Selected findings
A short, dated summary of what our research on ES, NQ and GC has found. Every number comes from a stored, logged result. Full papers with methods are in preparation.
How to read these numbers
A call on the nearer of two levels is right more often by geometry alone. With 3 points to the target and 5 points to the other level, a random walk reaches the target first 62.5% of the time, with zero edge. So we never show a hit rate on its own. Each one comes with:
- the base rate: how often the outcome happens anyway;
- the geometry reference: what distance and time alone predict for the same moments;
- the room: how far price still was from the target when the call was made.
Periods: models are fitted on 2022–2023, checked on the following years, and confirmed once on a locked year (October 2025 to September 2026) that was not looked at until the end. Since late September 2026, results are logged forward, live. Each situation was defined by a trader and checked by hand against real sessions.
The profile shape is the map, not the forecast
In three separate tests, the shape label added nothing or almost nothing once the geometry was known. One example: when price sits in the low-volume gap between two balance areas of the same session, the chance that it returns to the area it came from matches a simple random-walk formula built from the two distances almost exactly.
| Return to origin, 2022–23 | ES | NQ | GC |
|---|---|---|---|
| Observed | 0.636 | 0.606 | 0.622 |
| Distance-only formula | 0.629 | 0.619 | 0.625 |
What it means: the shape tells you where the targets are. What moves the odds is where price sits on that map and what is happening at that price.
Volume pulls price toward it
When price trades outside the value area, heavy volume at the current price (relative to the point of control) goes with a return to the point of control, and thin volume goes with escape. On ES, the held-out year showed about 26% returns in the thinnest group against 63% in the heaviest.
The same pattern appeared in the gap between two balance areas: a heavy route back to the area price came from went with more returns, about 10 points above the random walk in the heaviest fifth against about 1 point in the thinnest.
This reverses our own starting hypothesis that surplus volume outside value signals escape. We report it the way the data came out.
Headline hit rates flatter; room tells the truth
Confident "return to the point of control" calls are right about 79–82% of the time on the held-out year, but almost all of them fire next to value, where geometry already says the same thing. With real room to the target (more than 1 point on ES and GC, more than 5 on NQ) the rate is 66–68%.
Measured against distance and time at the moment each call fires, the outside-value model showed no call type with an edge beyond about +2 points. Its accuracy is real, and distance and the clock already account for it.
Where an edge survives: rare, strong calls
In the gap between two balance areas, strong calls (probability 0.7 or more) that price returns to its origin beat the geometry reference in every period tested. Compared at the same number of calls on the locked year:
| Strong return calls, locked year | ES | NQ |
|---|---|---|
| Model | 72% | 70% |
| Geometry, same number of calls | 59% | 65% |
| Edge over distance + time | +7.6 pts | +9.3 pts |
| Base rate at the setup | 58% | 54% |
These calls are rare (about 0.2 to 0.5 per session), and the ES confidence interval just touches zero. Too few GC calls to judge. The forward log will tell whether the edge holds.
What we withdrew
A broader "which level first" model appeared to show that volume along the route attracts price toward any level. A check of the live engine against the research data found that one input had used trades the session only printed later. Re-tested with honest inputs, the claim did not pass, so we withdrew it and took the model off our platform.
We include this on purpose. A research shop that never reports a mistake is not looking hard enough.
Limits
- The models estimate which level is reached first, not how far price runs afterwards.
- No untouched historical data is left. From here, only the forward log is an honest test.
- Results describe historical behaviour. They are not a forecast of future returns and not a trading system. See the disclaimer.