LBTrading

Wallet: 0xa6db8383b756f1e58c504ff4179e19427dd73155 Window: 2026-07-01 to 2026-08-16 (47 calendar days, 47 active) Universe: 45,771 trades (all BUYs, zero SELLs) · 10,208 markets · 4,989 events · $641,421 gross turnover

The Punchline

LBTrading is a systematic sports market maker running a both-sides spread-capture strategy at industrial scale, augmented by a directional tilt that becomes meaningful at high conviction levels. The wallet deployed $641,421 across 47 days, buying both sides of 6,217 markets (60.9% of its universe) and directionally buying one side only on the remaining 3,991. Zero sells: every position is held to resolution.

On trading P/L alone, the strategy lost -$443.19 across 44,815 resolved BUYs (-0.07% ROI). That is not the story. The maker rebate income of $1,882.39 from 40 rebate events flipped the account to +$1,951.56 net. The business model is: generate enormous maker-side flow, collect the rebate, use the spread-capture discipline to keep the trading P/L near zero, and bank the difference. The directional layer (high-dominance markets) adds positive skew on top when the operator's event-outcome signal fires.

This is the anatomy of a professional Polymarket market-making operation, not a gambler. The numbers at every level confirm it: calibrated paired costs, layered dominance ratios with rising win rates, consistent daily activity across 200+ markets, and maker-rebate income that constitutes the majority of net profitability.

What He Trades

The universe is broad: every major sport, significant esports coverage, and miscellaneous live-event props.

The "Other" category dominated by volume ($371K) encompasses the esports-heavy portion of the book: Counter-Strike map winners, LoL game handicaps, Dota 2 series results, Valorant match winners, plus CFL, World Cup props, and Korean soccer. The CSV sample makes this explicit: a July 11 session shows simultaneous trading across HLE vs LYON game handicaps (multiple price points), CS2 FaZe vs PARIVISION map handicaps, MLB Kansas City vs Baltimore totals, Norway vs England World Cup corners and exact scores, UFC rounds totals, and CFL. LBTrading is one of the only wallets on Polymarket covering all of these simultaneously.

The book has no meaningful geographic or category anchor. It follows event schedules globally.

The Order of Operations: One Event, Trade by Trade

The Argentina vs. Switzerland World Cup match (July 11-12) illustrates the strategy end-to-end. The event spawned dozens of sub-markets: match winner, exact scores, first-half totals, corners, team totals, second-half results, player props (Messi goals, Fernandez assists), penalty shootout, extra time, and spread handicaps. LBTrading entered all of them.

The pattern repeats across all sub-markets. Every market gets both Yes and No entries at whatever prices the book offers. For each paired position, the combined cost is calculated to be near or below $1.00 (capturing the spread). For directional markets, he tilts toward the more likely outcome by allocating more capital. The July 12 Argentina session shows 30+ distinct BUY fills across 15 sub-markets within a 15-minute window, all while simultaneously placing fills on Athletics vs White Sox MLB totals.

The operator is not watching individual games. The bot is scanning for sub-$1.00 paired-cost opportunities across all active markets simultaneously, placing fills wherever the spread exists, and applying a sizing tilt based on a pre-game (or live-updated) probability estimate.

Why It Works: The Math

Three stacked mechanisms:

Mechanism 1: Spread capture on paired positions

Median paired cost:    $0.9967 per pair
Expected payout:       $1.00 (one side always wins)
Spread per pair:       $0.0033 (33 basis points)
% of book paired:      60.9% of 10,208 markets = 6,217 markets
Theoretical spread P/L = 6,217 × avg_paired_capital × 0.0033

With average paired deployment of roughly $50-100 per market, this generates $1,000-$2,000 of gross spread income over the window before directional noise. Actual realized spread P/L from the decomposition is +$802.48.

