Unusual flow meets unusual skill. Ten Polymarket wallets on the current monthly leaderboard, scored.

The opening tension

Unusual Predictions, the Polymarket-focused extension of Unusual Whales, surveils prediction- market flow in real time against four public criteria: fresh accounts placing large concentrated bets (potential insider), traders with sustained historical outperformance (smart money), anomalously large single positions (whale trade), and over- sized concentration relative to Kelly (conviction signal). Each criterion is a flow flag first and a skill claim second.

The hard question behind every flow flag is the same: does the wallet moving the size have structural skill, or is it a single concentrated bet that happens to be winning? Flow surveillance alone cannot answer that, because flow surveillance does not have a calibration cohort to score against. Edge Score does. Convexly's published V3b composite is fit on 8,656 Polymarket wallets with five or more resolved positions, has an out-of-fold Spearman of +0.514 against signed log PnL, and passes a Fama-French bootstrap null at p less than 0.0001 across 10,000 permutations.

This post sits the two axes side by side. Ten wallets, picked from the single most-referenced wallet list on crypto-Twitter, scored on both.

How the ten were picked

Source: the Polymarket monthly profit leaderboard at polymarket.com/leaderboard/overall/monthly/profit, scraped 2026-04-20. The top ten wallets by monthly realized profit. These are the wallets whose screenshotted PnL numbers circulate on CT every day and the wallets any flow-surveillance product would be flagging right now on size and concentration. The list is not curated; it is the public top ten in rank order.

UW criteria scoring

Each wallet was scored against five proxies for the Unusual Predictions surveillance criteria, computed from the public analyzer response:

1. FRESH: fifteen or fewer unique resolved positions (concentrated book, not a prolific trader).
2. ALL-IN: biggest event responsible for at least 50% of total realized PnL.
3. OVER-SIZED: Kelly sizing efficiency greater than 1.5x (above-Kelly concentration).
4. HIGH-HIT: win rate at least 60% with a negative skill Brier (wins at a high rate on forecasts that price above own base-rate calibration; the conviction-hitter pattern rather than a pure-calibration one).
5. ANOMALOUS: realized PnL at least $1M on the all-time record.

A wallet needs two or more flags to count as matching the Unusual Predictions category. Three or more is a high-confidence match; the skill signal and the flow signal are both present.

One disclosure on the five thresholds. Fifteen positions, 50% concentration, 1.5x Kelly, 60% win rate, and $1M realized PnL are the cutoffs that approximate the UW public language. None of them are robustness-tested; shifting any threshold by plus or minus 50% would move a flag here or there. The headline "5 of 10" and "4 of 10" counts are directionally stable, not statistically decisive.

A second disclosure on sample sizes. Polymarket wallet PnL has a Hill alpha tail index of roughly 1.28 on our cohort (heavy-tailed; the growth-rate variance is effectively undefined). Skill inference from small numbers of resolved positions is unreliable in this regime. Where n is below the five-position threshold the Edge Score paper specifies, the per-wallet composite should be read as provisional rather than as a confirmed skill signal.

The ten wallets

#1  0x4924...3782 ·  $6.99M monthly profit

63 resolved positions, 55.6% win rate, $926,013 all-time realized PnL, Edge Score 47.9 (posture 83.0, conviction 37.0, discipline 32.6). Skill Brier minus 0.006, roughly baseline. Biggest event: Blackhawks vs Devils at 41.6% of all-time PnL. Kelly sizing efficiency 0.14. The monthly leader is also the lowest Edge Score in the ten, and the only wallet scoring zero UW criteria. Verdict: this is a high-frequency, thinly-sized book that compounded into $7M of monthly profit without any of the classic surveillance flags. Interesting operational profile; not an Unusual Predictions match.

