyclical).
Outcome. realised range of the following 60 minutes vs the day's median 60-minute range Test. Median 60-minute range under pro-cyclical at least 20 % larger than under dampening, over 60 sessions.
-
Systematic flow estimate (CTA + vol-target)
Definition. Trend score from SMA 20/60/120/250 and 1/3/6/12-month momentum, vol-target allocation = 12 % / realised 20-day vol capped at 150 %; flow = change in exposure times assumed AUM (CTA 250 B$, vol-target 300 B$ on ES).
Outcome. next-day close-to-close return of ES Test. Days with estimated flow below -20 B$ show a median next-day return at least 0.15 % below days above +20 B$, over 120 days; otherwise the AUM assumptions are dropped from the widget and only the score is kept.
-
Implied event premium
Definition. Extra move in points priced between two expirations above the quiet forward-vol baseline (desk-reads.eventMoves).
Outcome. realised move on the event day (open to close) vs the implied event premium Test. Realised / implied ratio distribution: median between 0.7 and 1.3 over 20 events means the chain's event pricing is usable as a range; outside, the premium is shown as information only.
-
OPEX cycle tendency
Definition. Average return and range by trading-day offset to the third Friday, on the bars we hold (2 years).
Outcome. n/a (descriptive) Test. Descriptive only: no claim is made. The table is shown with its sample size; no offset is promoted to a rule unless a separate pre-registered test is added here.
Why publish a track record before it exists? Because a rule that only shows its good months is marketing. Both rules sit inside what a random grid produces in-sample; the forward measurement is the only evidence that will count, and it is the only kind we will ever show.
Methodology & honest limitations
Data
74 trading sessions, 17,999 timestamped snapshots (May 26 – Sep 1, 2026), NQ futures pricing with levels captured intraday, minutes apart. Levels are the same ones the Indicator publishes — nothing was recomputed after the fact.
Scoring
A touch = price within the touch threshold of a level. A hold = a bounce of at least the react threshold before any close beyond the break threshold. A level re-arms only after price leaves twice the touch distance, so one visit is never double-counted. Every headline figure is checked under a second, stricter parameter set — a result that vanishes when thresholds tighten is not reported as a finding.
What this is not
These are descriptive statistics about price behaviour around structural levels — not trading results. No entries, exits, P&L, slippage, spreads or decision latency are modelled, and nothing here is a claim that trading these levels is or was profitable. Snapshots are minutes apart, so fast wicks between snapshots under-count touches. Sessions are autocorrelated, so we deliberately publish no p-values.
Updates
The dataset grows every session. We re-run the measurement as data accumulates and update this page — including any finding that weakens. If a number here later degrades, you will see it degrade.
The levels behind these numbers
The Indicator draws these same levels — walls, flip, focus levels, battle zone and the regime read — live on your own chart, locked at each session's open. The free Morning Map shows the previous session's core map every trading day.
See the Indicator Education only · context, never signalsEducational market research only — not financial advice, not a recommendation to buy or sell any instrument, and not a prediction of future behaviour. Hold rates describe how price historically behaved near structural levels under specific measurement parameters; they are not trading performance, imply no profitability, and past behaviour does not indicate future results. Futures and options trading involves substantial risk of loss.