Bonereaper

Wallet: 0xeebde7a0e019a63e6b476eb425505b7b3e6eba30 Window: 2026-05-01 to 2026-06-14 (45 calendar days, 44 active) Universe: 2,694,878 trades, 37,228 unique markets, $45,064,788 gross BUY notional Account P/L (verified): +$458,492.70 (trading: +$88,606 + rewards/other: +$369,887)

The Punchline

Bonereaper is not a directional trader. The name implies an aggressor collecting carcasses on losing markets, but the data tells a different story: this is a crypto market-making and liquidity-farming operation, deploying capital on both sides of every short-duration BTC, ETH, and SOL Up/Down market, earning the bid-ask spread where available, expressing directional conviction asymmetrically when signal fires, and collecting Polymarket's liquidity-mining rewards at institutional scale.

The trading P/L of +$88,606 is genuine and positive (+0.20% ROI), but it is not the main event. The $369,887 in rewards income is the main event. At $8,407/day in rewards against $2,014/day in trading P/L, the rewards component is 4.2x the trading component. The bot is sized and structured to maximize rewards eligibility, and the directional alpha within the book is a secondary benefit.

What makes this wallet analytically interesting: the directional signal embedded within the market-making framework is real and measurable. When the bot tilts its allocation 3x or more to one side, that side wins 87.6% of the time across 18,692 markets. That is not luck. Something in the decision logic identifies the correct side with strong reliability, and the bot expresses that confidence through heavily asymmetric sizing rather than pure equal-weight coverage.

---

What It Trades

The universe is narrow: exclusively Polymarket crypto short-duration Up/Down markets. The CSV sample shows:

• btc-updown-5m-* (5-minute BTC Up/Down)
• eth-updown-5m-* (5-minute ETH Up/Down)
• sol-updown-5m-* (5-minute SOL Up/Down)
• btc-updown-15m-* (15-minute BTC Up/Down)
• eth-updown-15m-* (15-minute ETH Up/Down)
• btc-updown-4h-* (4-hour BTC Up/Down)
• Occasional daily and hourly horizons (e.g., bitcoin-up-or-down-june-12-2026-12am-et)

The slug patterns confirm a systematic approach: the bot monitors all active short-duration crypto markets simultaneously and deploys fills across all of them. In the June 12 snapshot in the CSV, within 3 minutes (04:27-04:30 UTC), the bot places fills on BTC 5m, ETH 5m, SOL 5m, BTC 4h, and BTC daily windows simultaneously. This is not a one-market-at-a-time strategy.

Asset breakdown by implied activity: BTC dominates volume (largest per-market notionals in top markets list: $9,834-$13,424 per market in top volume), ETH and SOL are secondary. The 37,228 unique markets at ~288 markets per calendar day (37,228/45 = 827 unique markets/44 active days) means the bot enters essentially every eligible BTC/ETH/SOL 5-min and 15-min window across the full observation period.

Both-sides rate: 89.2%. Of 37,228 unique markets, 33,223 had both outcome sides bought. This is the defining structural signal. No directional bettor buys both sides of the same market at 89.2% frequency. This is a market-maker.

---

The Order of Operations: One Market, Trade by Trade

This sequence is from the CSV sample, illustrating the bot's per-market behavior in a live 5-minute window.

Bitcoin Up or Down - June 12, 12:25AM-12:30AM ET (btc-updown-5m-1781238300):

Walk-through:

1. The bot enters the Up side early and heavily (04:27:12-04:27:31), buying 259 shares of Up at prices between $0.67 and $0.82 for roughly $167. This is the dominant side position, established in the first 20 seconds.

1. Simultaneously and shortly after, it starts buying Down at prices ranging from $0.09 to $0.28, spending roughly $35-40 on the losing side. This is the hedge. The paired cost on this market is approximately ($167 Up + $40 Down) / (resolution payout of $1/Up share). The paired cost concept matters here.

1. In the final 90 seconds, the bot makes its largest adds: $84 at $0.93, $100 at $0.95, and $45 at $0.89. These are near-certainty buys at prices that reflect an almost-confirmed Up outcome. The bot is locking in near-guaranteed profit on the tail end of a winning market.

1. The market resolves "Up". The bot's Up shares pay $1.00 each. The Down shares pay $0.00. Net result: profit from the Up side, partial loss on the Down hedge, net positive due to the dominant allocation.

