Doggystyie

Wallet: 0x0484e64092ba4108c2786b61e6fc052d3bf41b1a Window: 2026-06-04 to 2026-07-02 (29 calendar days, 28 active) Universe: 172,530 trades across 6,412 unique markets, $3,480,124 gross BUY volume Account P&L (Polymarket verified): +$58,791.50 total. Trading P&L: -$34,876.75. Rewards: +$93,668.25.

The Punchline

This wallet is a high-frequency both-sides liquidity farmer, not a directional bettor. It opens every new Bitcoin 5-minute Up/Down market window on Polymarket, buys both sides, holds to resolution, and collects whatever the dominant side returns while absorbing the guaranteed spread loss. The trading book is structurally negative by design: paying a small spread across 172,530 fills generates the volume that earns Polymarket's liquidity-mining rewards, which at $93,668 over 29 days dwarf the -$34,877 trading loss.

The strategy's profitability is entirely a function of the rewards program. Strip rewards out and the trading P&L is -1.0% ROI on $3.48M deployed. Add rewards back and the wallet nets +$58,791 (+1.69% on deployed volume, or about $2,030 per active day). This is not a prediction market edge. It is a volume-farming operation that uses prediction market infrastructure.

---

What He Trades

The universe is BTC Up/Down only:

• All 6,412 markets are btc-updown-5m-* or btc-updown-15m-* slugs
• Zero sports, zero politics, zero ETH, zero SOL
• 100% Crypto category, 100% Bitcoin

The CSV sample confirms the pattern without exception. Every row is Bitcoin Up or Down - [Date], [Time Window] ET. The bot covers consecutive 5-minute windows continuously: btc-updown-5m-1782996600, btc-updown-5m-1782996300, btc-updown-5m-1782996000, btc-updown-5m-1782995700 - each slug timestamp 300 seconds apart, indicating the bot processes every available window.

The hourly trade distribution shows near-continuous coverage: the lowest hour (13:00 UTC, 5,686 trades) still has significant volume, and the highest (0:00 UTC, 8,269 trades) is only 45% above the lowest. There is no sleep window. Unlike SirMartingale's sharp overnight gap, this bot runs 24/7/28 with no meaningful dead zones.

---

What He Does - The Mechanics of Both-Sides Farming

The structural signature is unambiguous:

• Both-sides rate: 99.5% (6,381 of 6,412 markets had both Up and Down purchased)
• Zero SELL trades in the entire dataset
• Median paired cost: $1.013 (the bot pays 1.3 cents above the $1.00 true cost on average per market)

Every time a new 5-minute window opens, the bot submits buy orders for both Up and Down. It never sells. It holds every position to resolution. One side pays $1.00, one side pays $0.00. The net per market is: paired_cost × total_shares_deployed - $1.00 × winning_shares. Since paired cost averages $1.013, the expected net is slightly negative for every market regardless of which side wins.

This is intentional. The guaranteed small loss on each market is the "fee" paid to generate volume. The reward income makes up for it many times over.

A representative trade cluster from the CSV (July 2, 8:40AM-8:45AM ET):

The market btc-updown-5m-1782996000 resolved Down. The bot bought:

• Down at $0.88 (26.62 USDC), $0.88 (26.62), $0.90 (27.19), $0.89 (26.91), $0.90 (27.19) - heavy Down side
• Up at $0.11 (3.51), $0.10 (3.19), $0.08 (2.55), $0.12 (3.82), $0.12 (3.82) - small Up hedge

Down wins. The bot collects $1.00 on each Down share and $0.00 on each Up share. Net: roughly +$3-5 on the market after accounting for the cost of the losing Up shares. This is a 3x+ dominance ratio market where the bot correctly identified Down as the likely winner and tilted heavily. At 96.8% win rate in the 3x+ bucket, these high-tilt markets are a meaningful secondary alpha source.

---

The Order of Operations - One Market, Trade by Trade

Bitcoin Up or Down - June 6, 1:20PM-1:25PM ET (btc-updown-5m-1780766400). Resolved: Down wins.

