Gabigol

Wallet: 0x885278f0e304bc2d53f805af2ab779cb6011c569 Window: 2026-05-15 to 2026-06-17 (34 calendar days, 34 active) Universe: 324,515 trades across 27,089 unique markets, $1,446,953 gross buy notional

The Punchline

This is a high-frequency, both-sides market-making bot running continuously across Polymarket's 5-minute Bitcoin, Ethereum, and Solana Up/Down markets. It buys both sides of the same market simultaneously, locking in a guaranteed spread whenever the combined entry cost is below $1.00, while also tilting directionally - putting 3x or more on one side when its signal is confident - and winning those directional bets 82.9% of the time.

The spread-capture layer earns roughly $34,785 in guaranteed P/L from 9,313 sub-$1.00 paired markets. The directional alpha layer earns the remaining realized trading P/L. And on top of the trade-level returns sits $6,608 of Polymarket liquidity rewards, bringing the account total to $36,798 over 34 days.

This is not a simple market maker content to collect spread. The 28-percentage-point gap between the 1x-1.5x dominant-side win rate (53.2%) and the 3x+ dominant-side win rate (82.9%) is the most important number in this dataset. It means the bot's asymmetric sizing decisions are highly predictive, not random. It has a signal.

What He Trades

The universe is Polymarket's short-duration Up/Down markets on three crypto assets, all at 5-minute resolution:

• btc-updown-5m-* - Bitcoin 5-minute windows (primary workhorse)
• eth-updown-5m-* - Ethereum 5-minute windows
• sol-updown-5m-* - Solana 5-minute windows

The CSV sample confirms all three slugs appear simultaneously in the same calendar windows, with the bot executing trades on multiple assets per 5-minute block. There is no evidence of 15-minute, 1-hour, or any other duration in the trade sample or slugs.

All 324,515 trades classify as Crypto. There is no political, sports, or non-crypto activity in this book.

The Order of Operations - One Market, Trade by Trade

The clearest illustration from the CSV sample is the *2:35AM-2:40AM ET* window on June 14, where the bot simultaneously entered Bitcoin Down, Ethereum Up (wrong side), and Solana Up (wrong side) in the same 5-minute block:

Walk-through: In this window, Bitcoin had clearly been moving down before 06:36. The bot allocates heavily to BTC Down (the dominant leg) at prices ranging from $0.77 to $0.93 - that is a near-certainty bet on a market already trending toward resolution. Simultaneously, it buys BTC Up at $0.14 - the insurance/hedge leg. The combined VWAP for BTC in this market is approximately ($0.14 on the small leg + $0.86 weighted average on the dominant leg) - under $1.00, so the guaranteed spread is locked.

ETH and SOL show the same pattern: a heavy dominant-side position at high prices (both Down, at $0.93), and a token long-shot allocation on the other side ($0.06-0.13). The markets resolved with BTC Down winning, ETH Down winning, and SOL Down winning - all dominant legs correct.

This is the structural template: find a 5-minute window where asset prices have moved enough that one side is priced $0.70-0.95, buy that side heavily, and buy the opposite longshot side for the guaranteed spread credit. If the signal is right (82.9% of the time at 3x+ asymmetry), the dominant leg pays out and the paired entry was sub-$1.00 so the whole trade is profitable regardless.

Why It Works - The Math

There are two stacked positive-EV mechanisms:

Mechanism 1: Guaranteed spread from paired entries

paired_cost = Up_VWAP + Down_VWAP
guaranteed_payout = 1.00 (exactly one side always wins)
guaranteed_spread = 1.00 - paired_cost  (when paired_cost < 1.00)

For the 9,313 markets where paired_cost < 0.97:
  median paired_cost = 0.969 (overall), lower in sub-0.97 subset
  guaranteed_spread per paired share ≈ 0.03-0.08 cents
  This is pure risk-free profit - no directional exposure needed.

pnl_decomp.spread_pnl ≈ +$34,785 (computed from paired markets)

Mechanism 2: Directional alpha from asymmetric sizing

Dominance bucket  | Count  | Dom_win_rate | Expected (pure MM)
1.0-1.5x          | 2,210  | 53.2%        | ~50-55%  (no signal)
1.5-2.0x          | 1,604  | 55.6%        | ~50-55%  (weak signal)
2.0-3.0x          | 1,961  | 60.6%        | ~55-60%  (moderate signal)
3.0x+             | 10,391 | 82.9%        | ~50-55%  (STRONG SIGNAL)

EV per 3x+ market (illustrative, $50 dominant / $10 hedge):
  Win case (82.9%):  +$50 payout - $50 cost = $0 on dominant
                     -$10 lost on hedge
                     +$3 guaranteed spread
                     Net: -$7 on wrong (hedge) shares, +$3 spread = small loss offset by...
  Actually: On $50 dominant at $0.85 → 58.8 shares
            Win: +$58.8 settlement, cost -$50, net +$8.8
            Hedge at $0.15 → 66.7 shares, cost -$10
            Hedge settles at $0: -$10
            Net: +$8.8 - $10 = -$1.2... 
  BUT: The spread component: paired_cost = $0.85 + $0.15 = $1.00 exactly
       When paired_cost < $1.00 (sub-$0.97 subset), guaranteed profit exists

Key insight: The dominant leg's ROI exceeds the hedge cost precisely 
because the 82.9% win rate dramatically exceeds the implied probability 
embedded in the sub-$0.50 pricing of the longshot leg.