Mechanism 2: Maker rebate income

Measured maker rebates:   $1,882.39 over 40 rebate events
Average rebate per event: $47.06
Total trades:             45,771
Implied rebate rate:      ~0.29% of notional

At 40 events across 47 days, rebate events are not daily, which implies they are paid in batches. The trading volume of $641K at a 0.29% maker rebate rate would produce $1,860 in rebates, consistent with the measured $1,882.39. This confirms the mechanism is standard maker rebate (not LP mining, as lp_rewards are only $3.60).

Mechanism 3: High-dominance directional accuracy

3.0x+ dominance bucket:
  Markets:              2,053
  Dominant-side wins:   1,558 of 2,001 resolved
  Win rate:             77.9%
  Mean paired cost:     $0.9966

2.0-3.0x dominance:
  Markets:              971
  Dominant-side wins:   602 of 953 resolved
  Win rate:             63.2%

1.5-2.0x dominance:
  Markets:              1,039
  Dominant-side wins:   598 of 1,022 resolved
  Win rate:             58.5%

1.0-1.5x dominance:
  Markets:              2,154
  Dominant-side wins:   1,118 of 2,110 resolved
  Win rate:             53.0%

The monotonic rise from 53% to 78% as dominance increases is the clearest signal in the dataset that there is real information embedded in the tilt. At 1x-1.5x (near-equal allocation), the dominant side barely beats coin-flip. At 3x+, the dominant side wins 77.9% of the time - far above the 63%+ that would be expected if the dominant leg were simply the market-implied favorite. LBTrading is correct on its high-conviction tilts at a rate that implies genuine predictive signal.

Phase 1: Trader Profile

Scale and activity:

Trade size distribution:

The size distribution is compact. The max is only 11.3x the median. The top-5% concentration at 24.6% is moderate. This is a clip-capped strategy: fills almost never exceed $98, and the typical fill is $8.70. The hard $98 cap appears to be a system parameter. The size distribution is not power-law; it is roughly uniform across a range of $5-$90 with concentration in the $5-$50 range.

Execution signature:

The median inter-trade gap of 417 seconds (7 minutes) between consecutive fills looks manual but is misleading. The p10 gap is 1 second, and 23.5% of consecutive fills are under 10 seconds. The distribution is bimodal: rapid burst sequences (bot-mode entry across multiple markets at once) and long idle periods (waiting for new events to open). The second-side lag median of 4,206 seconds (70 minutes) between first and second side of a paired market confirms that both sides are not placed simultaneously but within a multi-hour window.

Trading hours:

Trades occur at all 24 hours with a trough at 02:00-06:00 UTC (roughly 700-1,400 trades per hour vs peak of 3,276 at 17:00 UTC). Unlike SirMartingale, there is no hard sleep window. The bot runs continuously but at lower intensity during early morning UTC.

Phase 2: Core Strategy Identification

Both-sides participation: 60.9% of 10,208 markets. This single number classifies the strategy as A (Market Making / Spread Capture) with B (Directional Betting) as the secondary component.

The operator is NOT:

• A pure directional bettor (60.9% both-sides rate rules that out)
• A latency arbitrageur (no sell-side activity, no evidence of rapid entry-exit cycles)
• A copy-trader (too broad and too simultaneous across too many event types)
• A DCA accumulator (fills cluster in event windows, not spread across days)

The operator IS:

• A systematic market maker buying both sides to capture the paired spread
• A conviction-weighted allocator applying a probability model to determine tilt direction and magnitude
• A maker-rebate harvester operating at the volume level needed for meaningful rebate income

The zero-sell constraint is worth underscoring. Across 45,771 BUYs, there is not one single SELL. Every position is held to binary resolution. This is structurally different from an active exit manager like SirMartingale. LBTrading's edge does not come from exit timing; it comes from entry pricing and volume.

Phase 3: Dominance Ratio Analysis

The dominance ratio is the most important diagnostic for this wallet. It confirms the directional signal embedded inside the spread book.