Walk-through: The bot enters both sides continuously across the full 5-minute window, starting with roughly equal allocation and gradually tilting toward Down as Down's price drifts lower (indicating the orderbook sees Down as more likely). By the end of the window, the Down exposure at 81 cents is far heavier than the Up exposure at 12-20 cents. This is the dominance ratio signal at work: the bot is not symmetric, it has a mild directional read that results in a 3x+ tilt toward the winning side.

Down resolves as winner. The Down shares pay $1.00 each. Up shares pay $0.00. The market's net P&L is modestly positive because the bot correctly identified Down as the high-probability side.

This market was one of the 1,486 markets in the 3x+ dominance bucket (96.8% win rate). The best single market in the dataset paid +$71.94 on $648 deployed. The worst single market lost -$99.72 on $909 deployed. Both are within the tight structural bounds of the strategy.

---

Why It Works - The Math

<pre><code>Trading drag per market (simplified): Paired cost: ~$1.013 per dollar deployed per market pair Expected net: -$0.013 per $1.00 paired = -1.3% per market

Scale: $3,480,124 deployed over 29 days Trading P&L: -$34,877 = -1.0% of volume (slightly better than expected due to directional tilt alpha in high-dominance markets)

Reward income per day: $93,668 / 28 active days = $3,345/day Trading drag per day: $34,877 / 28 active days = $1,246/day Net daily profit: $3,345 - $1,246 = $2,099/day

Reward yield on volume: $93,668 / $3,480,124 = 2.69% of deployed volume Trading drag: 1.0% of deployed volume Net yield: 1.69% of deployed volume per 29 days

Break-even reward rate: if rewards fall below 1.0% of volume, strategy loses money Current margin: 1.69% net vs 1.0% floor = 69 basis points of cushion</code></pre>

The directional tilt provides secondary alpha. The dominance data shows:

The bot's directional signal is genuinely sharp. At 3x+ tilt, it wins 96.8% of the time. This is not random: it is reading something in the BTC price action or orderbook that predicts the 5-minute outcome with high accuracy. However, even this directional alpha cannot overcome the trading drag on its own - the primary profit engine is the rewards program.

---

Phase 1 - Trader Profile

Scale and Activity:

• 172,530 trades in 28 active days = 6,162 trades/day average
• $3,480,124 BUY notional, zero SELL notional
• 6,412 unique markets, 6,412 unique events (one cluster per market)
• No SELL activity anywhere in the dataset

Trade Size Distribution (extremely tight):

The P99-to-median ratio is 3.1x and the max-to-median ratio is 4.7x. This is the most compressed size distribution possible for a live trading book. There is no power-law; there is essentially no size variation. The bot fires clips of $10-60 with no large outliers. The max fill of $83.52 is barely 4x the median and appears to be double-fill aggregation (60 shares at higher prices).

Execution Signature:

• Median inter-fill gap: 9.0 seconds
• 51.3% of fills under 10 seconds
• 97.1% of fills under 60 seconds
• 100% under 3,600 seconds (all fills within the same hour)
• Mean gap: 14.9 seconds

The 9-second median and the 51% sub-10-second rate indicate automated execution. Fills within a market cluster happen rapidly, separated by 1-15 seconds, creating the multi-fill-per-minute pattern visible in every row of the CSV.

Both-sides participation:

The 27.5% sub-$1.00 paired cost markets are the genuinely profitable trading windows - the bot locked in a guaranteed spread gain on those. The 72.5% above $1.00 are the ones where it paid a spread. On average, $1.013.

Second-side lag: Median 4 seconds between entering first and second side of a paired market. This confirms intentional pairing, not opportunistic hedging.

Archetype: LIQUIDITY FARMER with secondary DIRECTIONAL OVERLAY

---

Phase 2 - Core Strategy Identification

This is unambiguously a both-sides liquidity provisioner, with a superimposed directional signal that creates asymmetric allocation within each market.

