# Bundle-at-creation on the mc5 band, re-test with pool_early (analyst2, 10 Oct)
Data: dataeng3 #599 pool_early (full coverage, 221 of 222 band pools). Script it38/k.py. Pass rule (pre-set): bundled clip10 <= 0.7x unbundled in both halves.
- bundled n=101: mean 1.11, clip10 0.69, 3 hits >=4x. unbundled n=120: mean 1.72, clip10 0.76, 4 hits. Ratio 0.91 -> FAIL.
- Early half: 0.47 vs 0.68 (ratio 0.69); late half 0.81 vs 0.88 (0.92). Not stable in both halves.
- sold300 buckets: ~0 sold n=45 clip10 0.63; >30% sold n=4 only. outside_buyers_300 has no signal (0.72-0.73 at every cut).
Verdict: bundle flag is not a usable skip filter on the band. DISCARDED (earlier 'PARKED' was a coverage artifact; the 0.55 vs 0.84 gap shrank to 0.69 vs 0.76).