The hedge leg is priced at $0.07-$0.25 (longshot), implying 7-25% win probability. But the dominant leg wins 82.9% when the bot tilts 3x+. The gap between the longshot's pricing (implying ~20% win for the other side) and its actual outcome (winning only 17.1%) is the spread the bot collects beyond the guaranteed entry-cost mechanism. The market is pricing the longshot too expensively relative to how often the 3x+ bot tilts get resolved in the dominant direction.

Phase 1 - Trader Profile

Scale and Activity

The 0 SELL trades is the first major structural fact. This is a hold-to-resolution book. Every position is entered and held until the 5-minute window closes and Polymarket settles the outcome. There is no SELL engine, no active exit management, no taking profits before resolution. This is categorically different from SirMartingale, who actively sold into post-move rallies.

Trade Size Distribution

The distribution is highly skewed: the median fill is only $1.38, but the top 5% of fills carry 54.3% of capital. This is a power-law size distribution - thousands of small fills (the longshot hedge legs at $0.05-$2.00) and a smaller number of large fills (the dominant-side near-certainty bets at $10-$390).

Execution Signature

80% of consecutive fills are under 10 seconds apart. The median gap is zero - same-second multi-leg fan-outs are the norm. This is unmistakably automated. No human places 9,500 fills per day at sub-second cadence. The 10-second mean (higher than the 0-second median) comes from the gaps between market windows - when one 5-minute window ends, there is a brief pause before the next opens.

Trading Hours

The hourly trade histogram is remarkably flat:

The range is 11,984 (hour 17) to 14,473 (hour 9) - a 21% spread across 24 hours. This is not a sleep-window bot. It runs essentially flat around the clock. The slight dip in hours 15-20 UTC (US afternoon/evening) likely reflects periods of lower Solana or ETH Up/Down market activity rather than any deliberate scheduling.

Archetype

This is a dual-mechanism market maker: guaranteed spread from paired entries on the structural layer, and a directional tilt at 3x+ asymmetry that wins at 82.9% rate on the signal layer.

Phase 2 - Core Strategy Identification

Both-sides participation: 59.7%

16,170 of 27,089 markets have both Up and Down buys from this wallet. That is the primary classification signal. Compare to SirMartingale's 0% - the strategies are structurally opposite.

The 40.3% of markets with only one side purchased falls into two explanations:

1. The bot entered only one side when the signal was extremely strong (one side already near $0.99, other side at $0.01 - no meaningful paired entry possible)
2. Market liquidity only permitted one side to be filled within the 5-minute window

Classification: A (Both-Sides Spread Capture) + B (Directional Betting)

The bot is not purely a market maker - the 3x+ dominance bucket's 82.9% win rate proves a directional signal. It is not purely a directional bettor - the 59.7% both-sides rate proves a structural spread-capture engine. The correct classification is a directionally-informed market maker: it enters both sides for the guaranteed spread, but tilts the allocation asymmetrically when confident.

It is NOT:

• A latency arbitrageur (no SELL trades, no exit before resolution)
• A copy-trader (no temporal lag pattern with another wallet)
• A DCA accumulator (entries are clustered in single 5-minute bursts, not distributed over time)
• A pure penny-floor bot (entries span $0.01 to $0.99, no single-cent concentration)

Phase 3 - Dominance Ratio Analysis

This is the most diagnostic phase for this wallet.

The discontinuity between the 2.0-3.0x bucket (60.6%) and the 3.0x+ bucket (82.9%) is sharp and highly significant. It is not a linear progression - it is a threshold effect. When the bot's allocation ratio crosses 3x, something qualitatively different is happening in its decision-making. The dominant side is winning 5 out of 6 times at that asymmetry level.

The mean paired cost rises monotonically with dominance ratio: from $0.760 at 1x-1.5x to $0.960 at 3x+. This makes sense structurally: when the dominant side is near-certain (say $0.90), the opposite longshot is near $0.10, and the combined cost is near $1.00. At equal allocations (1x), the bot is often buying at moderate prices on both sides (e.g., $0.45 + $0.45 = $0.90 paired cost). So high-dominance markets inherently have higher paired costs, lower guaranteed spread, and are held together by the directional accuracy of the signal.