The step from 63.2% to 77.9% between the 2x-3x and 3x+ buckets is decisive. At 2,001 resolved markets in the 3x+ bucket, the sample is large enough to be statistically robust. A 77.9% dominant-side win rate with a mean market implied probability of roughly 65-70% (since they are paying $0.9966 for a paired position leaning heavily one side) represents meaningful positive selection.

The paired costs are remarkably consistent across all dominance buckets ($0.997-$0.998), confirming the spread-capture discipline is maintained regardless of conviction level. The operator is not sacrificing spread income to express a directional view; it is adding the directional tilt on top of the spread income.

Phase 4: Entry Price Analysis

The win rate column is a near-perfect calibration: 5.1% wins on sub-$0.10 purchases, 92.5% wins on $0.90+ purchases. The market is pricing these outcomes correctly on average. This confirms LBTrading is not exploiting pricing errors in a particular price band.

The P/L column shows the losers: sub-$0.10 entries bleed -$818 on only $5,976 deployed (-13.7% ROI). These are the longshot legs of paired positions where LBTrading buys the "No" on a near-certain outcome for a few cents to complete the spread. The $0.30-$0.40 and $0.60-$0.70 bands also show negative P/L, reflecting the directional noise that spreads don't fully hedge.

Sub-bucket analysis: The $0.40-$0.60 zone holds the majority of capital ($274K, 43% of total). Within this zone, fills are spread across individual cents rather than pinned to a single price. This is consistent with an opportunistic market-maker buying whatever the book offers rather than anchoring to a specific fair-value price.

The price-band distribution confirms that this is not a single-tick bot like LIL222. Capital is distributed across all price points, with concentration in the coin-flip zone ($0.40-$0.60) where both-sides paired costs are easiest to minimize.

Phase 5: Category and Market-Type Breakdown

The NFL category is the standout on ROI (7.63%), but at only 163 trades and $3.8K deployed the absolute impact is small (+$289). The sample is the preseason coverage at the start of the window (July-August).

The core problem is the esports category. The 26,161 "Other" trades carrying $371K of capital bleed -$1,869 at -0.51% ROI. This is the paired-cost discipline not fully working in esports: liquidity is thinner, spreads are occasionally inverted, and the operator's probability model is less calibrated on Dota 2 and Valorant markets than on tennis or MLB. The CSV confirms heavy esports coverage (Counter-Strike, LoL, Dota 2, Valorant all appear in the top markets by volume).

Tennis at -$288 on $95.9K is nearly breakeven at -0.31%. The market-making mechanic is working but the directional tilt is not adding value in tennis.

MLB (+$692) and NBA (+$579) are the profitable sport categories in absolute terms, both with win rates just above 50%.

Phase 6: Timing and Execution Analysis

Hourly P/L and win rate:

Worst hours: 00:00 UTC (-$527), 22:00 UTC (-$556), 08:00 UTC (-$521), 05:00 UTC (-$417).

The hourly P/L is highly noisy. No single hour is consistently profitable by a meaningful margin. The best-performing hour by absolute P/L is 14:00 UTC (+$697 on 2,444 trades, +28 cents per trade), which corresponds to 10:00 ET (US sports morning). The 01:00 UTC hour (+$626 on 1,398 trades) benefits from overnight Asian sports.

Day-of-week analysis:

Monday (+$555, +0.93% ROI) and Friday (+$490, +0.58%) lead. Saturday (-$964) and Sunday (-$817) are the bleeding days. Weekend underperformance likely reflects esports tournament scheduling (heavy weekend esports volume, where the model is weakest) and reduced market liquidity that widens book crossing costs.

Burst patterns:

The second-side lag median of 4,206 seconds (70 minutes) is the critical timing insight. LBTrading does not enter both sides simultaneously. It typically enters the larger (dominant) side first, then returns 30-120 minutes later to fill the other side. This pattern suggests the operator uses a queue: monitor markets for opening spread conditions, fire dominant side, then opportunistically complete the pair when the other side reprices.

