Let me start with the conclusion — also the most valuable sentence in this entire post: I looked, and I didn’t find one. Under the combined constraint of “public + independently verifiable + over 2 years + retail + perpetual futures + net profitable after funding,” not a single trader survived verification. All five candidates were rejected, zero survivors. But more important than “didn’t find one” is the second sentence: this question, under the bar I set, may be fundamentally unfalsifiable. It’s not that the data is insufficient — it’s that the identity of “retail trader” cannot be proven from start to finish inside an anonymous on-chain wallet.

1. Why I Went Looking for Someone Like This

The LynxCrypto project has reached a point where my own MacdCross-4h strategy just passed its Walk-Forward final review — out-of-sample 5/5 profitable, MCPT p=0.003, DSR=1.000 — and is qualified to enter demo dry-run. But I’ve had this question in the back of my mind the whole time: does there exist a real person, a retail trader, who has been consistently profitable on crypto perpetual futures over the long term — and I mean long term, two years or more — in a way I can verify?

Not the kind of thing where you hear someone bragging — the kind where I can pull the data and double-check myself. If such a person exists, I want to reverse-engineer his trading behavior, extract his decision patterns, model his ability — and see whether it’s genuine alpha or just luck. If no such person exists, or if it’s fundamentally impossible to prove they exist, then I should rein in my expectations about “being able to stably profit” myself.

The academic consensus is right there: roughly 70-80% of crypto futures retail traders lose money, and among day traders in the entire Brazilian market who stuck it out for 300 days, 97% lost. But I wanted a verifiable counterexample, not a legend. So I set six bars, all of which must be met to count:

1. Public — not a track record handed to me privately, but something anyone can look up
2. Independently verifiable — I can pull the data and double-check myself, not relying on someone’s word
3. Over 2 years — not a survivor of a single market cycle
4. Retail — not institutional, not a market maker, not MEV, not exchange-internal
5. Perpetual futures — not delivery futures, not spot
6. Net profitable after funding — funding is the big cost in perpetuals; calling something “profitable” without accounting for funding is dishonest

All six must hold simultaneously. The bar is high, I know.

2. How I Searched: Not a Casual Google

A casual Google search on this kind of question is just lying to yourself. I used 16 agents and three search engines cross-checked — Grok, Tavily, and Metaso — across six angles:

• On-chain perpetual DEXs (Hyperliquid, GMX, dYdX, Drift, Vertex, Aevo) with verifiable wallets
• Centralized exchange leaderboards (Binance, Bybit, OKX) + BitMEX historical leaderboards (2016–2020, the originator of perpetuals, snapshots on archive.org)
• Academic empirical research
• Chinese retail communities (Chinese retail traders are the main force in futures)
• Available datasets (specifically looking for trade-level public data suitable for reverse engineering)
• Audit/regulatory/social channels with third-party verification trails

After scanning, I didn’t just take things at face value. I ranked all candidates by “evidence strength” — on-chain verifiable wallets > leaderboard snapshots > audits > academic > social — took the top few, and dispatched agents one by one to use WebFetch to actually pull leaderboards, wallets, block explorers, and papers for verification. Then came another round: a skeptic agent took the surviving candidates and independently went online to hunt for disconfirming evidence (searching names + “market maker”/“scam,” searching platforms + “leaderboard reset”). If it couldn’t be confirmed, it was ruled not established.

Three layers: scan → verify → rebut.

3. Five Candidates, How They All Died

They all share a common pattern of death — one look and you can tell these aren’t humans:

• 3 to 5 trades per second, simultaneously across 70+ tokens — a human can’t do this
• Positions routinely $30M to $125M — whale-scale
• 100% win rate, large volume of zero-hash maker fills — this is a market-making/arbitrage/liquidation machine, not directional speculation
• Time active generally under 2 years — Hyperliquid only launched in November 2023; the longest candidate only barely hits 2.6 years because the platform itself is young
• Most are net-negative recently — the most telling case is that “White House Whale”: a peak +$33M “100% win rate” snapshot that turned into −$17.6M when the market reversed. This is a living textbook case of survivorship bias: you see the screenshot from when he was profitable, but not the aftermath after he got liquidated.

All five carry the fingerprints of institutional/algorithmic operations, not human ones.

4. Why the “Retail” Bar Is a Dead Knot

This is the most fundamental blocker, and also the most honest point. All on-chain wallets are anonymous addresses with no KYC labels. You can’t distinguish:

• A genuine retail trader manually placing orders with $5,000
• A quant team testing strategies with $5,000
• A large institution’s sub-account camouflaged as a small account
• A criminal trading with stolen funds

There’s an example that illustrates this perfectly: on-chain detective ZachXBT unmasked a well-known “anonymous whale” on Hyperliquid — it turned out to be a British fraudster named William Parker. So anonymous whale does not equal retail, and you must never default to assuming otherwise. Hyperliquid simultaneously hosts market-making vaults, institutional market makers like Jane Street, MEV bots, and quant teams — all using the same KYC-free self-custody wallets, with no on-chain labels to tell them apart.

This means: the proposition “does there exist a long-term profitable retail perpetual futures trader” is, within the boundaries of data I can access, neither confirmable nor falsifiable. It sits in an unfalsifiable state — not “there isn’t one,” but “can’t prove there is, and can’t prove there isn’t.” This is the ceiling of this research, and I have to acknowledge it honestly.