Key observation: The bot is not predicting outcomes ex-ante with perfect accuracy. It is continuously assessing which side is winning and increasing its exposure to that side throughout the window. The 87.6% win rate at 3x+ dominance reflects that by the time the bot has deployed 3x+ asymmetry, the market outcome has often become clear enough to justify that tilt.

---

Why It Works: The Math

The economics have two layers:

Layer 1: Spread capture on balanced fills

When the bot enters a market with roughly equal allocations to both sides (dominance ratio 1.0-1.5x, representing 5,645 markets), the paired cost averages $0.948. This means it pays $0.948 on average to lock in a $1.00 certain payout regardless of outcome.

<pre><code>Spread per paired share set: $1.00 - $0.948 = $0.052 (5.2%) Markets in this bucket: 5,645 Mean paired cost: 0.9483

Expected P/L from pure spread: 5,645 markets * ~$200 average paired notional * $0.052 per dollar = approximately $58,700 over the window from spread capture alone (before directional losses on excess unhedged shares) </code></pre>

This estimate is rough, but it explains a significant portion of the +$88,606 trading P/L.

Layer 2: Directional alpha from asymmetric allocation

When the bot tilts heavily (3x+ dominance, 18,692 markets), the dominant side wins 87.6% of the time. With a mean paired cost of 1.058 in this bucket (paying $1.058 to guarantee $1.00 regardless), the spread capture is negative here. The profit comes entirely from the directional accuracy.

<pre><code>Markets at 3x+ dominance: 18,692 Dominant-side win rate: 87.6% Expected win rate at random: 50% Directional lift: +37.6 percentage points

For a 3:1 dominant/non-dominant allocation with paired cost 1.058: If dominant side wins (87.6%): net gain from direction If dominant side loses (12.4%): net loss from direction

EV per market at 3x dominance (simplified): 0.876 * (gain) + 0.124 * (loss) > 0 Because gain >> loss when 3x allocated to dominant side </code></pre>

Layer 3: Liquidity rewards

$369,887 over 44 days = $8,407/day. At 37,228 markets and $45M deployed, the bot is a top-tier liquidity provider by volume. Polymarket's reward structure favors wallets that are consistently present, quote tight, and provide two-sided depth. This bot does all three at scale.

---

Phase 1: Trader Profile

Scale and activity:

• 2,694,878 BUY trades, zero SELL trades
• $45,064,788 gross BUY notional
• 37,228 unique markets / 37,228 unique events
• 44 of 45 calendar days active
• 61,247 trades per active day; 2,552 per hour average

Trade size distribution:

The distribution is highly power-law at the top: the top 5% of fills carry 63% of capital, and the max is 1,270x the median. This is not bounded sizing. It is a tiered allocation model: many small fills for market presence and spread capture, occasional large fills to express strong directional conviction or anchor deep favorites.

Execution speed:

• Median inter-fill gap: 2.0 seconds
• 79.2% of fills under 10 seconds
• 96.8% under 60 seconds
• Essentially 100% under 1 hour

This is a fully automated bot. The 2-second median gap across 2.69M fills is only achievable with persistent WebSocket connections and automated order management.

Buy vs. sell ratio: 100% BUY, 0% SELL. The wallet holds all positions to resolution. No active exit management. This is pure buy-and-hold market-making, consistent with strategies that earn spread at entry and collect at resolution.

Archetype classification:

---

Phase 2: Core Strategy Identification

Both-sides participation: 89.2%

33,223 of 37,228 markets had both outcomes purchased. This is the single most important structural fact about this wallet. No directional bettor does this. The 10.8% of markets that are one-sided represent either early fills where the second side was never added, markets where the directional signal was overwhelming, or operational gaps.

Paired cost analysis:

• Median paired cost: $1.016
• Mean paired cost: $1.013
• Pct sub-$1.00: 44.9% (sub-parity, locked-in profit regardless of outcome)
• Pct sub-$0.97: 36.1% (at least 3% guaranteed return per paired unit)

44.9% of both-sides markets are locked-in winners. When the bot buys both outcomes for less than $1.00 combined, it cannot lose on that market regardless of resolution. The other 55.1% are directional bets where the dominant-side conviction must compensate for the overpayment on the hedge.

Classification:

The strategy is A (Both-Sides Spread Capture / Market Making) as the primary archetype with B (Directional Betting) as the secondary. The both-sides rate of 89.2% firmly places it in market-making territory. The dominance ratio data showing 87.6% correct at 3x+ confirms the directional signal is real and contributing to P/L.

