# a2 #1277 / #1280 (11 Oct 2026)
## #1277 clone-flood (nm2 names x pool_paths, cap_first<150k, P(max_cap>=1M) by # same-name tokens born earlier; it1277/c.py)
- nohook v4 usdg: 0 prior 1.0% (47/4511), 1-4 1.4%, 5+ 2.6%. nohook v3 weth: 1.3% / 4.1% / 3.7%. nohook v4 weth: 2.3/1.5/2.3. nohook v4 native: 1.3/2.4/2.6.
- Clones do NOT reduce P(1M) inside any nohook cell; v3 weth and usdg clones are higher in both halves (v3 weth 5+: 2.9% H1, 4.3% H2 vs 1.0/1.7%). Hooked cells: 0 prior best (hook weth 5.2% vs 1.7%/1.3%) but names are unique by launcher there: venue confound, and hooked-native 0-prior is a H1 spike (125/232 vs 5/169).
- Verdict: clone-order is no filter and no edge (base rate 1-4%). Lift in v3 weth clones is small-base, parked (weak, could be recycled names of old runners; symbol-collision test not run).
## #1280 holders at 1M -> reach 3M (ladder.db, 423 non-glitch 1M crossers, it1277/l.py)
- 235 reach 3M. Fewer holders at 1M = MORE 3M (holders<=7: 158/212 vs 77/211; top10>0.96: 133/190 vs 67/192); inverted in H2 too only in magnitude (holders<=109 68/107 vs 27/105). Median 7 holders at a 1M cross = tiny concentrated tokens, i.e. jump/wash prints (known artifact), not organic runs. Nothing to enter on; no usable gate (cross is already known only after the print).
- Verdict: discard as a runner predictor. Possible use only as a fake-cap flag (<=2 holders), not a signal.