Classification:

• A (Both-Sides Spread/Volume Capture): Primary - 99.5% both-sides rate, zero sells, hold-to-resolution
• B (Directional Overlay): Secondary - dominance ratios up to 10x+ on individual markets, 96.8% win rate at 3x+

NOT:

• A pure spread capper (many paired costs above $1.00)
• A directional bettor (buys both sides every market, no SELL engine)
• A copy trader (no lag signature, 5-minute windows move too fast)
• A latency arbitrageur (no SELL leg to capture spread)

The core value proposition: Generate Polymarket liquidity-mining rewards by deploying volume into BTC 5-minute markets continuously. Use a directional signal (likely BTC spot price or orderbook mid) to tilt the allocation toward the favored outcome within each market, recovering some of the spread cost through directional wins.

---

Phase 3 - Dominance Ratio Analysis

The dominance data tells the most interesting structural story in this wallet. The bot is not making random bets - it has genuine predictive power when it tilts.

Three observations:

1. The win rate at 60.4% for near-equal allocation (1.0-1.5x) is already significantly above 50%. Even when the bot makes nearly equal bets on both sides, it is putting slightly more money on the right side.

1. The win rate climbs monotonically and dramatically with conviction: 60% to 80% to 87% to 97%. This is one of the strongest dominance-win-rate curves in the PR&R dataset.

1. The mean paired cost decreases as dominance increases ($1.016 at 1x-1.5x vs $1.006 at 3x+). High-conviction markets also have tighter spreads. This makes sense: when the bot has strong directional signal, one side of the orderbook is probably thicker and cheaper, making the paired entry more efficient.

The paired cost distribution across all buckets is tightly centered around $1.01, with outliers in both directions (some below $0.65, some above $1.26). The outliers below $1.00 represent markets where the bot captured a genuine spread gain. The outliers above $1.10 represent markets where liquidity was thin and the bot paid significantly to enter both sides.

---

Phase 4 - Entry Price Analysis

The price-band distribution is remarkably smooth and covers the full range:

Every single price band produces negative trading ROI. This is the structural characteristic of the strategy: it is not trying to win on individual price bands. The win rates are perfectly calibrated (4.7% at sub-$0.10, matching ~5% probability; 94.9% at $0.90+, matching ~95%) - the market is pricing Bitcoin outcomes correctly and the bot is paying fair prices.

The ROI is most negative in the $0.40-$0.70 band (-1.07% to -1.12%) and less negative at the extremes (-0.49% at sub-$0.10, -0.80% at $0.90+). This is expected: near-certainty bets (very cheap or very expensive) have thinner spreads on the orderbook. The coin-flip zone has the widest spreads.

The sub-bucket concentration check: both Up and Down fills appear at nearly every price from $0.01 to $0.99 across the 29-day sample. No single cent dominates. This confirms the bot is not anchored to a specific price - it accepts whatever the orderbook offers when it fires each window.

---

Phase 5 - Category and Market-Type Breakdown

There is one category:

All 6,412 markets are Bitcoin 5-minute or 15-minute Up/Down windows. The 49.8% overall win rate is slightly below 50% because the losing side of each paired market is by definition a full loss, and the paired cost is slightly above $1.00.

The CSV sample shows a handful of btc-updown-15m-* appearances (e.g., btc-updown-15m-1780766100 traded concurrently with the 5-minute window at 17:25 on June 6). These are minor allocations, typically 5-10 shares at $0.11-$0.89, appearing to hedge or supplement the primary 5-minute coverage. The bulk of volume and trade count is 5-minute windows.

---

Phase 6 - Timing and Execution Analysis

Hourly distribution (UTC):

The trade count by hour shows near-continuous coverage with no dead zones:

The bot is most active in the early UTC morning (roughly midnight-6am UTC, which is 8pm-2am Eastern). The lightest period is 13:00-15:00 UTC (9am-11am Eastern). There is no sleep window. This 24/7 profile is consistent with an automated system with no human supervision required overnight.

Hourly P&L: All hours produce negative trading P&L, ranging from -$88 (21:00 UTC) to -$2,228 (01:00 UTC). The variation tracks volume: higher-volume hours lose more dollars but at similar percentage rates (~-1.0% consistently). No hour has structural positive trading P&L.

Day-of-week P&L:

Sunday and Saturday show the highest trading losses in both absolute and percentage terms, despite having lower win rates. This may reflect thinner liquidity on weekends (higher paired costs), or a different mix of market conditions. However, since the rewards income likely scales with volume regardless of day, the weekend trading drag does not necessarily mean weekends are worse for overall profitability.