---

Phase 3: Dominance Ratio Analysis

The pattern is unmistakable. As the bot allocates more asymmetrically, the dominant side wins more often. This is the signature of a real directional signal being expressed through sizing, not random allocation.

At 1.0-1.5x (near-equal), the 57.0% win rate is consistent with slightly better than random coin-flipping. At 3x+, the 87.6% win rate means the bot has effectively identified the winner before or during the market window.

The mean paired cost inversion is critical: the 3x+ bucket has mean paired cost 1.058 (above parity), meaning pure spread capture is negative in that bucket. The bot overpays $1.058 to guarantee $1.00. It only earns money in this bucket if the dominant side wins, and it does, 87.6% of the time. This confirms the allocation is driven by directional signal, not spread hunting.

The likely signal: the bot reads the spot price trajectory during the window. As BTC or ETH moves up or down within the 5-minute window, the bot identifies the winning outcome with increasing confidence and tilts its allocation toward it. By the time 3x or more capital is on one side, the price move is clear enough that the dominant side wins 87.6% of the time.

---

Phase 4: Entry Price Analysis

The price band structure reveals the market-making and directional logic simultaneously.

The $0.00-$0.70 bands are the hedge/non-dominant fills. The bot buys cheap outcomes (Down when Up is clearly winning, Up when Down is clearly winning) as partial hedges. These lose money as expected on a resolved basis, but they contribute to the paired-cost spread capture on both-sides markets.

The $0.80-$1.00 bands are where the real directional money is made. The $0.90-$1.00 band alone accounts for $27M of the $45M deployed (60% of capital in a 10% of the price range). This is the bot loading heavily into near-certain outcomes, particularly late in winning market windows. The best evidence for the "accumulate into momentum" strategy is this band: $27M at $0.90+ to collect $125K in trading P/L.

The win rate calibration is near-perfect across the spectrum: 4.9% at $0.01-$0.10, climbing to 96.6% at $0.90-$1.00. This wallet is not beating the market's implied probability; it is matching it almost exactly. The edge comes from the paired-cost mechanics and the asymmetric sizing, not from buying mispriced odds.

Sub-bucket inspection: The $0.90-$1.00 band is where the bulk of capital sits. Within this band, the CSV shows fills at $0.88-$0.97 in winning windows, with max single fills of $100-$200 (the $5,004 max is likely a deep-favorite late-window fill). This is not a single-tick concentrator; it is spread across all prices in the favorite zone.

---

Phase 5: Category and Market-Type Breakdown

100% of activity is Crypto. No sports, no politics, no other categories. The single-category structure is consistent with a specialized market-making operation that has built its infrastructure around one asset class.

Within crypto, by implied duration: The CSV shows fills on 5-minute, 15-minute, 4-hour, and daily markets. The 5-minute markets dominate by trade count (most volume), the 15-minute markets are secondary, and the longer-duration markets appear less frequently. This aligns with the liquidity-farming incentive: shorter-duration markets cycle faster and provide more reward opportunities per unit of capital deployed.

SOL presence is a key distinguishing feature from comparable wallets. The CSV explicitly shows sol-updown-5m-* alongside BTC and ETH. Bonereaper covers a broader asset universe than a pure BTC/ETH arbitrageur. SOL markets carry less competition and potentially higher reward eligibility per dollar deployed.

---

Phase 6: Timing and Execution

Hourly trading activity:

The hour histogram shows volume at all 24 hours with relatively even distribution. No hard sleep window exists. The busiest hours are 13:00-15:00 UTC (3,021/hour peak) and the quietest are 07:00-11:00 UTC (approximately 2,400/hour). This is a 24/7 operation with moderate diurnal variation, consistent with an automated system that follows market availability rather than operator schedules.

Hourly P/L:

The P/L volatility by hour is large relative to the overall total, which is expected for a market-making operation where individual hour results depend heavily on whether that hour contained strongly trending or rangebound crypto price action.

Day-of-week P/L:

Wednesday is the only negative day (-$8,685). Thursday, Saturday, and Sunday show the highest ROI. The weekend premium is consistent with reduced competition from other market makers, allowing the bot to capture more favorable spreads.

Burst patterns and accumulation windows:

The CSV shows same-second multi-fill bursts: on the June 12 BTC 5m market, the bot places 8-12 fills within a 2-3 second window as the market opens, then spaces out subsequent fills over 2 minutes. The initial burst covers both sides to establish the position, and the subsequent fills represent ongoing monitoring and adjustment. The median inter-fill gap of 2 seconds across all markets reflects these bursts averaging down against the longer inter-burst gaps.

