Do liquidation clusters act as magnets?

Every 15 minutes Blockmetric takes the liquidation map of the tracked Hyperliquid positions of each coin (real positions, not a model). Within ±15% of the mark it ranks the buckets by USD: long liquidations below the mark, short liquidations above it. The 3 largest buckets that hold at least $5M each are recorded as "big clusters" with the distance from the mark to the near edge of the bucket. The same bucket (coin, side, price) is recorded at most once per 24 hours. A cluster counts as reached when the Hyperliquid mark price touches the bucket (longs: mark at or below the top edge; shorts: mark at or above the lower edge). The mark is sampled every 30 seconds; the time of the touch comes from these samples while the collector runs, otherwise from the stored 5-minute highs/lows of the same samples (time rounded to the 5-minute bucket; the bucket in which the cluster was observed is skipped). Wicks between two samples are not seen, so reach rates are a lower bound. A horizon (4 h, 24 h, 72 h) is counted only for clusters observed at least that long ago, so open observations never inflate the hit rate. Observations are checked for 72 hours. Past results do not predict future prices; no investment advice.