#2   HorizonSplendidView  0x0222...8ff7 ·  $4.02M monthly profit

8 resolved positions, 62.5% win rate, $4,896,518 all-time PnL, Edge Score 83.5 (posture 94.7, conviction 39.9, discipline 93.6). Biggest event: Manchester City vs Nottingham Forest at 48.4%. Skill Brier minus 0.060 (observed Brier above own baseline; overconfident forecast posture). UW flags: FRESH, HIGH-HIT, ANOMALOUS (three of five). Verdict: strong match on three criteria. The posture and discipline pillars are both above the 93rd percentile of the cohort; the book is sharp and focused. Worth watching on the Unusual Predictions dashboard.

#3   reachingthesky  0xefbc...f9a2 ·  $3.74M monthly profit

9 resolved positions, 44.4% win rate, $3,741,725 all-time PnL, Edge Score 81.1 (posture 48.3, conviction 56.2, discipline 92.2). Biggest event: Chelsea vs Paris Saint-Germain at 86.2%. Kelly sizing efficiency 1.02 (right at Kelly). Skill Brier plus 0.076 (at the cohort median; observed Brier beats own baseline). UW flags: FRESH, ALL-IN, ANOMALOUS (three). Verdict: the classic concentration story. Edge Score is in the 81st percentile but driven almost entirely by discipline plus event concentration, not calibration. A flow surveillance product would light up on the PSG position size; Edge Score confirms it as one structural bet rather than a repeatable edge.

#4  0x2a2c...9bc1 ·  $2.89M monthly profit

44 resolved positions, 47.7% win rate, $22,133 all-time realized PnL, Edge Score 99.1 (posture 60.4, conviction 100.0, discipline 46.2). The edge case. All-time PnL is marginally positive across 44 resolved positions, but a single in-progress position (Magic vs Pistons) has contributed $189,783 of unrealized PnL that the analyzer treats as the wallet's "biggest event" at 857% of realized total. The 99.1 Edge Score reads high because conviction percentile sits at 100, but the composite is driven by one open position rather than by a stable resolved track record. UW flags: ALL-IN only. Verdict: do not copy-trade on a single unresolved position. This is the textbook example of why conviction percentile alone is not a signal; it has to be combined with posture or discipline to matter.

#5   beachboy4  0xc2e7...be51 ·  $2.67M monthly profit

21 resolved positions, 71.4% win rate, $4,313,730 all-time PnL, Edge Score 79.3 (posture 92.3, conviction 49.8, discipline 73.3). Biggest event: Liverpool vs Newcastle at 71.5%. Skill Brier minus 0.044. UW flags: ALL-IN, HIGH-HIT, ANOMALOUS (three). Verdict: sports specialist with a real win-rate advantage on 21 resolved positions (the largest usable sample in the ten). Posture percentile in the 92nd is the strongest conviction-above-own-base-rate reading in the analyzable set. Match on three criteria; best candidate for the smart-money category in this cut.

#6   majorexploiter  0x0197...9f3c ·  $2.42M monthly profit

3 resolved positions, 100% win rate, $3,667,734 all-time PnL, Edge Score 94.2 (posture 99.8, conviction 47.4, discipline 99.0). Biggest event: Arsenal vs Chelsea at 65.9%. Skill Brier minus 0.159. UW flags: FRESH, ALL-IN, HIGH-HIT, ANOMALOUS (four of five). Verdict: the highest flag count in the ten, but the sample is tiny. At three resolved positions this wallet is below the five-position threshold the Edge Score paper specifies for stable skill inference; the 94.2 composite should be read as provisional rather than as a confirmed skill signal. A 100% win rate on three binary outcomes is 1.73 standard deviations from a coin flip; any surveillance product flagging this wallet on flow is correct, but the skill confirmation is pending more resolved bets.

#7   RN1  0x2005...75ea ·  $2.25M monthly profit

The public analyzer returned "No resolved positions yet. Come back after some markets close." for this wallet. The leaderboard position is driven by unrealized PnL on open positions that the Edge Score layer cannot score until they resolve. UW flags: cannot be computed. Verdict: excluded from the scoring count; represents a real limit of any skill-measurement layer that requires resolved outcomes.