Second-side lag: Median 21 seconds. The bot typically enters the dominant side first, then adds the non-dominant hedge within 21 seconds. This is intentional: wait for directional confirmation before hedging, rather than entering both sides simultaneously.

---

Phase 7: Filter Experiments

Most important finding: the high-conviction filter delivers +$157K above baseline. Restricting to markets where the dominant side carries 2x+ the capital of the non-dominant side, keeping only the dominant leg, generates +$245,909 in trading P/L on $33.7M deployed (+0.73% ROI). This is 2.78x the unfiltered P/L on 74.7% of the unfiltered capital.

The standard price band filter (30-70%) is severely destructive here. Applying it cuts the dominant-side high-price fills that generate most of the positive P/L, leaving only the hedge/non-dominant/moderate price fills that are net negative. The loss from this filter is -$116,771 vs baseline.

The hour exclusion filter slightly reduces P/L. The bot is 24/7, and cutting the worst 4 hours (12, 14, 17, 22 UTC) removes 480K trades but also removes some of the best directional opportunities embedded in those hours. Net -$23K, not material.

Summary for replicators: The actionable filter is the dominance ratio filter at 2x+. Dominant-side fills on asymmetrically sized markets are the profitable subset. The price band filter that works for directional bettors actively hurts this market-making book.

---

Phase 8: Rolling Window Consistency

The weekly stair from report_data shows 5 of 7 weeks positive on trading P/L. The account P/L (including rewards) is positive every single day with a monotonically increasing cumulative line from $4K on May 1 to $458K on June 14.

Critical distinction: The daily cumulative P/L line in the data (pnl_daily) shows account P/L (including rewards), which climbs every day without exception. The trading P/L alone has down weeks (Week 6: -$28K) and down days. The rewards smooth the account line and make it appear more stable than the underlying trading book.

Rolling 15-day analysis: The rolling15 data shows positive 15-day windows through late May, then turning negative in mid-June (June 9 window: -$733, June 14 window: -$27,048). This late-period deterioration in trading P/L reflects June's weaker crypto price action, where rangebound markets reduce directional alpha on the asymmetric fills. However, rewards income continued throughout, keeping the account total positive.

---

Phase 9: P/L Decomposition

The P/L decomposition from report_data provides spread_pnl = -$685,278 and hedge_tax = $7,654,329. This is the structural cost of market-making: the bot spends $7.65M on non-dominant/hedge positions that lose their full value at resolution, but these positions contribute to the paired-cost spread capture and to the liquidity-farming eligibility. The negative spread_pnl figure reflects that many both-sides markets are entered above parity (mean paired cost $1.013), meaning the spread itself is a drag.

Where the +$88K trading P/L actually comes from:

1. The 44.9% of both-sides markets where paired cost < $1.00: approximately 14,922 markets with guaranteed positive resolution regardless of outcome
2. The high-conviction asymmetric fills at $0.90+ (the $125K P/L from the $0.90-$1.00 band)
3. Partially offset by small losses on the hedge/non-dominant fills ($0-$0.70 bands: aggregate -$30K)

The dominant source of the +$88K trading P/L is the $125K from near-certain late-window dominant-side accumulation, partially offset by losses on the hedge legs and the overpriced spreads on the 55.1% of markets where paired cost exceeds $1.00.

---

Phase 10: Strategy Specification

One-sentence summary: A 24/7 automated crypto market-making bot that covers both sides of every short-duration BTC/ETH/SOL Up/Down market, skews allocation 3x+ toward whichever side the price action is confirming, and earns Polymarket's liquidity-mining rewards as the primary income stream alongside spread capture and directional alpha on asymmetric fills.

Edge sources:

1. Sub-parity paired costs on 44.9% of both-sides markets (guaranteed spread capture)
2. Directional signal that achieves 87.6% accuracy when allocation reaches 3x+ (asymmetric sizing alpha)
3. Liquidity rewards at $369,887 over 44 days for consistent two-sided market coverage (primary income, not replicable at small scale)

Critical parameters for the replication playbook:

• Both-sides entry required: cover all eligible short-duration crypto markets
• Dominant-side allocation: when spot price moves decisively, tilt allocation 3x+
• Price range: dominant fills across full range including $0.90-$1.00 late-window accumulation
• Duration: 5-minute and 15-minute primary, 4-hour and daily secondary
• Assets: BTC, ETH, SOL
• Schedule: 24/7, no sleep window
• No sell trades: hold all positions to resolution

Weaknesses: Rewards eligibility requires institutional-scale volume. The trading book alone at small scale generates +0.20% ROI on deployed capital before rewards. Rangebound crypto markets reduce directional alpha on the 3x+ bucket. June 1-7 week demonstrates the vulnerability: -$28K trading P/L when crypto moved without clear directional momentum.

