universe

subscribed

150

score version wallet-v0

on the bench

1,003

picked, not listened to

retired

512

dropped for silence

took part in signals

146

at least one

#	address	score	PnL	consist.	winrate	select.	recency	repro.	signals	for smarta	scored
21	0xe737…0547	72.2	0.41	0.93	0.57	0.78	0.97	0.87	123	-1.0% /86	6d
22	枯坐p小将	71.8	0.69	0.97	0.80	0.00	0.97	0.87	48	-27.0% /48	5h
23	0x481f…8972	71.5	0.90	0.40	0.93	0.63	0.82	0.73	28	-18.2% /8	2d
24	0x85a3…92fa	71.3	0.71	0.67	0.61	0.87	0.99	0.47	13	+4.4% /12	10d
25	0xae0c…dad1	70.9	0.42	0.97	0.45	0.71	0.99	0.88	126	-1.1% /67	19d
26	0x5b19…f719	70.9	0.70	0.50	1.00	0.53	0.96	0.82	8	-14.6% /8	4d
27	0x0718…01ef	70.9	0.41	0.83	0.66	0.78	0.99	0.81	70	-0.3% /68	2d
28	0x144f…966e	70.8	0.48	1.00	0.30	0.71	0.98	0.88	114	-0.5% /69	20d
29	0x2960…3623	70.7	0.38	0.93	0.51	0.78	1.00	0.84	265	-0.2% /66	17h
30	0x9ae7…7c69	70.7	0.44	0.97	0.39	0.78	1.00	0.80	57	-1.1% /56	2d

how the board is built bench, retirement, weights

score is how the wallet trades on its own: six dimensions from its leaderboard history. for smarta is how its clusters moved — median price change an hour after the setup, over 168 hours, measured clusters beside it. A wallet can trade well and still reach us only on its worst entries. Unmeasured says so instead of printing a zero.

The board lists the 150 addresses the ingestor listens to, best score first; 1,003 more wait on the bench for a slot. Without that reserve, dropping a silent address would shrink the universe instead of improving it. Two days of silence means retirement, with the reason recorded on calibration.

Dimensions are normalised to 0..1 and summed with weights. Win rate is damped by sample size — a hundred percent over seven tokens is a sample size, not skill — and it lowers a wallet in the ranking rather than throwing it out.