I tried to measure how "copyable" Hyperliquid + Polymarket wallets really are. Almost none are, and the reason is kind of obvious in hindsight.
Something about copy-trading has bugged me for a while: every leaderboard ranks traders by PnL, but PnL tells you basically nothing about whether *you* could reproduce it. So I spent way too long building a thing to actually measure that.
Pulled the whole Hyperliquid leaderboard (~39k accounts), narrowed to the few thousand that are actually active in a sane size band (no mega-MM/HFT), and scored the ~1,000 that are even mechanically copyable on the stuff that actually eats your returns when you follow someone:
- can the edge survive you filling *late* (copy-lag)?
- does it hold up de-levered to a size you'd really run, or is the return just leverage?
- is there real alpha, or is it levered beta to BTC?
- does it work in more than one market regime?
- repeatable, or one lucky trade carrying the whole record?
Only about 180 made it through — call it 1 in 5. Fewer than 40 I'd call genuinely solid. So ~80%+ of "copyable" HL wallets fall apart once you account for reality.
The wallets that look best are usually winning on something you can't replicate — queue priority, speed, or leverage. (I ran the same test on Polymarket out of curiosity, and it's even worse there: almost none survive, because most of the "profitable" wallets are market-makers sitting on resting orders you can't copy. Different venue, identical trap.) Which is basically the whole problem with "just copy smart money."
Data's here if you want to poke at it (free, read-only, public on-chain data, no signup, I'm not selling signals): https://blacklight.observer — it opens on the Hyperliquid list with the full funnel (screened → copyable → how many actually survive), sorted by realizability. You can filter it, open any wallet's full breakdown, or paste your own address. (Counts drift a bit day to day since it re-runs, but the shape holds.)
Mostly posting because I want people to find holes in it. If a wallet you *know* is good scores low, or some obvious junk scores high, tell me — that's the feedback I actually need. Happy to run any address in the comments.