"Edge-hunt37 — on-chain timing: signal or momentum proxy?" · Quant Trading Research & Paper-Trading Lab

PUELL = miner-revenue / 365d-MA — mechanically ~= price/1y-MA. corr(PUELL_z, MOM365_z) = +0.791.

#Legs (long BTC when z>0 / z<0, flat otherwise)

leg	exposure	FULL net SR	IS net SR	OOS net SR	OOS B&H	FULL net DD	OOS net DD	flips
PUELL z>0	44%	1.23	1.40	0.97	0.60	-69%	-50%	812
ADDR z>0	59%	0.89	1.31	0.00	0.60	-85%	-83%	1367
HASH z>0	78%	1.04	1.38	0.33	0.60	-80%	-75%	1103
FEE z>0	50%	1.05	1.43	0.25	0.60	-88%	-71%	745
NVT z>0	53%	0.65	0.91	0.16	0.60	-93%	-81%	1591
PUELL z<0 (reverse)	47%	-0.33	-0.14	-0.67	0.60	-99%	-90%	812
ADDR z<0 (reverse)	39%	-0.10	-0.15	-0.02	0.60	-84%	-51%	1368
HASH z<0 (reverse)	21%	-0.32	-0.10	-0.56	0.60	-86%	-76%	1100
FEE z<0 (reverse)	49%	0.16	0.20	0.09	0.60	-81%	-57%	744
MOM365 z>0 (control)	43%	-	1.15	0.89	0.60	-50%	-61%	195

#Incremental value over momentum (the decisive test)

corr(PUELL_z, MOM365_z) = +0.791 (proxy-ness)
partial corr(PUELL_z, fwd1 | MOM365_z) = +0.0913
partial corr(PUELL_z, fwd21 | MOM365_z) = +0.0735

#Block placebo (PUELL z>0, 90d blocks shuffled)

OOS observed net SR 0.97 vs null mean 0.13 / p95 0.66 -> p 0.005
IS observed net SR 1.40 vs null mean 0.85 -> p 0.040
DSR(K=9) OOS net vs B&H 0.60: 0.71 | haircut +0.55

#PUELL quintiles -> mean BTC fwd21

Q	mean fwd21	n
Q1	+0.16%	914
Q2	+1.84%	913
Q3	+3.08%	914
Q4	+5.46%	913
Q5	+8.06%	914

#MOM365 quintiles -> mean BTC fwd21 (the proxy)

Q	mean fwd21	n
Q1	+0.06%	917
Q2	+1.97%	916
Q3	+3.08%	917
Q4	+6.03%	916
Q5	+7.50%	917

#Verdict: REJECT

PRIMARY PUELL z>0: OOS net-vs-BH +0.37, placebo p 0.005, IS edge -0.15, partial-IC +0.091, PUELL 0.97 vs MOM365 0.89

#Reading (near-miss, but not a new edge)

The placebo is significant (p=0.005) and PUELL beats the MOM365 control in BOTH windows (IS 1.40 vs 1.15, OOS 0.97 vs 0.89), so the timing is not noise.
But corr(PUELL_z, MOM365_z) = +0.79: PUELL is a noisier version of the price/1y-MA trend oscillator. The effect is the well-known trend-following / drawdown filter (BTC above trend -> hold), which the live gate60 family already runs — not a new on-chain edge.
It fails the pre-registered bar because the long/flat overlay gives up bull-market return (IS net SR 1.40 < B&H 1.55) while the OOS window happens to reward de-risking (0.97 > 0.60). Sign-flip IS/OOS = regime, not skill.
Verdict REJECT. Trigger for a frozen re-run: a genuinely price-orthogonal on-chain signal (PUELL is not it), or a shadow log of the PUELL gate as a BTC risk overlay only. No live wiring.

Legs (long BTC when z>0 / z<0, flat otherwise)

leg

exposure

FULL net SR

IS net SR

OOS net SR

OOS B&H

FULL net DD

OOS net DD

flips

PUELL z>0

44%

1.23

1.40

0.97

0.60

-69%

-50%

812

ADDR z>0

59%

0.89

1.31

0.00

0.60

-85%

-83%

1367

HASH z>0

78%

1.04

1.38

0.33

0.60

-80%

-75%

1103

FEE z>0

50%

1.05

1.43

0.25

0.60

-88%

-71%

745

NVT z>0

53%

0.65

0.91

0.16

0.60

-93%

-81%

1591

PUELL z<0 (reverse)

47%

-0.33

-0.14

-0.67

0.60

-99%

-90%

812

ADDR z<0 (reverse)

39%

-0.10

-0.15

-0.02

0.60

-84%

-51%

1368

HASH z<0 (reverse)

21%

-0.32

-0.10

-0.56

0.60

-86%

-76%

1100

FEE z<0 (reverse)

49%

0.16

0.20

0.09

0.60

-81%

-57%

744

MOM365 z>0 (control)

43%

1.15

0.89

0.60

-50%

-61%

195

mean fwd21

+0.16%

914

+1.84%

913

+3.08%

914

+5.46%

913

+8.06%

914

mean fwd21

+0.06%

917

+1.97%

916

+3.08%

917

+6.03%

916

+7.50%

917

Reading (near-miss, but not a new edge)

The placebo is significant (p=0.005) and PUELL beats the MOM365 control in BOTH windows (IS 1.40 vs 1.15, OOS 0.97 vs 0.89), so the timing is not noise.

But corr(PUELL_z, MOM365_z) = +0.79: PUELL is a noisier version of the price/1y-MA trend oscillator. The effect is the well-known trend-following / drawdown filter (BTC above trend -> hold), which the live gate60 family already runs — not a new on-chain edge.

It fails the pre-registered bar because the long/flat overlay gives up bull-market return (IS net SR 1.40 < B&H 1.55) while the OOS window happens to reward de-risking (0.97 > 0.60). Sign-flip IS/OOS = regime, not skill.

Verdict REJECT. Trigger for a frozen re-run: a genuinely price-orthogonal on-chain signal (PUELL is not it), or a shadow log of the PUELL gate as a BTC risk overlay only. No live wiring.