#8   sovereign2013  0xee61...3deb ·  $2.03M monthly profit

Same result as RN1; the analyzer reports no resolved positions have landed. Both #7 and #8 are showing up on the monthly profit leaderboard on unrealized PnL alone. This is worth flagging in any surveillance product that sorts by monthly profit without distinguishing realized from unrealized. UW flags: cannot be computed.

#9  0x777d...5143 ·  $1.80M monthly profit

11 resolved positions, 72.7% win rate, $248,554 all-time PnL, Edge Score 77.3 (posture 88.3, conviction 39.1, discipline 89.1). Biggest event: Trail Blazers vs Spurs at 46.6%. Skill Brier minus 0.025. UW flags: FRESH, HIGH-HIT (two of five). Verdict: borderline match. A tight, fresh book with a credible 72% win rate on eleven positions. Not concentrated enough to trigger ALL-IN and not large enough to trigger ANOMALOUS, but posture and discipline are both in the 88-89th percentile range. Probably a skilled trader; not yet at Unusual Predictions scale.

#10  0x37c1...c53a ·  $1.70M monthly profit

28 resolved positions, 42.9% win rate, minus $268,894 all-time PnL, Edge Score 58.7 (posture 66.6, conviction 41.1, discipline 63.5). Negative on the all-time record. Kelly sizing efficiency 0.90. Skill Brier plus 0.037 (modestly beats own baseline; calibration is not the problem here). UW flags: none of five. Verdict: the monthly- leaderboard appearance is driven entirely by unrealized gains on open positions. The realized track record is net-negative. A wallet that will show up on any flow-based surveillance product this month, but which Edge Score does not rate highly and which a disciplined copy-trader should skip.

What the ten reveal as a group

Pattern 1. Five of ten match two or more UW criteria. HorizonSplendidView (three), reachingthesky (three), beachboy4 (three), majorexploiter (four), and 0x777d...5143 (two) all have multi-criterion matches. The other five either fail the scoring filters (#1 and #10 hit zero UW flags) or cannot be analyzed because their resolved-position history is empty (#7 and #8). The single edge case is #4, which has Edge Score 99.1 on the strength of one unresolved position.

Pattern 2. Four of ten match three or more UW criteria.HorizonSplendidView, reachingthesky, beachboy4, and majorexploiter each hit three or more proxies. Three of the four (HorizonSplendidView, reaching- thesky, beachboy4) have enough resolved positions (8, 9, 21) to support a skill read. Majorexploiter clears four flags on only three resolved positions and is the one case in the four where the flow signal is strong but the skill signal is not yet statistically supportable. Any flow surveillance product should surface all four; a skill-scoring layer confirms three of them and flags the fourth as pending more resolved bets.

Pattern 3. Every analyzed wallet's biggest event is a sports contract. Not a 2024 election market. The monthly leaderboard in April 2026 is dominated by sports-specialist books, not political whales. The election-cluster narrative that dominated Polymarket coverage in late 2024 is a historical artifact; the current month is Premier League, Champions League, NBA, and NHL. Any surveillance product whose criteria are optimized for the 2024 political cluster is currently missing the center of the action.

Pattern 4. Monthly profit and all-time skill are not the same distribution. The monthly rank-1 wallet has Edge Score 47.9 (below the cohort median). The monthly rank-4 wallet has Edge Score 99.1 on one unresolved position. The monthly rank-10 wallet has a net-negative all-time record. Flow-based monthly rankings are heavy-tailed and include wallets whose recent number is driven by unrealized PnL or by a single big position. Edge Score, fit on resolved history, does not move with the month. The two measures disagree more than a naive reader would expect.

