What I learned watching the top 20 Polymarket wallets for 90 days

I spent 90 days building and running polycopy — a bot that watches the top 20 Polymarket wallets by volume, detects when they make meaningful new positions, and copies the trade. The bot works. The lesson I learned is more complicated than "copy winners."

The naive theory is wrong

The naive theory of whale copying goes like this: the largest wallets have the best information, so copying them is free alpha.

This falls apart almost immediately when you look at the actual trade history. Most top-20 wallets by volume are not consistently profitable across all market types. They have edge in specific corners.

The most common patterns:

Early arb wallets. These wallets show up in the first 2–4 hours of a market's life, when pricing is often wrong by 5–15% because there aren't enough informed traders yet. Their edge is being early, not being more informed. By the time you observe their position and copy it, the market has already repriced toward fair value. You're buying after the mispricing is gone.

Niche-market specialists. Some wallets consistently win on very specific market categories — obscure geopolitical questions, local elections, niche sports leagues — where they plausibly have a genuine information advantage. A $50k position on a specific municipal election outcome from a wallet that has won 15 similar markets is worth paying attention to.

High-volume hedgers. Some large wallets are using Polymarket as a hedge for a real-world position. Their trading looks correlated with the outcome they care about but they're not trying to maximize prediction market returns. Copying them is unprofitable.

Lucky streaks. A 5-market winning streak is consistent with a 50/50 trader. Some wallets that look like geniuses in a 30-day window are noise.

The 0x7c3db7 case

One wallet — I'll use the first 6 characters, 0x7c3db7 — was the most interesting in the 90-day observation window. This wallet placed above-market-size positions on climate and weather-related markets with a win rate that was statistically hard to explain as luck: roughly 68% over 40+ markets during the observation window.

Three hypotheses for that edge:

1. Access to private meteorological ensemble forecasts earlier than public models
2. A systematic approach to reading NOAA/ECMWF ensemble spreads that most traders don't do
3. Genuine luck that will regress

I don't know which is true. What I do know is that copying 0x7c3db7's weather-market positions, with appropriate latency and sizing adjustments, generated positive returns in the observation window. Whether that persists is an open question.

The broader point: when you find a wallet with apparent edge in a specific domain, the useful question is "what's their information source?" not "should I copy everything they do?"

Why naive copying loses

Even if you identify a wallet with genuine edge, three problems compound:

Latency. polycopy polls the Polymarket Data API. The whale places a trade; the API reflects it; polycopy detects it and places a copy. By then, 2–15 seconds have passed. In a liquid market, a meaningful trade by a known whale moves the price immediately. You're buying after the price impact.

For markets with thin books, the price impact is even worse. A $10k trade in a $50k liquidity pool moves the market 3–5 cents. By the time you're copying, you're crossing that spread.

Sizing. A wallet with a $200k bankroll taking a $5k position is sizing at 2.5%. The same position as a fraction of a $1k bankroll is 500% — obviously impossible. Naive proportional sizing doesn't work. polycopy uses a fixed-fraction of your configured max balance, not proportional replication.

Selection bias. You observe the wallets that are big now. Most of them became big by winning. The ones that lost are gone. You're sampling survivors, which biases your estimate of edge upward.

Exit timing. Copying entry is tractable. Copying exit is harder. Does the whale hold to resolution? Exit early? Their position management is invisible until they submit a transaction. polycopy holds to resolution by default, which is not always what the whale does.

What polycopy adds

Given these problems, what does polycopy actually do?

Three things that reduce (but don't eliminate) the naive-copying failure modes:

Wallet filtering by domain. polycopy tracks per-wallet win rates by market category. If a wallet has strong historical performance on sports markets but random performance elsewhere, it only copies sports trades from that wallet. This requires enough history to be reliable (50+ trades in a category), so it's limited at first but improves over time.

Minimum liquidity check. If a market has less than $X in available liquidity at the copy price, polycopy skips the trade. This prevents copying into thin books where price impact would make the trade negative EV before you even get filled.

Staleness filter. If the wallet's trade is older than N seconds (configurable, default 45s) by the time polycopy detects it, the trade is skipped. Old signals have already been arbitraged away.

None of these fully solve the latency problem. polycopy is most useful for wallets with sustained edge in liquid markets where the price impact of copying is small relative to the edge. Those wallets exist — but they're maybe 3–5 of the top 20, not all 20.

What I'd do differently

If I rebuilt polycopy from scratch, I'd focus on two things:

Better wallet classification upfront. Instead of treating all top wallets the same and filtering by domain after the fact, I'd do manual analysis of each wallet's history before including it in the copy list at all. A wallet with 60%+ win rate over 100+ same-category trades, in markets with at least $50k of liquidity, is worth copying. Everything else probably isn't.

Separate copy from position management. The entry signal from whale observation is decoupled from exit decision. Use the whale to identify the market and direction; use your own signal to decide when to exit. A whale holding to resolution when the outcome is already obvious is leaving money on the table (or holding for strategic reasons you don't understand).

The honest summary

Whale copying is not free alpha. The wallets with genuine edge are a small minority of the high-volume set, their edge is domain-specific, and execution latency makes copying expensive in thin markets. polycopy is a tool for capturing the small real signal in whale activity — not a shortcut that bypasses the need to think about market structure.

The 90-day observation window was worth more for what it taught me about how different wallet types approach the market than for the direct trading returns it generated. If you're building prediction market bots, watching how the best wallets behave is valuable research even if you never copy a single trade.