Where an edge survives
Two narrow places beat distance and time on the held-back year: the rare, strong return calls between two balance areas, and calls outside value made with real room to the target in some markets. Both are now being tested forward.
The question
Once every call is compared with what distance and time already say at the moment it fires, is anything left?
Between two balance areas: rare, strong return calls
The model's most confident calls that price returns to the area it came from beat the distance-and-time reference in every period tested: by 8 to 19 points in the test years in all three markets, and on the held-back year:
| Strong return calls, held-back year | ES | NQ |
|---|---|---|
| Model, right | 72% | 70% |
| Geometry, same number of calls | 59% | 65% |
| Base rate at the setup | 58% | 54% |
| Edge over distance + time at the call | +7.6 pts | +9.3 pts |
These calls are rare, and the ES confidence interval just touches zero. There were too few GC calls to judge.
Outside value: calls with real room
Calls made right next to the target carry no edge (see Headline hit rates flatter). Calls made with real room are different. The rules were written in advance, then checked once on the held-back year:
| Return calls with room, held-back year | Edge over distance + time | Verdict |
|---|---|---|
| ES, overnight | +9.5 pts | Confirmed |
| ES, New York session | +5.2 pts | Confirmed |
| NQ, New York session | +8.0 pts | Confirmed |
| NQ, overnight | positive, interval includes zero | Not conclusive |
| GC, overnight | n/a | Failed |
Where it held, the typical move against the call was well below the room to the target.
Not shown yet
- Calls on the farther level between two balance areas showed an edge on the held-back year but not in the years before. Unproven.
- Escape calls outside value show no edge over distance and time in any market or session.
Why this is a forward test, not a conclusion
Part of the held-back year had been looked at in an earlier study before this test. No untouched historical data is left, so results logged forward since late September 2026 are the final judge.
How these numbers are produced and compared: our method. Every change to this finding is dated in the research log. Historical results only; not a forecast and not a trading system. See the disclaimer.