Phase 7: Filter Experiments

Full filter analysis in the Filters tab. Summary:

The hour-exclusion filter produces the only genuine lift: +$1,503 in trading P/L improvement by skipping hours 00:00, 03:00, 19:00, and 22:00. This is meaningful but does not change the fundamental picture.

Phase 8: Rolling Window Consistency

The rolling window analysis shows significant instability. Only 49% of 7-day windows are green, and the worst 7-day window reaches -$1,301. This is not a strategy with a smooth, consistent edge on trading P/L alone. The weekly data shows alternating profitable and unprofitable weeks:

The account-level cumulative P/L (which includes rebates) climbs steadily from $9.61 on July 1 to $1,951.56 on August 16, consistent with rebate income accumulating in batches throughout the window.

Phase 9: P/L Decomposition

The hedge tax figure of -$239,709 is technically the sum of all losing-side USDC spent, not a net number. The spread P/L of +$802 represents the guaranteed component from paired positions that closed with sub-$1.00 combined cost. The net of these two (-$443) confirms the book is slightly trading-side negative even after capturing all spread income.

Why the spread income doesn't fully offset the hedge tax: The median paired cost of $0.9967 implies only 33 basis points of guaranteed income per pair. With average paired notional per market of roughly $80-100, that's only $0.26-$0.33 per paired market. Across 6,217 paired markets this produces $1,600-$2,100 gross, but the directional noise (when the dominant side loses) overwhelms it. The solution in the actual business model is the rebate income that doesn't depend on outcome accuracy at all.

Phase 10: Strategy Specification Summary

One-sentence summary: A systematic 24/7 sports market maker that buys both sides of 61% of its markets to capture sub-$1.00 paired costs, applies a probability-weighted directional tilt (3x+ dominance markets show 77.9% accuracy) across 200+ markets daily, and profits primarily from maker rebates on $641K+ of monthly trading volume.

What works: Maker rebate income at scale, spread discipline (median $0.9967 paired cost), high-dominance directional accuracy (77.9% at 3x+), MLB and NBA categories, weekday operation.

What drags: Esports category (-$1,869 on $371K deployed), weekend operation (-$964 Sat, -$817 Sun), hours 00:00 and 22:00 UTC.

What replicators must build: A volume-generating bot capable of 40,000+ trades per 47 days to access meaningful rebate income. Without that scale, the trading P/L alone is negative. See full playbook for the runnable spec.

Wallet: 0xa6db8383b756f1e58c504ff4179e19427dd73155 Window: 2026-07-01 → 2026-08-16 (47 active / 47 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: 60.90% (6,217 of 10,208 markets)
• Median paired cost: $0.9967
• Mean paired cost: $0.9972
• Paired cost % under $1.00: 62.1%
• Paired cost % under $0.97: 2.1%
• Median 2nd-side hedge lag: 4206s

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: 30 of 47 (63.8%)
• Rolling 7-day P/L range: -$1,301 → +$1,081
• Rolling 15-day windows green: 28 of 47 (59.6%)
• Rolling 15-day P/L range: -$684 → +$1,003

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 2026-08-16 23:17 UTC.

Wallet: 0xa6db8383b756f1e58c504ff4179e19427dd73155 Window: 2026-07-01 to 2026-08-16 Baseline: 44,815 resolved BUYs · 47.2% WR · $623,728 deployed · -$443 trading P/L · -0.07% ROI Account P/L (with maker rebates): +$1,951.56 verified

The headline result

No filter produces a reliably profitable trading book in isolation. The base strategy loses -$443 on trading alone across 47 days, and the standard filter battery does not reverse that. One filter (hour exclusion) provides genuine lift of +$1,503. One filter (top category: NFL) shows elite ROI but on a tiny sample. The remaining filters are destructive, neutral, or structurally inapplicable.

The deeper insight: this strategy's profitability does not come from filter-selectable trade quality. It comes from maker rebate income that requires volume. Any filter that cuts volume also cuts rebates. The optimal operating mode is maximum eligible volume (excluding demonstrated bad hours/categories) rather than hunting for a high-quality subset.

