Verified profile of Polymarket trader swisstony (proxy wallet 0x204f72f35326db932158cba6adff0b9a1da95e14, pseudonym “Frail-Possible”, no linked X). Re-investigated 2026-06-19 via a Claude workflow over the live Polymarket Data API + our pmv2 Supabase DB; every figure was recomputed from raw trade/position files and all claims adversarially verified. He is an algorithmic HFT sports market-maker bot, not a human discretionary sportsbook bettor — and he is not copyable by our copy-trade op.

Verdict — NOT copyable (and supersedes prior framing)

NOT copyable for our copy-trade op. swisstony’s edge accrues to the resting maker: a follower mirroring his fills crosses the spread as a taker, pays away the ~1% that is the strategy, then loses on gas/fees on top. He is also 0% crypto, so there is no fit with our crypto-updown first-mover lane (he runs multi-leg, two-sided positioning vs. our single-leg directional copy). Our ~30-min REST trigger lag is fatal against a bot doing up to 13 fills/second. Only legitimate use: treat him as a flow / liquidity-concentration signal, never a trade to mirror.

This note CORRECTS / SUPERSEDES a prior characterization. An earlier session framed swisstony as a “favorite-lay / draw-fade sportsbook”-style discretionary bettor. That framing is wrong and is now retired. The trade-by-trade evidence below shows a machine running full-board liquidity provision; the apparent “favorite bias” is favorite-premium harvesting by a market-maker, not a human’s directional lay/fade book. Do not reuse the old “sportsbook bettor” framing in future sessions.

Who he is — leaderboard standing

MetricValue
Identityproxy wallet 0x204f72f35326db932158cba6adff0b9a1da95e14, pseudonym “Frail-Possible”, no linked X
Rank by lifetime volume#1 on all of Polymarket (~$990M)
Rank by PnL#5 ($9.93M)
Category~100% SPORTS — not in the top-1000 of crypto, politics, or culture
Recent — monthrank #24, +$672,804
Recent — weekrank #10, +$991,897
Margin (PnL / volume)~1.00% all-time, 0.45% month, 1.13% week

A ~1% margin on ~$990M of lifetime turnover is the signature of a volume-and-spread business, not a high-conviction directional bettor. He earns thin and often, not fat and rare.

Why he is a bot, not a human (decisive evidence)

Measured over a single 3.3-hour trade window:

  • 3,497 trades in 3.3h ≈ ~1,055 trades/hour.
  • Up to 13 fills inside a single 1-second timestamp.
  • 56.5% of trades share a timestamp with another trade.
  • All 3,497 trades have unique transaction hashes — this is the decisive point: it rules out “one human click that split into many maker fills under one tx.” Each fill is its own on-chain transaction → an automated agent placing/refreshing orders continuously.
  • 100% BUY in-window — zero SELLs. He exits by resolution or by an offsetting BUY on the other side, not by selling.
  • Breadth: 64 events / 325 distinct outcomes in 3.3 hours. No human watches and prices 325 outcomes by hand at that cadence.

For Agents

Market-making is inferred, held at medium confidence — the Data API exposes no explicit maker/taker flag. The inference rests on: unique-tx-per-fill, sub-second fill bursts, full-board two-sided saturation, and sub-$1.00 combined Yes+No entries (a locked spread only a resting maker captures). Treat “maker” as a strong inference, not a logged fact.

Strategy — full-board liquidity provision

He saturates every sub-market of a few live matches: moneyline / draw / BTTS (both-teams-to-score) / over-under totals / spread / halves. He is not picking winners — he is quoting the whole board and collecting the spread.

  • Snapshot was World-Cup-soccer dominated: 97.5% of trade notional.
  • Two-sided positioning dominates: 74.4% of notional sits in conditions he holds on both outcomes.
  • 72.5% of two-sided open markets have a combined Yes+No entry below 0.992) → a mechanical locked spread that only the resting MAKER earns. (A taker crossing in pays >$1.00 combined and gives away this edge.)
  • Price barbell: most trades are cheap (entry < 0.90** — he loads big size into near-certain favorite legs.

Source of profit

Volume × spread + favorite-premium harvesting — NOT directional skill. This is the crux for our scoring:

Why our engine reads edge ≈ 0 while his ledger prints ~1%

Our favorite-bias-stripped edge metric oscillates around ZERO for swisstony: −0.0064 / +0.0036 / −0.0071, consistency 0, W/L 40/37 on n_resolved = 77. That is expected and correct: our scoring deliberately strips the favorite premium, which is exactly the income stream he lives on. So an engine looking for directional edge sees nothing, while his realized ledger compounds ~1% per unit volume. The two are not in conflict — they measure different things. Do not “fix” the engine to score him highly; he has no directional edge to capture.

Risk profile — single-match prop concentration

His tail risk is concentration into one match’s prop markets, not directional drawdown:

  • One match (Canada–Qatar) was 46.1% of his open book and accounted for the entire −$363k unrealized loss in the snapshot.
  • Within that match: main lines +412.8k → classic “right on the result, wrong on the total.”
  • Sizing is a power law: median trade 219k. Size tracks liquidity / order flow, not forecast confidence.

Caveats (scope of the evidence)

Snapshot-specific, not lifetime

  • The trade sample is a single ~3.3h, ~3,500-trade window (the Polymarket Data API offset cap). So “100% BUY” and the World-Cup mix are snapshot-specific, not lifetime facts.
  • Open PnL figures (incl. the −$363k) are unrealized mark-to-market, not realized.
  • n_resolved = 77 is small — the edge metric carries wide uncertainty.
  • The model uses fee = 0.
  • No explicit maker/taker flag in the data — the market-making conclusion is inferred (medium confidence), per the agents note above.

Why this matters for our op

  1. Copy-trade lane (copy-trade-lane): mirroring his fills is structurally losing — you become the taker on his spread. Skip.
  2. First-mover / cohort lane (cohort-firstmover-nats-migration): no fit — he’s 0% crypto, multi-leg two-sided, and our ~30-min REST trigger lag can’t track 13 fills/sec.
  3. Scoring (pmv2-two-tier-trader-scoring): he is the canonical example of a high-volume, high-PnL wallet that our favorite-stripped edge correctly scores ≈ 0. Use him as a regression case: “did we accidentally start rewarding favorite-premium harvesting?”
  4. Best use: a flow / liquidity-concentration signal — where the smartest market-maker on the platform is parking size — not a directional copy target.