Pattern 5. Kelly sizing efficiency is mostly below 1. Of the eight wallets where Kelly sizing computed cleanly, only one (reachingthesky, 1.02) is sized at or above Kelly. The other seven sit between 0.14 and 0.90. This is what one should expect from rational operators in this market. Full Kelly is catastrophic under Polymarket's fat-tailed payoff distribution (Hill alpha around 1.28, so the growth-rate variance is effectively undefined); a fractional Kelly in the 0.2-to-0.5x range is the defensible choice. Reading the cluster as "under-sized" misstates the distribution. The right read is that leaderboard whales are applying a correct fat-tail discount; the stereotype that they over-size their book does not hold here.

The testable hypothesis

A falsifiable claim for any reader: wallets at the top of the Polymarket monthly profit leaderboard during a given month will have lower Edge Score values than wallets in the top 100 of the all-time cohort ranked by the same composite. In this ten-wallet cut, the direction is consistent; the mean Edge Score across the eight analyzable monthly-leaders is 77.6, while the top 100 all-time file mean is 97.2. The gap of roughly 20 points is directionally supportive but not statistically decisive at n=8, given the fat-tailed distribution the score is drawn from. The claim needs a wider window (multiple months, hundreds of wallets) to be confirmed or refuted. Any reader can begin that check by running arbitrary leaderboard addresses through convexly.app/tools/polymarket-wallet- analyzer and comparing against the published top-100 file at convexly.app/research/edge-score-top100.json.

What to do with this

The operational frame, stated plainly: monthly-leaderboard whales are tail draws from a Hill-alpha-1.28 distribution. Edge Score is fit on the body of the wallet distribution, the 8,656 wallets with five or more resolved positions. By construction, a composite fit on the body does not and should not predict who wins any given month. Skill scoring is a filter on the middle of the distribution, not a predictor of the tail. Any product that claims to predict the monthly leader is fitting noise.

For a reader who trades Polymarket: do not copy- trade off the monthly leaderboard. Half of the current top-10 either hit zero structural-skill criteria or cannot be analyzed on resolved history. The four wallets that do match three or more criteria are distinguishable, but even there the samples are small (n=8, 9, 21, and 3 resolved positions). The operational rule is the same as the five rules in the operator's takeaway: run the pillar breakdown, skip anything without a resolved track record, size fractional Kelly, expect the ones you most want to copy to be survivors rather than predictors.

For a reader running a surveillance product: the right use of a skill-scoring layer is to deprioritize the zero-flag wallets (#1 and #10) and the unresolved-history cases (#7 and #8), not to try to rank the high-confidence matches. A flow filter that returns four-out-of-ten at three or more criteria is doing useful work on the false- positive side; a skill layer makes that filter tighter by explicit exclusion of untestable candidates. That is the realistic contribution of Edge Score on top of an Unusual Predictions feed: negative filtering of the tail, not positive prediction of it.

For Convexly: any surveillance product that wants the skill axis as a complement to the flow axis can consume the Edge Score API directly. The live analyzer at convexly.app/tools/polymarket-wallet-analyzer takes any Ethereum address and returns a full pillar breakdown in under thirty seconds. The methodology is frozen-coefficient and reproducible with one command from the research page.

Disclosure

Equity-options and crypto flow trackers defined the flow-plus- skill surveillance category in prediction markets. This post is Convexly's independent analysis. No partnership between Convexly and Unusual Whales exists as of 2026-04-20. The ten addresses were taken directly from the Polymarket monthly profit leaderboard at polymarket.com/leaderboard/overall/monthly/profit on 2026-04-20. Every per-wallet number traces to a live response from convexly.app/tools/polymarket- wallet-analyzer captured on the same date. The 8,656-wallet cohort and the Edge Score V3b coefficients are frozen in the methodology paper at convexly.app/research/edge-score-methodology-v1.

Survivorship note. Both the 8,656-wallet training cohort and the top-10 monthly leaderboard are drawn from wallets currently active and currently visible. Wallets that blew up and stopped trading do not appear in either set. Ruin is not observed here; the distribution of outcomes is conditioned on survival. Any "skill" claim in this post is a claim about the surviving subsample, not a claim about the full population that ever traded Polymarket.