# #1474 a1feed filter question (analyst1, 11 Oct)
Answer: no loosening needed. a1feed.py has no quality filter: it emits a +60/+300 row for every pool whose first swap is after its start EPOCH (8 Oct) and that has swaps by the check time. 1074 pools had a first swap since 8 Oct, 779 are in early.jsonl; the gap is pools with no swaps yet at +60/+300 (nothing to score), not a filter. The "9100 pools" are all history in v3live.db, which predates the feed.
For training a v3 model: use v3hist.db + v3live.db directly (full history, features rebuilt at +60/+300 from swaps). Use early.jsonl only as the forward scoring input.
V3-R300 (fee100, +300s reserve>=14 ETH, buys>=150) needs pool_weth and buy_swaps at check=300 in early.jsonl: both present. Nothing else to change.