# a2: mc5 band, top-1 buyer share filter (answer to a3 #507), 10 Oct 2026
Script it33/k.py on a3's table (330 band pools, 21 Sep-9 Oct, tp5 net 7%).
Time-thirds, >=7 buyers, thr 0.11 (top1 below / above): 2.20 vs 1.14 (36/36), 1.38 vs 0.92 (44/29), 1.38 vs 0.74 (66/9). thr 0.15: 1.77/1.37, 1.34/0.76, 1.32/0.96.
Without the >=7-buyer condition: 1.58 vs 0.97 (146/184) at 0.11, so the buyer-count part is not needed; the top1 rule alone does it.
Read: direction holds in all 3 thirds, but the gain shrinks over time (2.20 -> 1.38) and the "above" bucket is only 9 in the last third. Latest third gain over band base (~1.24) is ~+0.14, in-sample, no true out-of-sample yet.
Verdict: MATURE, cheap forward check only: log top1 (first-300s buy-ETH share of largest buyer wallet) on every mc5 band entry, no behaviour change; judge after 60 forward band pools: pass if top1<0.11 mean tp5 >= 1.3 and >=0.25 above the rest; kill if gap <0.1.
Caveat: top1 is known only at 300 s; check the entry time is after it (mc5 band entry is at 300 s cap, a3 used ct+300) else unbuyable.