The 3x+ bucket has the largest market count (10,391) and the highest mean paired cost ($0.960). The bot is most active in the highest-confidence markets, which are also the ones that are approaching resolution. This is consistent with a signal that reads real-time price movements in the underlying assets and identifies near-resolved windows.

The median second-side lag of 81 seconds (time between entering the first and second side of a market) is meaningful. It is short enough to be within the same 5-minute window, but long enough to suggest the bot sometimes enters the dominant side first, observes price movement, and then adds the hedge leg - rather than entering both simultaneously.

Phase 4 - Entry Price Analysis

Price band distribution by capital deployed:

The $0.90-$1.00 band holds 60.2% of all buy capital ($871,636) and generates slightly negative trading P/L (-$3,286). This is the dominant near-certainty leg of paired trades - the bot is buying at $0.90-$0.99, winning 96.4% of the time, but the per-share profit is tiny ($0.01-$0.10 at settlement) and the combined cost of both legs means the spread-P/L lives in the guaranteed-entry math, not in the per-trade directional return.

The $0.00-$0.30 bands are the hedge/longshot legs of those same paired trades - tiny capital (2.8-3.7% each) but positive P/L because when they occasionally win, the payouts are large multiples of the small investment.

Sub-bucket concentration check: There is no single-cent concentration. The per-cent histogram (implied by 324,515 trades across a wide distribution) shows broad usage of price points. This is not a penny-floor bot or a single-tick stale-price sniper. The bot is bidding opportunistically wherever the orderbook offers paired entries within its target paired-cost range.

Phase 5 - Category and Vertical Breakdown

The standard category framework collapses to one row:

The 36.6% overall win rate on resolved buys is not a good metric for this strategy in isolation - it averages together the 5.8% win rate on longshot hedge legs and the 96.4% win rate on near-certainty dominant legs. The meaningful metric is paired P/L, not per-trade win rate.

Within the Crypto vertical, the CSV confirms three active asset sub-verticals: BTC, ETH, and SOL. All appear to be running in 5-minute windows exclusively. The top-volume markets by the data (best/worst market lists) are split across ETH and BTC, with SOL appearing in the best-by-PNL list (Solana Up or Down - June 16, 4:15AM-4:20AM ET: +$560 on $47 volume, 39/39 wins - a pure dominant-leg win where the SOL market was near-resolved).

Assessment: MODEST on trading ROI (+2.09%). But when rewards income is included, total account ROI on $1.45M notional is +2.54%. At this volume and capital turnover, the absolute dollar returns (+$36,798) justify the infrastructure.

Phase 6 - Timing and Execution Analysis

Hourly P/L distribution:

Hour 12 (noon UTC, 8am Eastern) is the only hour with a meaningful negative P/L (-$1,050). This is the US market open - a period of high volatility in crypto markets where rapid price moves may result in the bot's dominant-side bets resolving incorrectly more often. The higher win rate at hours 3-6 UTC (40.3%, 40.6%) suggests the early-morning UTC hours are when paired entries at sub-$0.97 cost are most available (lower competition, slower markets, more mispriced longshots).

Day-of-week breakdown:

Wednesday leads in absolute P/L (+$7,774, 3.95% ROI). Saturday is the weakest despite the highest trade count (58,561 trades, 0.31% ROI, 42.0% win rate). The Saturday anomaly deserves attention: win rate is the highest of any day (42.0%) yet ROI is the lowest (0.31%). This implies Saturday has more longshot-heavy books - high win rates on small hedge positions but many dominant-leg losses, perhaps because Saturday crypto markets are more volatile and the directional signal is less reliable.

Accumulation window per market:

Each market appears to be entered in a burst of 10-145 fills within a few minutes, all within the same 5-minute resolution window. The bot does not return to a market after it resolves. The time from first to last fill within a single market is typically 1-5 minutes. This is a one-touch, hold-to-resolution structure.

Phase 7 - Filter Experiments

The filters reduce absolute P/L but improve ROI efficiency in two cases. The price $0.30-$0.70 filter raises ROI from 2.09% to 4.94% by focusing on coin-flip zone trades and excluding both the near-certain dominant legs (large capital, thin margin) and the longshot hedge legs (small capital, low win rate). The exclude-worst-hours filter marginally improves ROI to 2.38% while removing only $1,101 of P/L from 55,015 trades.

The high-conviction filter (dom ≥ 2x, dominant leg only) actually reduces ROI to 1.81% despite a 64.3% win rate, because it includes the capital-heavy dominant-side bets in the 2-3x bucket where paired cost approaches $1.00 and per-share profit is minimal.

The combined filter achieves the best ROI (5.22%) but cuts absolute P/L in half.

Conclusion for filters: The base strategy's ROI is low (2.09%) because it intentionally deploys large capital on thin-margin near-certainty bets as part of the paired structure. Filtering for "better ROI" destroys the paired entry mechanics that generate the guaranteed spread. The filters are structurally misaligned with how this strategy works.