Filter results table

Filter-by-filter commentary

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

Applying the standard sweet-spot filter keeps 29,508 trades on $422,265 of capital. Win rate improves from 47.2% to 50.3%, which sounds encouraging. But the P/L worsens from -$443 to -$470, a $27 deterioration. ROI moves from -0.07% to -0.11%.

The mechanism: this filter discards the high-conviction both-sided pairs where LBTrading buys the near-certain leg at $0.80-$0.90 and the longshot leg at $0.10-$0.15. By removing all sub-$0.30 and post-$0.70 entries, the filter destroys the paired-cost structures that generated +$802 in spread P/L. What remains is the coin-flip zone without the paired income. Additionally, cutting from 44,815 to 29,508 trades (34% reduction) would cut maker rebate income proportionally from $1,882 to roughly $1,240, costing another $642 in account-level P/L.

Conclusion: do not apply this filter. It destroys both trading spread income and rebate income simultaneously.

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

This filter keeps only the 7,921 dominant-side BUYs from markets where LBTrading allocated 2x+ to one side. Win rate jumps to 73.3%, which is genuinely impressive. But trading P/L is -$787 on $195,383 deployed (-0.40% ROI) - worse than baseline.

Why? The filter discards the non-dominant sides that complete the paired structures. The dominant side alone pays out when correct but is missing the guaranteed spread income from the paired hedge. Furthermore, 73.3% win rate on a 0.40% loss ROI implies the average winning position pays too little relative to its entry price. Buying a market at $0.70 on average and winning 73.3% of the time produces roughly expected value of $0.733 per dollar at $1.00 resolution vs $0.70 cost, which should be +4.7% ROI. The actual -0.40% suggests some systematic over-pricing of the dominant side (buying slightly above fair value on average), which gets corrected out to near-zero by the market's calibration.

The win rate is real and impressive. The P/L is negative. The filter is destructive in practice.

3. Top category filter (NFL only) MEANINGFUL_LIFT (small scale)

NFL produces +$289 on $3,786 deployed (7.63% ROI) across 163 trades. This is genuine alpha: 55.2% win rate against fair-market expected win rates of roughly 49-51% for coin-flip zone trades implies positive selection. The NFL trades in this window are preseason (July-August), which historically shows higher pricing inefficiency.

But 163 trades is too small a sample to operate as a standalone strategy. The $3,786 of deployed capital generates perhaps $11 in maker rebates (0.29% rate). The total account contribution from NFL is roughly $300 of trading P/L plus $11 in rebates = $311.

Applied as a category *enhancement* (over-weight NFL within the broader book), this is actionable. Applied as a standalone strategy replacing the full book, it produces insufficient volume for meaningful rebate income.

4. Hour exclusion filter (worst 4 hours: 0, 3, 19, 22 UTC) MEANINGFUL_LIFT

This is the only filter that provides genuine, material trading P/L improvement. Excluding hours 00:00, 03:00, 19:00, and 22:00 UTC reduces trades from 44,815 to 38,576 (14% reduction) while improving trading P/L from -$443 to +$1,060, a swing of +$1,503.

The four excluded hours collectively generated -$1,408 in P/L on trades within those hours. Removing them converts the trading book from negative to positive (+$1,060). Win rate improves modestly from 47.2% to 47.6%.

The hour profiles of the worst performers:

• 00:00 UTC: -$527 on 1,603 trades (largely end-of-day sports events with resolved uncertainty)
• 22:00 UTC: -$556 on 2,437 trades (overlaps with European evening and US late sports)
• 19:00 UTC: -$174 on 1,578 trades
• 03:00 UTC: -$150 on 797 trades (thin overnight market with high adverse selection)

The trade-off: removing 6,239 trades also removes approximately $181 in maker rebate income (6,239/45,771 × $1,882). The net account-level benefit of the hour filter is approximately +$1,503 - $181 = +$1,322 per 47-day period.