Wallet: 0xeebde7a0e019a63e6b476eb425505b7b3e6eba30 Window: 2026-05-01 → 2026-06-14 (44 active / 45 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: 89.24% (33,223 of 37,228 markets)
• Median paired cost: $1.0163
• Mean paired cost: $1.0129
• Paired cost % under $1.00: 44.9%
• Paired cost % under $0.97: 36.1%
• Median 2nd-side hedge lag: 21s

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 45 (75.6%)
• Rolling 7-day P/L range: -$32,656 → +$41,385
• Rolling 15-day windows green: 38 of 45 (84.4%)
• Rolling 15-day P/L range: -$31,880 → +$70,442

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 2026-06-24 10:59 UTC.

Wallet: 0xeebde7a0e019a63e6b476eb425505b7b3e6eba30 Window: 2026-05-01 to 2026-06-14 Baseline (resolved-BUY trading view): 2,694,848 resolved BUYs, 47.8% win rate, $45,064,687 deployed, +$88,603 trading P/L, +0.20% ROI Account P/L (includes rewards): +$458,493 (rewards: +$369,887)

---

The headline result

One filter adds substantial lift. The standard "sweet spot" band filter is catastrophic. The category filter is irrelevant.

The high-conviction dominance filter at 2x+ is the single actionable refinement: it produces +$245,909 trading P/L on $33.7M deployed (+0.73% ROI) versus the unfiltered baseline of +$88,603 on $45M (+0.20% ROI). Applied in isolation, it would deliver 2.78x the trading P/L on 74.7% of the capital. Every other standard filter either hurts or is neutral.

---

Filter results table

---

Filter-by-filter commentary

1. Price band filter ($0.30-$0.70) → DESTRUCTIVE

Applying the standard mid-range sweet spot filter to this wallet causes catastrophic damage: trading P/L drops from +$88,603 to -$28,168, a swing of -$116,771.

The reason is structural. For this market-maker, the $0.30-$0.70 price zone is the "hedge zone": the bot buys non-dominant outcomes at mid-range prices as partial hedges. These are intentional loss positions that serve the paired-cost accounting and the liquidity-farming eligibility. Selecting only these trades strips out the dominant-side near-certainty fills at $0.80-$0.99 that generate +$128,952 of positive P/L, leaving only the cost centers.

The P/L by price band tells the story clearly: every band from $0.00-$0.79 generates negative or near-zero trading P/L. The entire positive P/L comes from the $0.80-$1.00 zone. The "$0.30-$0.70 sweet spot" rule is designed for directional bettors who buy before outcomes are clear. This wallet does the opposite: it accumulates heavily at high prices once the outcome is clear.

Recommendation: do not apply a price band filter to this wallet under any circumstances. It inverts the strategy's economics.

2. High-conviction dominance filter (dom 2x+, dominant side only) → MEANINGFUL_LIFT

This is the one filter that works. Restricting to markets where the dominant-side allocation is at least 2x the non-dominant allocation, and keeping only the dominant-side fills within those markets, yields:

• 937,837 fills (34.8% of total trades)
• $33.7M deployed (74.8% of total capital)
• +$245,909 P/L (+0.73% ROI)
• Win rate of 87.4%

The 87.4% win rate on this subset (versus 47.8% overall) reflects that the dominant-side bet in high-conviction markets is a near-certainty. The 0.73% ROI is 3.65x the unfiltered ROI, but on roughly three-quarters of the deployed capital, so the absolute P/L is 2.78x.

The mechanism: when the bot allocates 2x+ to one side, it has made an assessment that the spot price is confirming that direction. By the time the window resolves, that direction wins 87.4% of the time. The filter selects for the fills that embody the directional signal, removing the hedge fills that drag down the aggregate win rate.

For a replicator who wants to run this strategy at a fraction of the scale: only deploy capital on the dominant-side fills at 2x+ asymmetry, and skip the non-dominant hedge fills entirely. You lose the paired-cost spread capture on the sub-parity markets, but you gain a 3.65x improvement in ROI on the capital you deploy. At small scale where liquidity rewards are negligible anyway, this is the correct approach.

3. Category filter → NOT APPLICABLE

100% of the wallet's trades are Crypto. Applying the "top category" filter returns the exact baseline: 2,694,848 trades, same P/L, same everything. There is nothing to filter. The wallet is already perfectly specialized.

4. Hour exclusion filter (worst 4 hours: 12, 14, 17, 22 UTC) → MILDLY DESTRUCTIVE

Excluding the 4 worst-performing UTC hours removes 480,345 trades and $7.55M in capital, producing P/L of +$65,599 (+0.17% ROI) versus the baseline +$88,603 (+0.20% ROI). Delta: -$23,004.

This outcome might seem counterintuitive: shouldn't removing the worst hours improve the book? The issue is that these hours are not uniformly bad. Some fills within the "worst" hours are profitable high-conviction dominant-side fills. The negative performance in these hours comes primarily from the hedge fills, which are distributed evenly across all hours. Removing the hours removes the profitable fills along with the loss-making ones, resulting in a net negative.

Additionally, for a 24/7 market-making operation that relies on continuous presence for liquidity-farming rewards, excluding hours would reduce reward eligibility, compounding the damage at the account level even if the filter showed marginal trading improvement (which it doesn't).

Recommendation: do not implement hour-based scheduling cutoffs for this strategy. The bot should run 24/7, consistent with its current operation.

5. Combined filter (dom 2x+ + exclude worst 4 hours) → DESTRUCTIVE

Stacking the dominance filter with the hour exclusion produces -$27,313 (-0.35% ROI), which is worse than either the unfiltered baseline or either component filter alone. The combination reduces the sample to only the dominant-side fills in the good-hours subset (1,035,240 trades, $7.9M capital), which happens to exclude many of the high-conviction dominant fills that occurred during the "bad" hours.

This is the standard pitfall of stacking filters on market-making books: each filter independently makes sense conceptually, but combined they over-select on specific market conditions and lose the diversity that makes the strategy work.

---

The genuine actionable refinements

The standard PR&R filter battery is misaligned with this wallet's structure. The filters designed for directional bettors (price band, hour selection) are either neutral or destructive here. The genuine refinements available are:

The first is actionable now. The rest require data beyond the trade CSV.

---

Bottom line for replication

Three concrete recommendations:

1. DO apply the dominance filter at 2x+, dominant side only. It is the only filter that improves trading P/L, and it does so dramatically (+$157K). At small scale, this is the correct subset to trade.

1. DO NOT apply the $0.30-$0.70 price band filter. It destroys -$116K of trading P/L by stripping out the near-certainty late-window fills that are the primary profit source.

1. DO run 24/7 without hour-based cutoffs. The bot's value comes from continuous market presence (for rewards) and from covering all windows (for directional alpha). Hour-based filtering reduces both.

The fundamental lesson from these filter experiments: the standard filter battery assumes a directional bettor who needs to screen for high-quality signals. Bonereaper is a market-maker that already screens by allocating asymmetrically. The dominance ratio is the built-in filter. Apply it explicitly, and skip everything else.

Source wallet: 0xeebde7a0e019a63e6b476eb425505b7b3e6eba30 Strategy: Crypto short-duration both-sides market-making with asymmetric directional allocation and liquidity-farming Reference performance: +$458,493 account P/L over 44 days (trading: +$88,606 at +0.20% ROI; rewards: +$369,887)

---

One-paragraph operator brief

Build a Polymarket bot that enters every active BTC, ETH, and SOL short-duration Up/Down market (5-minute, 15-minute, optionally 4-hour) by buying both outcomes. During the window, monitor the spot price tape; when one side is clearly winning (spot price confirming directional movement), tilt the allocation 2x+ toward the dominant side. Accumulate heavily into near-certainty late-window positions at $0.85-$0.97. Hold all positions to resolution with no exits. Run 24/7 without sleep windows. Capital target: $50K-$500K working capital. The trading P/L alone (+0.20% ROI on deployed capital) is modest; the primary income at scale is Polymarket's liquidity-mining rewards, which require consistent two-sided market presence and high volume to earn meaningfully. At small scale, focus exclusively on the high-conviction dominant-side fills (dom 2x+) to achieve a 87.4% win rate and ~0.73% ROI on deployed capital.

---

1. Market Selection

Priority order by expected alpha per dollar deployed:

1. BTC 5m (highest volume, most reward-eligible)
2. ETH 5m (good volume, less competition than BTC)
3. SOL 5m (least competition, potentially highest reward-per-dollar)
4. BTC 15m (half the refresh rate of 5m, good for both strategies)
5. ETH 15m (secondary, lower volume)
6. BTC 4h and daily (occasional, low frequency)

Coverage goal: At full implementation, enter every eligible BTC 5m and ETH 5m window as soon as it opens. At minimum viable scale, enter every window where a directional signal is present.

---

2. Entry Logic

The strategy has two operating modes. Both should run simultaneously.

Mode A: Symmetric market-making (for rewards and spread capture)

def enter_market_symmetric(market):
    # Enter both sides early in the window
    if seconds_since_open(market) < 30:
        return None  # Too early, orderbook still calibrating
    
    if seconds_until_close(market) < 15:
        return None  # Too late
    
    # Calculate target allocation
    up_price   = market.up_side.ask_price
    down_price = market.down_side.ask_price
    paired_cost = up_price + down_price
    
    # Only enter if paired cost is reasonable
    if paired_cost > 1.10:
        return None  # Too expensive for symmetric play
    
    base_clip = min(10.0, available_capital() * 0.002)
    
    return {
        "Up":   base_clip * 0.5,  # Equal split as starting point
        "Down": base_clip * 0.5
    }

Mode B: Asymmetric allocation (for directional alpha)

def enter_market_asymmetric(market, spot_data):
    # Determine directional signal from spot tape
    direction = assess_spot_direction(spot_data, market.asset)
    
    if direction == "neutral":
        return None  # No directional signal, skip asymmetric mode
    
    # Compute dominance ratio based on conviction level
    # Higher conviction = more asymmetric
    conviction = get_conviction_score(spot_data, market)
    
    if conviction < 0.6:    # Low conviction
        dom_ratio = 1.5
    elif conviction < 0.75: # Moderate conviction
        dom_ratio = 2.5
    else:                    # High conviction
        dom_ratio = 4.0
    
    base_clip = min(50.0, available_capital() * 0.01)
    dominant_clip    = base_clip * dom_ratio / (dom_ratio + 1)
    nondominant_clip = base_clip * 1 / (dom_ratio + 1)
    
    # Only the dominant side for high-conviction (>= 2x filter)
    if dom_ratio >= 2.0:
        return {
            direction:      dominant_clip,
            other_side:     nondominant_clip  # Optional hedge
        }
    
    return None  # Below threshold, skip

---

3. Exit Logic

There are no exits. The wallet holds 100% of positions to resolution, confirmed by the zero SELL trades in the data.

This is a design choice, not an oversight:

• The strategy captures spread at entry (sub-$1.00 paired costs on 44.9% of markets)
• The directional alpha comes from near-certainty late-window accumulation, which only pays at resolution
• Selling early would cap the upside on the near-certain dominant leg while leaving the non-dominant hedge in place

The only exception is operationally forced exits if a market never resolves (operator timeout, technical issue). Build a cleanup module that handles unresolved positions after 48 hours.

def exit_logic(position):
    # Hold to resolution - no active exits
    return None

def cleanup_stale_positions():
    for position in open_positions:
        if position.market.window_close + timedelta(hours=48) < now():
            # Market likely stuck; manually close or flag for review
            flag_for_manual_review(position)

---

4. Sizing Model

The sizing model has three tiers corresponding to the three types of fills in the reference wallet:

Recommended sizing by bankroll:

The reference wallet achieved $45M in monthly turnover on approximately $5-15K of instantaneous working capital by cycling fills across thousands of markets. High-frequency cycling is the key: deploy a small base, collect at resolution (5-15 minutes), redeploy immediately.

Critical sizing rule for small-scale replicators: If you cannot afford to cover both sides of a market with the hedge leg, skip the symmetric mode entirely and run only the asymmetric (high-conviction) mode. One-sided dominant-leg fills at 2x+ asymmetry generate +0.73% ROI and 87.4% win rate. This is far better than the 0.20% from the full market-making book, and it requires roughly 60% less capital per market.

---

5. Both-Sides Allocation

For the full market-making strategy, the both-sides allocation follows this framework:

Baseline (symmetric):

• Entry: 50/50 allocation between Up and Down when no directional signal
• Purpose: establish market presence, qualify for liquidity rewards, lock in sub-$1.00 paired costs when available

Directional tilt (asymmetric):

• When conviction is moderate (0.6-0.75): 2.5x dominant side vs. non-dominant
• When conviction is high (>0.75): 4x+ dominant side vs. non-dominant
• Late-window accumulation (final 60 seconds of a winning window): load heavily at $0.85-$0.97

The correlation between dominance ratio and accuracy from the data:

Do not reach for 3x+ unless the directional signal is very strong. At 3x+, the mean paired cost exceeds $1.00, so you are overpaying for the hedge leg. You need the 87.6% win rate to compensate. Only go there when you have high conviction.