5. Didn’t Find a Person, but Dug Up Two Genuinely Usable Datasets

I didn’t find a person, but I found two datasets I can use to dig on my own — and that’s the most concrete takeaway from this entire exercise. Even without a ready-made “god-tier trader” to copy, as long as there’s trade-level public data, I can do behavioral reverse engineering, decision-pattern extraction, and ability modeling myself.

① Hyperliquid L1 Info API — the cleanest raw data source. Trade-level fills, positions, liquidations, and funding rates are all natively recorded on-chain, pullable without authentication. Key endpoints:

• userFills / userFillsByTime — trade-level fills, including closedPnl, fee, direction, leverage, coin, and timestamp
• userFunding — funding rate cash-flow time series
• userNonFundingLedgerUpdates — deposit/withdrawal timeline

The net profit formula is directly computable: sum(closedPnl) - sum(fee) + sum(funding) - sum(withdrawals). 3.4 years of history. This is the only source among all candidates where funding is recorded trade-by-trade on-chain and net profit can be independently calculated.

② CoinLobster’s 60 pre-filtered wallets — the best “wallet discovery entry point.” It has one filter: daily average trades under 30. This under-30-trades-per-day threshold happens to exclude exactly the HFT/bots that killed all five candidates — which is precisely where all five trader candidates died. But note that its own profit ranking excludes funding rates and carries affiliate referral links, so net profit must be independently recalculated by taking its wallet addresses back to the Hyperliquid API — you cannot trust the numbers on its leaderboard.

One as the address entry point, one as the raw data. Combined, they form a self-built reverse-engineering pipeline.

6. The Most Critical Pit: Leaderboards Don’t Include Funding

This is the one practical lesson I think is most worth remembering from this entire post. The HyperStats leaderboard and CoinLobster’s profit ranking both exclude funding rates. In perpetual futures, funding is often the single largest cost or revenue item:

• In a positive-funding market, longs pay continuously — a position that’s directionally profitable may end up net-negative after funding
• Conversely, a wallet that’s directionally losing money might show “profit” on the leaderboard by collecting funding

Third-party analyst Renesis put it bluntly: “funding rate revenue is often the largest contributor to top-of-leaderboard PnL,” and most leaderboard rankings “are an illusion.” So any profitability judgment must be independently recalculated from userFills(closedPnl + fee) + userFunding — you cannot trust any leaderboard’s headline number.

This directly affects my own MacdCross backtest — funding costs must be actually computed, not conveniently ignored, otherwise the backtested profit won’t match live trading.

7. Academia Points the Same Way

It’s not just me seeing it this way:

• Someone analyzed 43,618 Hyperliquid addresses, of which only 12 made it into deep analysis — a strict pass rate of roughly 0.03% (12/43618, derived value)
• An SSRN paper (6701738) directly demonstrates: cross-sectional alpha screening using OHLCV and funding signals fails on crypto perpetuals
• There’s also @hydromancerxyz claiming 29% profitability over an 8-month window — but I could not independently confirm the original source for this claim, so I’ve downgraded it to circumstantial evidence and do not use it as a quantitative basis

All verifiable evidence points in the same direction: on the perpetual futures market, long-term net profitability (including funding) for retail traders is extremely rare, and possibly fundamentally impossible to publicly verify. This is fully consistent with the “97% of retail traders lose” and “honest expectations about real alpha” findings in my own project.

8. So What Can I Do Next

I didn’t find a person, but the path forward is clear. Against my three original goals:

• Trading behavior reverse engineering: Use CoinLobster’s 60 wallets as the address entry point, pull fills + funding + ledger from Hyperliquid one by one, and reconstruct the complete “open position → hold (funding accrual) → close/liquidation” lifecycle. Filter criteria: daily average under 30 trades, concentrated on single instruments, no zero-hash maker fills, account under $1M, trade interval over 10 seconds (human pace).
• Decision-pattern extraction: Align trade sequences to the market timeline (paired with OHLCV and market-wide funding rates), use decision trees to extract “what operation under what market state” rules, use sequence pattern mining to find repeated operations, and cluster into categories like “trend-following / contrarian / funding arbitrage / event-driven.”
• Trader ability modeling: Compute time-weighted rate of return (TWRR) to avoid deposit/withdrawal distortion, use beta decomposition to separate directional alpha from funding carry income, run Walk-Forward rolling-origin out-of-sample tests, and Monte Carlo permutation tests to judge whether profitability is significantly better than random.

The tech stack is what I defined in my project: Jupyter + vectorbt as the brain, Freqtrade as the hands.

The concrete next step: pull data for all 60 wallets in full, independently recalculate net profit including funding, filter out the subset that meets “over 2 years + net profitable + low frequency + small account + no maker fills,” and then do ability modeling on that subset. If that subset turns out to be empty — that itself is the final honest conclusion of “no verifiable counterexample exists,” not a failure.

9. An Honest Ending

Not finding one isn’t a failure, it’s a conclusion. The inability to identify retail identity is the ceiling of this research, so all conclusions can only be presented conditionally — “if this wallet is retail, its net profit is Y” — rather than “this retail trader has ability.” This is the most honest answer I can give.

The full report (8 sections, including deduplicated ranking of 37 trader candidates, detailed rejection of each of the 5 candidates, endpoint-level documentation of the two datasets, and an online verification checklist for external citations) is in the local file lynxcrypto-perp-trader-hunt-2026-09-06.md. Six bars, three layers of verification (scan → verify → rebut), six search angles — anyone who wants to double-check can pull it themselves.

Consistent with the direction of my entire project: don’t believe in get-rich myths. First get the costs and funding rates right, build a falsifiable system, and let it die in front of you on historical data first.