See Filters tab for full commentary.

Phase 8 - Rolling Window Consistency

The rolling window trajectory shows a significant early drawdown. The cumulative account P/L went from +$1,710 on May 15 down to -$5,083 by May 17 - a ~$6,800 peak-to-trough swing in the first 3 days. This was the wallet's worst stretch. After May 20, the account P/L turned positive and climbed steadily, reaching $36,798 by June 17.

Week-by-week:

Week 20 (May 15-17):   +$462 - slow start, early drawdown period
Week 21 (May 18-24):   +$6,730 cumul $7,192
Week 22 (May 25-31):   +$3,817 cumul $11,009
Week 23 (Jun 1-7):     +$7,198 cumul $18,207 - best week in absolute terms
Week 24 (Jun 8-14):    +$6,722 cumul $24,929
Week 25 (Jun 15-17):   +$5,261 cumul $30,190 - truncated week, strong finish

The ramp pattern (weak start, accelerating later) is consistent with a market maker that builds out its paired-market coverage over time, or whose signal calibration improved during the observation window. Week 23 (June 1-7) was the peak performance week despite not being the highest-volume week.

Phase 9 - P&L Decomposition

The hedge tax of $199,748 is the cost of buying the losing sides of markets where the dominant side won. This is enormous relative to the $30,190 net P/L - it means the bot spends heavily on "insurance" that mostly expires worthless. But the guaranteed spread from paired entries offsets much of this: when paired_cost < 1.00, the hedge's cost is already baked into the locked-in spread.

The true decomposition is:

Guaranteed spread (sub-$1.00 paired entries):  +$34,785 (estimated)
Directional wins on dominant legs:             Large positive
Losing directional bets (both sides):          Large negative
Net trading P/L:                               +$30,190
Liquidity rewards:                             +$6,608
Account total:                                 +$36,798

The liquidity rewards at $6,608 represent 18% of the account total. At $194/day, this is consistent with Polymarket's known LP reward structures for high-volume market makers operating in the 5-minute Up/Down series.

Phase 10 - Strategy Specification

One-sentence summary: A 24/7 automated market maker that simultaneously buys both Up and Down sides of 5-minute BTC, ETH, and SOL Up/Down markets, locking in guaranteed spread on sub-$1.00 paired entries while tilting allocation asymmetrically up to 10x+ on the directionally-confident side, winning those tilted bets 82.9% of the time via an undisclosed signal.

What works: The 3x+ dominance bucket (10,391 markets, 82.9% win rate). The spread-capture layer on the 9,313 sub-$0.97 paired markets. The liquidity rewards from continuous 24/7 market presence. Wednesday market conditions (3.95% ROI vs 0.31% on Saturday).

What drags: Hour 12 UTC (US market open, -$1,050 P/L). Saturdays (high trade volume, lowest ROI). The $0.90-$1.00 capital band (-$3,286 trading P/L on $871K deployed) - these near-certainty bets earn almost nothing directionally and rely entirely on the paired entry math for profitability.

What replicators must understand: The strategy's 2.09% trading ROI on $1.45M of notional is thin. The absolute profit ($30,190 trading + $6,608 rewards = $36,798) requires large-scale capital deployment. The undisclosed signal driving 82.9% accuracy on 3x+ tilts is the core IP and is not derivable from the trade CSV alone.

Wallet: 0x885278f0e304bc2d53f805af2ab779cb6011c569 Window: 2026-05-15 → 2026-06-17 (34 active / 34 calendar days) Methodology: Cash-flow P/L = -buy_usdc + sell_usdc + remaining_share_payout. Resolved shares settle at $1 (win) / $0 (loss); open positions marked at last price.

Phase 1 - Trader Profile

Scale

Trade-size distribution (USDC per fill)

Inter-trade gap, same (market, outcome)

Phase 2 & 3 - Both-Sides Participation, Dominance Curve

• Both-sides rate: 59.69% (16,170 of 27,089 markets)
• Median paired cost: $0.9694
• Mean paired cost: $0.8966
• Paired cost % under $1.00: 57.6%
• Paired cost % under $0.97: 50.2%
• Median 2nd-side hedge lag: 81s

Dominance buckets

Phase 4 - Entry-Price Analysis

Phase 5 - Category & Vertical Breakdown

Phase 6 - Timing & Execution

Net P/L by hour (UTC)

Phase 8 - Rolling Window Consistency

• Rolling 7-day windows green: 34 of 34 (100.0%)
• Rolling 7-day P/L range: +$462 → +$10,322
• Rolling 15-day windows green: 34 of 34 (100.0%)
• Rolling 15-day P/L range: +$462 → +$16,447

Weekly P/L

Phase 9 - P/L Decomposition

Phase 10 - Top Markets by Volume

Top 10 winners by P/L

Top 10 losers by P/L

Report generated 2026-06-22 08:24 UTC.

Wallet: 0x885278f0e304bc2d53f805af2ab779cb6011c569 Window: 2026-05-15 to 2026-06-17 Baseline: 324,515 BUY trades · 36.6% WR · $1,446,953 deployed · +$30,190 trading P/L · +2.09% ROI Account total (including rewards): +$36,798

The headline result

Two filters marginally improve ROI efficiency. Three filters reduce absolute P/L substantially. One is a structural no-op. None of them capture the strategy's actual edge.

The reason is architectural. This bot's edge comes from two sources: (1) guaranteed spread from paired entries where Up_VWAP + Down_VWAP < $1.00, and (2) directional accuracy on the heavily-tilted 3x+ markets. Both sources are largely invisible to per-trade filters that look at individual buy prices or individual market win rates in isolation. A filter that removes the $0.90-$1.00 band "improves" ROI by stripping out low-margin dominant legs - but those legs are load-bearing components of paired trades whose other half is the profitably-priced longshot leg.

The single most actionable filter insight: the hour-exclusion filter for hour 12 UTC provides small, genuine lift at negligible cost. Everything else is noise or damage.

Filter results table

Filter-by-filter commentary

1. Price $0.30-$0.70 (the "sweet spot") → DESTRUCTIVE

The $0.30-$0.70 band contains 90,565 trades on $227,849 of capital, with a 47.5% win rate and +4.94% ROI. That looks better than the 2.09% baseline ROI. But applying this filter costs $18,941 of absolute P/L - 63% of the baseline.

Why? Because this filter strips out two critical components of the strategy simultaneously:

First, it removes the near-certainty dominant legs ($0.70-$1.00 range, 60.2% of all capital). Those large-clip dominant-side bets are where the bot places its highest-confidence directional tilts. The 3x+ bucket's 82.9% win rate lives primarily in this zone - markets where one side is priced $0.85-$0.95 because the outcome is almost certain. Removing them eliminates the bulk of the capital-weighted directional alpha.

Second, it removes the deep longshot hedge legs ($0.00-$0.30 range, ~10% of capital). Those small clips at $0.05-$0.25 are the sub-$0.20 purchases that form the cheap "other side" of high-dominance paired entries. Their occasional wins at 6x-20x the entry price contribute meaningfully to absolute P/L. Removing them severs the pairs.

The $0.30-$0.70 zone that remains is the bot's most contested territory - moderate-dominance markets where paired entries hover near $0.90-$0.95 combined cost and directional accuracy is only 53-61%. This is the weakest part of the strategy, not the strongest. Filtering down to it improves ROI% by concentrating on a smaller capital base, not by finding better trades.

Do not apply the $0.30-$0.70 filter to this strategy.

2. High-conviction filter (dom ≥ 2x, dominant leg only) → DESTRUCTIVE

This filter keeps only the dominant leg of markets where the dominance ratio is 2x+. It covers 95,507 trades (the dominant-side fills in the 2x+ buckets: 1,961 markets at 2-3x and 10,391 markets at 3x+), capturing $928,573 in capital at a 64.3% win rate and +1.81% ROI.

The ROI is worse than baseline (1.81% vs 2.09%) despite the dramatically higher win rate (64.3% vs 36.6%). This paradox has a clean explanation: the dominant leg at high dominance ratios is priced $0.80-$0.97 per share, so winning returns only $0.03-$0.20 per share. The capital-weighted P/L from these large clips is thin. When you strip out the paired structure and look only at the dominant leg in isolation, you see a lot of correct directional bets that earn very little per dollar deployed.

The filter as designed (dominant leg only, dom ≥ 2x) eliminates the hedge tax by definition - you're not counting the losing hedge legs. But it also eliminates the guaranteed spread from the paired entry math. You can't have the spread without both legs.

This filter is structurally misaligned with a paired entry strategy. It would be more useful for a pure directional betting book.

3. Top category filter (Crypto only) → NOT APPLICABLE

100% of trades are Crypto. The filter is identity-equivalent to baseline. P/L: +$30,190. ROI: +2.09%. Delta: $0.

Single-vertical book. The filter framework's category dimension is meaningless here.

4. Exclude worst 4 hours filter → MEANINGFUL LIFT (small)

The four worst hourly P/L buckets are hours 1, 12, 13, and 20 UTC. Excluding them removes 55,015 trades and $223,842 of capital, at a cost of only $1,101 in P/L. The remaining 269,500 trades generate $29,089 at 2.38% ROI - a genuine 29 basis point improvement over baseline.

The lift is genuine but small in absolute terms. Hour 12 UTC (noon UTC, 8am Eastern US) is the meaningful outlier: -$1,050 P/L on 14,119 trades. This is the US equity market open, when BTC/ETH volatility spikes and the bot's near-certainty calls may be slightly less reliable. The three other excluded hours (1, 13, 20 UTC) have small positive or near-zero P/L and their exclusion is marginal.

Recommendation: Excluding hour 12 UTC alone (not all four) would capture most of this lift with a smaller reduction in trade count. The bot is presumably 24/7 by design; selective hour suppression would require deliberate scheduling changes.

5. Combined filter (price $0.30-$0.70 + exclude worst hours) → DESTRUCTIVE (net negative)

The combination achieves the highest ROI in the filter battery at 5.22% on $193,121 of capital. But it cuts absolute P/L from $30,190 to $10,079 - a $20,111 reduction. As a capital-efficiency exercise on a theoretical sub-portfolio, this looks interesting. As a recommendation for this strategy, it destroys two-thirds of the profit by gutting the paired structure.

The combined filter's only legitimate use case is for a replicator who wants to build a simpler directional-only sub-strategy targeting the $0.30-$0.70 zone of this wallet's activity, without attempting to replicate the full paired market-making engine. Even then, the $10,079 P/L on $193K capital represents a thin absolute return on meaningful operational complexity.

What filters would genuinely help (if you had the right data)

The standard filter battery is structurally misaligned with this strategy. The dimensions that would actually improve selection require data outside the trade CSV:

The dominance ≥ 3x filter (keeping both legs of those markets) is the one genuinely executable improvement. The 3x+ bucket contains 10,391 markets where the dominant side wins 82.9% of the time. If a replicator could identify and enter only those markets (both sides, maintaining the paired structure), the strategy would generate better capital efficiency than the full 27,089-market book.

Bottom line for replication

Three concrete recommendations from filter analysis:

1. Exclude hour 12 UTC from the bot's activity schedule. It costs $1,050 in P/L on 14,119 trades. The operational disruption is minimal for a fully automated system, and the 8am Eastern open's volatility is a known execution risk for market makers in short-duration windows.

1. Do not apply the $0.30-$0.70 price filter. It improves the optics (4.94% ROI) while destroying 63% of the absolute profit by removing the load-bearing components of the paired structure.

1. Focus replication capital on the 3x+ dominance bucket. That bucket has 10,391 markets, 82.9% dominant-side accuracy, and is where the bot earns most of its directional alpha. Building a smaller version of this strategy that only enters markets where the signal is at 3x+ confidence would achieve better capital efficiency at lower volume.

The real filter that matters for this strategy is not price band or hour exclusion. It is the directional signal filter - the model that decides when to tilt 3x+ on one side. That model is the core IP and is not derivable from the trade data.

Source wallet: 0x885278f0e304bc2d53f805af2ab779cb6011c569 Strategy: Both-sides paired market-making with directional tilt on short-duration crypto Up/Down markets Reference book: $1,446,953 BUY notional · +$30,190 trading P/L · +$6,608 rewards · +$36,798 account total over 34 days

One-paragraph operator brief

Build a 24/7 automated market maker that simultaneously enters both Up and Down sides of Polymarket's 5-minute BTC, ETH, and SOL Up/Down markets. Target markets where the combined VWAP of both sides is below $0.97, locking in at least 3 cents of guaranteed spread per paired share. When your directional signal is highly confident, allocate 3x-10x+ more capital to the favored side - the reference wallet achieves 82.9% dominant-side accuracy at 3x+ asymmetry. Hold all positions to resolution (no active exits). Run continuously with the only exception being hour 12 UTC (US market open), which has negative trading P/L. Earn baseline spread P/L from the guaranteed-entry math, supplemental directional P/L from accurate 3x+ tilts, and potential Polymarket LP rewards from continuous high-volume market presence. Expect 2-5% ROI on deployed notional plus rewards income at scale.

1. Market Selection

Why 5-minute windows exclusively: Longer windows have more time for the orderbook to equilibrate, reducing the paired-entry spread available and giving directional signals more time to be arbitraged away. 5-minute windows cycle frequently (12 per hour per asset), maximizing the number of entry opportunities and ensuring positions turn over rapidly.

Why BTC + ETH + SOL simultaneously: All three appear in the CSV sample with concurrent entries in the same 5-minute blocks. Running all three assets multiplies the number of available markets per hour threefold, allows capital diversification across assets, and may capture windows where one asset's volatility is low (smaller spread available) while another's is high (wide spread).

2. Entry Logic

The entry decision has two layers: the structural paired-entry gate and the directional tilt decision.