5. Combined filter (NFL + exclude worst hours) MEANINGFUL_LIFT (limited scale)

The combined filter keeps 123 NFL trades outside the worst 4 hours: 53.7% win rate, $2,692 deployed, +$293 trading P/L, 10.9% ROI. This is the highest ROI of any filter in the battery.

The problem is the same as the standalone NFL filter: 123 trades per 47 days generates about $8 in maker rebates. The combined filter cannot function as a standalone strategy. As an anchor for a broader multi-category book, it confirms which specific execution conditions produce the best outcomes.

What filters would actually help (data not available)

The standard PR&R filter battery targets dimensions like price band, category, hour, and dominance. None of these fully unlock the available edge because the real alpha on this wallet is in rebate mechanics and paired-cost precision, both of which require external data:

The esports exclusion and weekend exclusion are the two highest-impact computable filters from the available data. Applying both would swing trading P/L from -$443 to approximately +$1,338, potentially making the book profitable on trading alone before rebates.

Bottom line for replication

The standard filter battery does not work as intended on LBTrading because the strategy's profitability mechanism (rebate income) is orthogonal to trade quality. Three recommendations:

1. Apply the hour filter. Exclude 00:00, 03:00, 19:00, and 22:00 UTC. This is the only filter in the standard battery that improves both trading P/L and is unlikely to destroy a proportional amount of rebate income (the excluded hours are low-quality on both dimensions).

1. Over-weight NFL and MLB, under-weight esports. The esports category is the main drag. Reducing esports exposure by 50% while maintaining comparable volume in MLB and NBA would substantially improve the trading P/L without reducing total volume.

1. Do not apply the dominance or price-band filters. Both destroy the paired-cost structures that generate spread income, and both reduce the volume needed for maker rebates. The dominant-side win rate of 77.9% is impressive but cannot be isolated from the non-dominant paired side without losing the spread income.

Source wallet: 0xa6db8383b756f1e58c504ff4179e19427dd73155 Strategy: Systematic sports market making with spread-capture pairs, conviction-weighted directional tilt, and maker-rebate harvesting Reference book: $641,422 deployed over 47 days · -$443 trading P/L · +$1,882 maker rebates · +$1,952 net account P/L

One-paragraph operator brief

Build a Polymarket market-making bot that operates 24/7 across sports (MLB, NBA, NFL, Soccer, UFC) and selectively across esports (Counter-Strike, LoL), buying both sides of each market at paired costs below $1.00 to lock in the guaranteed spread, while applying a probability-weighted tilt to allocate more capital to the more likely side. Run at sufficient volume (40,000+ fills per 47-day period) to generate meaningful maker rebate income. Avoid the four worst UTC hours (00, 03, 19, 22) and reduce esports exposure to limit the category drag. At full scale, expect trading P/L near breakeven or slightly positive, with the account bottom line driven by maker rebates of approximately $1,800-$2,000 per 47-day period on $600K+ of turnover.

1. Market Selection

Category prioritization rationale: MLB generates +$692 on $77.9K (0.91% ROI), NBA +$579 on $35.2K (1.72%), NFL +$289 on $3.8K (7.63%). Esports bleeds -$1,869 on $371.5K (-0.51%). The replicator's first improvement over the reference book is reducing esports to at most 15% of volume rather than the 58% the reference book allocated.

Market depth guidance: The reference book deploys in markets across every granularity level: match winner, exact score, first-half totals, corners, spreads, player props. Across a single big event (like Argentina vs Switzerland), LBTrading entered 30+ distinct sub-markets. Replicating this breadth requires a market scanner that indexes all available sub-markets for each tracked event.

