Deduplication and anti-Sybil in simple terms means filtering out large clusters of addresses who are making transitions (often low value transactions) which are performed for the purpose of farming an airdrop, or are just not relevant enough to be considered a legitimate user.
Other examples are bots which are creating thousands of wallets and making $0,01 transactions on memecoins to boost the trending scores of algorithm of platforms like DexScreener with transacting counts and maker counts, in an attempt to make a coin trending. Nearly of these are filtered out by excluding wallets with less than < $100 in transaction volume, which would filter out 11,361,557. Some blockchain analysis tools can be used to filter out some additional wallet clusters, which is deduplication or anti-Sybil measures.
I do believe most of the 3,885,443 users with $100+ volume are legitimate, as Jupiter had 955,000 users one year ago at the 2nd of November 2023 snapshot - and a 4X increase in the amount of users on Jupiter over the last year is very realistic.
About high volume (arbitrage) bots, per my proposal they would not get more than 9,045 JUP + 10 JUP for the $10M+ tier. Even if they did $100M or $500M or Billions in volume. That is the reason for making it tiered, and not linear.
@meow wrote the following about the bots in last year airdrop announcement: