Thesis under test: volatility is forecastable; direction is not.
Entries permitted only when a volatility-expansion gate fires.
BTCUSDT · ETHUSDT · SOLUSDT · BNBUSDT · XRPUSDT · 1h / 15m ·
$100.0 start · 0.10% fee ·
0.050% slippage · 2% risk
Rendered 2026-08-05 08:37:50 UTC · auto-refresh 30s
01Thesis verdict — does the gate beat no gate?
GATE LIFT IS NOT SIGNIFICANT - best gated +0.0044R vs control +0.0011R is only 0.2sigma. Indistinguishable from noise; thesis unproven.
Layer-1 entry expectancy, no trade management. Judged on gross
edge — the only comparison that isolates prediction from cost arithmetic.
This, not P&L, is the headline: if gating entries does not raise the raw
entry edge above the ungated control, the premise of the whole lab fails.
A lift must clear 2σ to be called a result; the σ shown uses a
deliberately optimistic standard error (overlapping forward-return windows mean
the true error is wider), so anything failing this bar certainly fails a stricter one.
Control gross
0.0011R
Best gated gross
0.0044R
Gate lift (gross)
0.0033R
Significance
0.20σ
Round-trip cost
−0.2401R
Signals (ctrl / gated)
18096 / 16432
DIRECTION MODEL HAS NO PREDICTIVE CONTENT
Control gross expectancy is 0.0011R — statistically
indistinguishable from zero over 18096 signals. Before costs, T3 +
Squeeze predicts direction no better than a coin. The entire net loss of
-0.2391R per trade is the 0.2401R round-trip
toll. This is consistent with the second half of the thesis —
direction is not forecastable — and it means no amount of gate tuning
can rescue this direction model: a gate can only choose when to bet,
it cannot manufacture an edge that isn't there.
02Variant ladder — does strictness buy performance?
Variant
Gates
Signals
L1 gross
L1 cost
L1 net
Trades
PF
Win rate
Net P&L
CONTROL (ungated)
0
18096
0.0011R
−0.2401R
-0.2391R
9880
0.51
24.3%
-$958.35
v1 ATR-rising
1
8517
0.0044R
−0.2348R
-0.2304R
5951
0.49
24.2%
-$874.27
v2 +vol-expansion
2
4371
-0.0055R
−0.2007R
-0.2061R
3335
0.49
25.6%
-$690.35
v3 +ADX
3
3196
-0.0138R
−0.1969R
-0.2107R
2504
0.48
25.1%
-$593.21
v4 +squeeze-release
4
348
-0.0439R
−0.1995R
-0.2434R
347
0.40
23.3%
-$135.21
PF does not improve monotonically v1→v4. Extra gate conditions are not
buying performance in a consistent way; that weakens the "stricter gate = better" story.
03Promotion shortlist — cleared all four layers
Empty. No account cleared all four validation layers.
This is the expected and honest result — the funnel is built to reject, and
0 of 50 accounts survived to be considered.
04Validation funnel
50
Accounts
2
L1 entry edge
−48 died
0
L2 management
−2 died
0
L3 sizing
−0 died
0
L4 execution
−0 died
L1 fixed-horizon entry edge (≥30 signals, expectancy>0) ·
L2 trail sweep, out-of-sample PF ≥1.15 on ≥25 test trades,
robust on ≥3 symbols ·
L3 max drawdown ≤25% under fixed-2% sizing ·
L4 edge must survive next-open fills, not just close fills.
05Conditional edges — dead overall, alive in one regime
Account
Regime
Trades
PF
Expectancy
P&L
Died at
v2 ETHUSDT 1h
dow=Thu
46
1.27
0.116R
$3.69
L1
v3 XRPUSDT 1h
dow=Thu
32
1.25
0.063R
$4.75
L1
v3 ETHUSDT 1h
dow=Thu
40
1.19
0.070R
$2.53
L1
v2 BNBUSDT 1h
dow=Wed
68
1.15
0.065R
$3.97
L1
Hypotheses only, not promotion candidates: these were found by
slicing after the fact, so each needs ≥30 in-regime trades
(enforced) and a fresh out-of-sample test before it means anything.
06Capacity to validate — how long to a live track record
Account
Symbol
TF
Trades
Trades/month
Months to 30
control
BNBUSDT
15m
979
165.9
0.2
control
BTCUSDT
15m
1017
172.4
0.2
control
ETHUSDT
15m
1022
173.2
0.2
control
SOLUSDT
15m
975
165.2
0.2
control
XRPUSDT
15m
1014
172.3
0.2
v1
BNBUSDT
15m
599
101.7
0.3
v1
BTCUSDT
15m
590
100.3
0.3
v1
ETHUSDT
15m
609
103.6
0.3
v1
SOLUSDT
15m
590
100.3
0.3
v1
XRPUSDT
15m
591
100.4
0.3
v2
BNBUSDT
15m
326
56.3
0.5
v2
BTCUSDT
15m
338
58.0
0.5
v2
SOLUSDT
15m
320
55.0
0.5
v2
XRPUSDT
15m
327
56.6
0.5
Showing the 14 fastest-accumulating of
50 accounts with trades. Anything over
12 months cannot realistically be validated live.
07False-positive context
50 accounts tested ·
~2.5 expected to look good by chance ·
0 actually passed
Passes are within what luck alone would produce -- treat as noise.
At α=0.05, testing 50 variants means
roughly 2.5 will clear a bar like this on luck alone.
A pass count at or below that baseline is indistinguishable from noise.