2. Entry Logic

The entry decision combines two checks: the paired-cost check (is there a guaranteed spread to capture?) and the directional check (which side gets the larger allocation?).

def evaluate_entry(market, clob_snapshot):
    yes_ask = clob_snapshot.yes_best_ask
    no_ask  = clob_snapshot.no_best_ask
    
    paired_cost = yes_ask + no_ask
    
    # PRIMARY GATE: Only enter if paired cost is below $1.00
    # Targeting sub-$0.98 for meaningful spread; anything below $1.00 is acceptable
    if paired_cost >= 1.005:
        return None  # no guaranteed spread, skip
    
    # DIRECTIONAL TILT: Use probability model to determine dominant side
    fair_prob_yes = probability_model(market)   # external model output
    
    if fair_prob_yes > yes_ask + 0.05:
        dominant_side = "Yes"
        underdog_side = "No"
    elif (1 - fair_prob_yes) > no_ask + 0.05:
        dominant_side = "No"
        underdog_side = "Yes"
    else:
        dominant_side = None   # near-coin-flip, use equal allocation
    
    return EntryPlan(
        dominant_side=dominant_side,
        paired_cost=paired_cost,
        yes_ask=yes_ask,
        no_ask=no_ask
    )

Important: do not anchor to a specific price band. The reference book uses all price points from $0.01 to $0.99. The paired-cost check across both sides is the entry gate, not the individual side price.

3. Sizing Model

The reference book uses a hard cap at $98 with a median of $8.70. The sizing appears to vary by market quality (paired cost depth) and conviction level (dominance ratio).

Base sizing framework:

Working capital requirements:

The rebate rate of approximately 0.29% of notional means a $300K-scale operator earns roughly $700-$900 per month in rebates. Below $100K of daily throughput, rebate income is insufficient to offset the directional noise in trading P/L. The strategy requires scale to work.

4. Paired Position Management

The reference book enters first and second sides with a median lag of 4,206 seconds (70 minutes). This is a staged pair entry, not simultaneous.

Recommended paired entry sequence:

def manage_paired_entry(market, entry_plan):
    # Step 1: Enter dominant side first
    fill_dominant = submit_buy(
        market=market,
        outcome=entry_plan.dominant_side,
        max_price=entry_plan.dominant_ask + 0.01,  # allow 1 cent slippage
        usdc=entry_plan.dominant_size
    )
    
    # Step 2: Queue the non-dominant side entry
    # Wait for price to remain stable, then complete the pair
    schedule_delayed_entry(
        market=market,
        outcome=entry_plan.underdog_side,
        max_price=entry_plan.underdog_ask + 0.02,
        usdc=entry_plan.underdog_size,
        delay_min=30,    # minimum 30 minutes
        delay_max=240,   # maximum 4 hours before market resolves
        condition=lambda: paired_cost_still_valid(market)
    )

Why staged entry: Entering both sides simultaneously at the same timestamp signals to the book that you are a pure spread-capture bot with no directional information, potentially causing adversarial requotes. Staging the entry 30-120 minutes apart mimics genuine two-sided interest and preserves maker status on both fills.

Pair completion gate: Only complete the pair if the paired cost at the time of second entry is still sub-$1.005. If prices have moved such that the pair is no longer profitable, cancel the second-side order and take the directional exposure only.

5. Exit Strategy

The reference wallet has zero sells. All positions held to binary resolution. This is the intentional architecture.

Hold-to-resolution rationale: In the reference book's market universe (sports outcomes, match totals, exact scores), the typical market resolution time is hours to days, not seconds. Selling early would require finding a buyer at a price above your entry, paying a taker fee, and losing the paired-cost guarantee. The economics favor holding.

Exception case (do not implement by default): If a position's market becomes clearly one-sided before resolution (e.g., a match goes to 4-0 in the first half when you hold the Over 2.5 at $0.45), selling the winning side at $0.95 could lock in 2x+ returns. The reference book does not do this, and adding it would require a separate exit engine with live score monitoring. The base replication does not include this.

6. Hours Schedule

7. Category Configuration

Category weights (recommended for replication vs reference book):

The rebalancing toward MLB, NBA, and NFL at the expense of esports is the single most impactful configuration change available. If the esports -$1,869 bleed were eliminated and replaced with proportional expansion of MLB (which runs +0.91% ROI), the trading P/L would swing from -$443 to approximately +$1,000 at the same total volume.