Layer 1: Paired entry gate (spread capture)

def should_enter_paired(market, up_book, down_book):
    """
    Check if both-sided entry locks in guaranteed spread.
    """
    # Get best available ask on each side
    up_ask = up_book.best_ask()     # price to BUY Up
    down_ask = down_book.best_ask() # price to BUY Down
    
    paired_cost = up_ask + down_ask
    
    # Gate: only enter if paired cost locks in meaningful spread
    if paired_cost > 0.97:
        return False  # spread too thin or negative
    
    guaranteed_spread = 1.00 - paired_cost
    # At $0.97 paired cost, guaranteed spread = $0.03 per paired share
    # Reference wallet's median paired cost = $0.969 (both-sides markets)
    
    return True, guaranteed_spread, up_ask, down_ask

Layer 2: Directional tilt signal

This is the undisclosed core IP. From the dominance data, we know that when the bot allocates 3x+ to one side, it wins 82.9% of the time. The signal that drives this asymmetric allocation is not encoded in the trade CSV. Observable proxies that may approximate it:

def compute_directional_tilt(asset, window_start_time):
    """
    Compute confidence-weighted allocation ratio for Up vs Down.
    The reference wallet's 3x+ markets win at 82.9% -- this function
    needs to identify those near-certain outcomes.
    
    Observable signals to explore:
    1. Recent price trend: if BTC moved +0.3% in the last 2 minutes,
       the 5-minute window is likely to close "Up" -- tilt heavily Up
    2. Market price itself: if one side is already at $0.85+, the market
       has already priced the near-certain outcome -- follow the market
    3. Time within window: if 4 minutes have elapsed and one side is $0.90,
       the remaining time has very little probability of reversal
    4. Realized volatility: low vol → more predictable direction
    """
    # Placeholder: use market price as the directional signal
    up_price = market.up_side.mid_price
    down_price = market.down_side.mid_price
    
    if up_price >= 0.85:
        dominant_side = "Up"
        tilt_ratio = up_price / (1.0 - up_price)  # e.g., 0.90/0.10 = 9x
    elif down_price >= 0.85:
        dominant_side = "Down"
        tilt_ratio = down_price / (1.0 - down_price)
    else:
        dominant_side = None
        tilt_ratio = 1.0  # equal allocation (the 1x-1.5x bucket)
    
    return dominant_side, tilt_ratio

Entry price discipline: The bot does not anchor to a single price point. It walks the orderbook, filling at whatever prices are available up to the target notional. The CSV shows entries spanning $0.01 to $0.99 across 324,515 fills, with no concentration at a single tick.

3. Sizing Model

The reference wallet's size distribution has a bifurcated structure: many small fills (the hedge legs) and fewer large fills (the dominant legs).

Dominant leg sizing:

def size_dominant_leg(bankroll, market, tilt_ratio, confidence):
    """
    Size the high-confidence dominant side.
    """
    # Base allocation as fraction of bankroll
    base_pct = 0.001  # 0.1% of bankroll per market at 1x confidence
    
    # Scale with confidence, but cap at 0.4% of bankroll
    confidence_scalar = min(tilt_ratio / 3.0, 4.0)
    target_usdc = bankroll * base_pct * confidence_scalar
    
    # Hard cap: reference wallet's max single fill is $390
    target_usdc = min(target_usdc, 390)
    
    return target_usdc

Hedge leg sizing:

The hedge leg is sized to maintain the paired structure (buy some of both sides) while minimizing capital at risk on the non-confident side.

def size_hedge_leg(dominant_usdc, tilt_ratio):
    """
    Size the hedge/insurance leg.
    Reference wallet's median tilt ratio (3x+) implies:
    if dominant = $30, hedge ≈ $10 (3x ratio)
    if dominant = $90, hedge ≈ $10 (9x ratio, small absolute hedge)
    """
    # Hedge is dominant_usdc / tilt_ratio, with a minimum floor
    hedge_usdc = max(dominant_usdc / tilt_ratio, 0.50)
    return min(hedge_usdc, 50.0)  # cap hedge at $50

Bankroll-scaled sizing guide:

The reference wallet deployed $1.45M over 34 days with a max single fill of $390 and median fill of $1.38. The implied operating bankroll is approximately $150-200K (the wallet turns over its capital roughly 7-10x over the window).

4. Both-Sides Allocation Logic

The core structural innovation of this strategy is the simultaneous both-sides entry. The allocation ratio between dominant and hedge legs is where the directional signal lives.

def allocate_sides(market, signal_confidence):
    """
    Full entry allocation logic for a single 5-minute market.
    
    signal_confidence: float in [0, 1], where 0.82 = 82% confident in Up
    """
    # Step 1: Determine dominant side
    if signal_confidence > 0.60:
        dominant = "Up"
        hedge = "Down"
        dominant_conf = signal_confidence
    elif signal_confidence < 0.40:
        dominant = "Down"
        hedge = "Up"
        dominant_conf = 1.0 - signal_confidence
    else:
        # Low confidence: roughly equal allocation (1x-1.5x bucket)
        dominant = "Up"   # arbitrary for coin-flip markets
        hedge = "Down"
        dominant_conf = 0.50
    