---

6. Bankroll Math

The trading P/L component:

Reference wallet performance:
  Capital deployed/month:    ~$45M (buy notional)
  Net trading P/L/month:     +$88,606
  Trading ROI:               +0.20% on deployed capital

At the high-conviction filter subset (dom 2x+, dominant leg only):
  Capital deployed/month:    ~$33.7M
  Net trading P/L/month:     +$245,909
  Trading ROI:               +0.73% on deployed capital

Working capital required:
  The bot recycles capital every 5-15 minutes (market window duration)
  To deploy $1M/day in notional requires ~$10K-20K instantaneous capital
  (Cycling rate: 1 cycle per 5 minutes = 12 cycles/hour = 288 cycles/day)
  1M / 288 cycles ≈ $3,500 average position size per cycle

Rewards estimation at different scales:

*Rewards estimates are rough approximations; Polymarket's reward structure is nonlinear and not publicly specified. Small-scale operators receive disproportionately less than linear scaling suggests.

The brutal math at small scale: Below $1M/month in trading volume, rewards income is negligible. The trading P/L alone at 0.20% ROI on $100K deployed = $200/month. This is only worth running if you are building toward scale or if you apply the high-conviction filter to achieve 0.73% ROI ($730/month on $100K deployed). Still modest, but the economics improve dramatically as volume grows.

---

7. Hour Scheduling

Run 24/7. There is no sleep window in this strategy.

The reference wallet trades all 24 hours with minimal diurnal variation (busiest: 13-15 UTC with ~3,000 trades/hour; quietest: 07-11 UTC with ~2,400 trades/hour). The variation is 25%, not the 100% gap seen in true US-session operators.

Day-of-week scheduling:

Wednesday is the one anomalous day. The -0.16% ROI may reflect particular crypto market microstructure on Wednesdays (equity-market mid-week effects, derivatives expiry patterns). Consider reducing position sizes by 50% on Wednesdays if you want to reduce variance, but do not shut down entirely because rewards eligibility requires continuous presence.

---

8. Operational Requirements

Infrastructure minimum spec:

• VPS colocated in AWS US-East-1 or equivalent low-latency endpoint near Polymarket infrastructure
• 1Gbps network connection
• Python/Rust bot with async I/O (asyncio or tokio)
• PostgreSQL or equivalent for trade logging

---

9. Risk Profile

Position limits:

MAX_USDC_PER_MARKET  = bankroll * 0.01   # No single market > 1% of bankroll
MAX_USDC_OPEN_TOTAL  = bankroll * 0.50   # Total open exposure < 50% of bankroll
MAX_SINGLE_FILL      = bankroll * 0.002  # No single fill > 0.2% of bankroll

---

10. Diagnostic checklist

Run weekly to verify the bot is functioning correctly:

---

What this playbook deliberately does NOT include

• Price band filters ($0.30-$0.70). These destroy -$116K of trading P/L by removing the near-certainty dominant fills that generate all the positive P/L. Never apply them.

• Hour-based sleep windows. This strategy requires 24/7 presence for rewards eligibility. Shutting down for even 6 hours/day would reduce rewards income by 25%. There is no evidence in the data that any hours should be excluded.

• Active exit management (SELL trades). The wallet holds 100% to resolution. Adding SELL exits would cap the upside on late-window near-certainty positions and complicate the paired-cost mechanics. Do not add sells.

• Non-crypto markets. The bot's entire infrastructure, signal design, and reward eligibility is calibrated for crypto short-duration markets. Expanding to sports or politics would require a completely different signal model and would dilute the liquidity-farming rewards concentration.

• Kelly sizing or full-bankroll deployment. At 0.20% ROI, Kelly sizing is essentially zero. The correct approach is fixed fraction with hard caps per market and per fill.

• Copy-following other wallets. Short-duration crypto markets are self-contained; by the time another wallet's fill is visible, the market is already resolving. Copy-trading adds zero alpha here.

The strategy's core insight is that being consistently, reliably present across every crypto short-duration market is worth more than any clever directional call. Build for uptime, not for prediction.