8. Probability Model Requirements

The directional tilt at 3x+ dominance wins 77.9% of the time. To replicate this accuracy, the replicator needs a probability model that outperforms the Polymarket CLOB by at least 8-12 percentage points on its high-conviction calls.

Minimum viable probability model inputs:

Without a probability model, the strategy degrades to pure spread-capture with equal allocation (no dominance tilt). Equal allocation still generates maker rebates but loses the 77.9% high-conviction win rate that drives positive directional P/L.

Model accuracy target: The probability model needs to achieve at least 70% accuracy on its 3x+ dominance calls (the reference book hits 77.9%) to justify the larger dominant-side allocation. Below 60%, the tilt strategy subtracts from the spread-capture base.

9. Risk Profile and Controls

Monitoring checkpoints (weekly):

10. What This Playbook Deliberately Does Not Include

No sell engine. The reference book holds all positions to resolution. Adding a sell engine would require live event monitoring infrastructure and would cost taker fees on every exit. The economics of sell-before-resolution only make sense on positions with 40%+ unrealized gain, which requires a live scoring pipeline. Build that separately if desired; the base replication does not need it.

No leverage or cross-margin. Each position is fully funded at entry. The strategy's capital efficiency comes from rapid cycling of positions (sports events resolve within hours to days) rather than leverage.

No stop-loss. With a hard $98 clip maximum and a paired-cost structure, the maximum loss on any single market is bounded. Stop-losses would require selling at a taker cost before resolution and would destroy the paired-spread mechanics.

No price anchoring. The standard PR&R advice to trade only in the $0.30-$0.70 zone is actively harmful here. The paired structure requires buying both $0.05 longshot legs and $0.95 near-certainty legs to complete the spread. Applying a price filter destroys the paired economics and converts the book from spread-capture to directional-only, which loses money on trading without a perfect probability model.

No category concentration. Despite NFL having the highest ROI (7.63%), the NFL schedule is seasonal and insufficient to generate the volume needed for meaningful maker rebates. The multi-category breadth is architecturally necessary, not optional. Reduce weak categories; do not eliminate them entirely.

# Pseudocode skeleton - outline only
async def run_market_maker():
    prob_model = load_probability_model()   # Pinnacle-calibrated pre-game probs
    clob_feed  = connect_polymarket_ws()    # real-time L2 orderbook
    score_feed = connect_sports_api()       # live scores for in-play adjustment
    
    while True:
        hour = utc_hour(now())
        if hour in (0, 3):                  # pause hours
            await asyncio.sleep(60)
            continue
        
        volume_factor = get_volume_factor(hour, day_of_week())  # from schedule above
        
        for market in scanner.active_markets(categories=ALLOWED_CATS):
            snap = clob_feed.snapshot(market)
            paired_cost = snap.yes_ask + snap.no_ask
            
            if paired_cost >= 1.005:        # no spread to capture
                continue
            
            fair_prob = prob_model.predict(market, score_feed.state(market))
            plan = size_allocation(fair_prob, snap, volume_factor)
            
            # Enter dominant side now
            await submit_buy(market, plan.dominant_side, plan.dominant_size)
            
            # Schedule non-dominant side for 30-120 minutes later
            schedule_pair_completion(market, plan, delay_minutes=randint(30, 120))
        
        await asyncio.sleep(5)   # scan cycle

# Target: 40,000+ fills/47 days = ~850 fills/day
# Expected account P/L at reference scale: +$1,500-$2,500/47 days
# Primary driver: maker rebates at ~0.29% of notional
# Secondary driver: 77.9% dominant-side accuracy on 3x+ conviction markets

Run 22 hours per day (pausing worst hours). Reconcile daily. Expect account P/L driven 80-90% by maker rebates. Monitor the trading P/L weekly; if it trends below -$1,000 per 47-day period before rebates, the probability model or category mix needs adjustment.