    # Step 2: Compute tilt ratio
    if dominant_conf >= 0.90:
        tilt_ratio = 10.0   # 3x+ bucket, high end
    elif dominant_conf >= 0.80:
        tilt_ratio = 5.0    # 3x+ bucket
    elif dominant_conf >= 0.70:
        tilt_ratio = 3.0    # upper 2-3x bucket
    elif dominant_conf >= 0.60:
        tilt_ratio = 2.0    # 1.5-2x bucket
    else:
        tilt_ratio = 1.2    # 1x-1.5x bucket
    
    # Step 3: Size dominant and hedge legs
    dominant_usdc = size_dominant_leg(bankroll, market, tilt_ratio, dominant_conf)
    hedge_usdc = size_hedge_leg(dominant_usdc, tilt_ratio)
    
    # Step 4: Verify paired entry gate
    up_usdc = dominant_usdc if dominant == "Up" else hedge_usdc
    down_usdc = dominant_usdc if dominant == "Down" else hedge_usdc
    
    return up_usdc, down_usdc, dominant, tilt_ratio

Dominance ratio targets by confidence tier:

5. Exit Strategy

There is no exit. This is the simplest possible exit rule: hold every position to resolution.

def manage_position(position):
    """
    All positions held to market resolution.
    No SELL orders ever submitted.
    Polymarket auto-settles at $1.00 (winner) or $0.00 (loser).
    """
    return "hold_to_resolution"

The reference wallet has 0 SELL trades across 324,515 fills. This is not a coincidence or oversight - it is a deliberate structural choice. The guaranteed spread from paired entries is only realized at settlement, not from active exit. If you sell a winning position before resolution, you forego the settlement payout and break the paired entry economics.

Why no early exit works here: Unlike SirMartingale's strategy (which profits from selling into post-move rallies), this bot's profit comes from (1) guaranteed spread at settlement and (2) correct directional calls that pay out at $1.00. There is no "rally to sell into" in a binary market that will resolve at exactly $1.00 or $0.00.

Stop-loss rule: There is no per-position stop-loss because the maximum loss per market is bounded by the combined entry cost (dominant_usdc + hedge_usdc). In practice this is $10-$450 per market. No individual loss can blow up the book.

6. Hour Scheduling

The overall schedule is effectively 24/7 except for hour 12 UTC. The reference wallet runs at near-uniform volume across all 24 hours, with no detectable sleep window.

Day-of-week adjustments:

Saturday is the weakest day (42.0% WR but only 0.31% ROI on 58,561 trades). Consider reducing dominant-leg allocation by 30% on Saturdays while maintaining the hedge-leg structure to preserve guaranteed spread. Wednesday is the strongest day (3.95% ROI) - no adjustment needed.

7. Operational Requirements

LP rewards eligibility: The +$6,608 in rewards income implies Polymarket has classified this wallet as a liquidity provider. This classification is not automatic. At $1.45M/month of buy notional and 27,089 markets, the wallet qualifies as a high-volume participant. Reaching out to Polymarket's LP program team is necessary to access rewards at scale.

8. Risk Profile

Maximum daily drawdown at reference scale: The worst single day in the 34-day window was approximately -$3,677 (cumulative daily P/L on May 16 was $3,677 below May 15). At reference scale ($1.45M/month), peak daily drawdown is approximately 0.25% of notional. This is exceptionally contained for a strategy of this size.

9. Diagnostic Checklist - Is the Bot Still Working?

Run weekly:

10. What This Playbook Deliberately Omits

The directional signal. The most important component of this strategy - the model that achieves 82.9% accuracy on the 3x+ asymmetric allocations - is not described here because the trade CSV does not reveal it. The playbook describes the structural shell (paired entries, sizing ratios, asset universe, hold-to-resolution exit) that we can observe. The signal is the core IP that cannot be reverse-engineered from settlement outcomes alone.

Active exit management. This strategy holds to resolution. Adding a SELL engine would fundamentally change the mechanics and require recalibration of the entire paired structure.

Leverage or borrowing. The reference wallet operates on spot USDC with no apparent leverage. Adding leverage would convert the bounded per-market loss structure into an unbounded one - destroying the primary risk management property of this strategy.

Cross-market correlation trades. The bot treats each 5-minute BTC, ETH, and SOL window independently. It does not appear to cross-hedge between assets (e.g., going long BTC Up and short ETH Up as a correlation play). Adding such cross-market structure would require a separate signal framework.

Longer-duration windows. The 15-minute, 1-hour, and 4-hour BTC/ETH markets are excluded. Longer windows have lower turnover, more time for information to be incorporated into prices, and therefore less opportunity for the paired-entry spread to be captured at sub-$0.97 combined cost. Do not extend the universe to longer windows without building a separate model for those market types.

The executable version of this strategy is: enter both sides of every 5-minute BTC/ETH/SOL window where paired cost is below $0.97, tilt toward the directionally-confident side using whatever signal you develop, hold to resolution, skip hour 12 UTC, and monitor dominant-side WR weekly. The structural layer is straightforward to build. The signal layer is the life's work.