Entry timing within market windows:

From the CSV, the bot's pattern within each 5-minute window is clear: it fires an initial burst of 6-15 fills across both sides in the first 60-120 seconds, then continues adding through the window. The 4-second median second-side lag means Up and Down fills are nearly simultaneous. By 2-3 minutes into the window, the bot has typically deployed most of its intended volume and stops adding. It never exits early - all positions hold to resolution.

---

Phase 7 - Filter Experiments

The standard filter battery is entirely destructive or inapplicable for this strategy. Every filter tested either makes no difference (category filter = identity) or degrades the trading ROI by reducing volume. This makes perfect sense: the strategy's value is derived from volume, not from selecting high-quality individual trades. Filtering removes fills that contribute to the reward-earning volume total without a compensating improvement in trading outcome.

The high-conviction filter is the closest thing to a "useful" filter: applying it to the 2x+ dominance markets (where the dominant side is identified) yields 91.7% win rate on $1.097M capital, with a -0.86% ROI (vs -1.00% baseline). But applying this filter in practice would reduce total volume by 68%, collapsing the reward income by roughly the same proportion while saving only $25,391 in trading losses. The net effect on account P&L would be strongly negative.

---

Phase 8 - Rolling Window Consistency

The trading P&L is negative and declining throughout the entire window. Every rolling 7-day and 15-day trading window is negative, without exception. The worst 7-day window hit -$13,875. This is structurally expected: the bot bleeds a small amount every day from the spread, and that bleeding compounds.

The account P&L (including rewards) tells a completely different story:

The account cumulative line is monotonically increasing, growing approximately $2,030 per day. The rewards income is steady, regular, and sufficient to more than offset the daily trading drag. Every day the bot operates adds positive value to the account, even though every day of trading loses money.

---

Phase 9 - P&L Decomposition

Spread P&L decomposition from pnl_decomp:

• Spread P&L (guaranteed gain from sub-$1.00 paired markets): -$40,705 (net negative, meaning the bot is paying more than it gains from spread)
• Realized total: -$34,877

The spread P&L being -$40,705 while realized total is -$34,877 means the directional alpha from tilted positions contributes approximately +$5,828 of positive P&L that partially offsets the spread drag. Without the directional signal, trading losses would be roughly 17% larger.

The entire profitability case rests on $93,668 of non-trading income. If Polymarket's rewards program did not exist, this strategy would lose $34,877 on $3.48M of capital, a complete waste of capital deployment. With rewards, it returns +$58,791 on the same capital base, or approximately $2,030/day in net income.

The rewards yield (~2.69% of volume) exceeds the trading drag (~1.00% of volume) by 169 basis points. That margin is the strategy's entire value proposition.

---

Phase 10 - Strategy Specification (short form; full detail in playbook)

One-sentence summary: A fully-automated 24/7 bot that buys both Up and Down on every Bitcoin 5-minute Polymarket window, holds to resolution, and earns its profit from Polymarket liquidity-mining rewards rather than from directional accuracy.

Edge source: Polymarket rewards program payout on high-volume both-sides participation in BTC Up/Down markets.

Secondary edge: Genuine directional signal (likely BTC spot-based) that creates asymmetric allocation within each market window, achieving 96.8% accuracy at 3x+ dominance.

What works: Continuous 24/7 operation across all BTC 5-minute windows. Tight sizing ($17-$50/fill). No exits before resolution. Both-sides coverage with mild directional tilt.

What bleeds: Trading P&L is -1.0% of volume by design. Saturday/Sunday show slightly worse trading ROI (-1.22% / -1.55%), possibly due to wider spreads in low-liquidity weekend sessions.

Replication prerequisite: Access to the Polymarket rewards program at sufficient yield to exceed the structural ~1.0% trading drag. Without confirmed reward rate, this strategy should not be run.

Wallet: 0x0484e64092ba4108c2786b61e6fc052d3bf41b1a Window: 2026-06-04 → 2026-07-02 (28 active / 29 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: 99.52% (6,381 of 6,412 markets)
• Median paired cost: $1.0135
• Mean paired cost: $1.0114
• Paired cost % under $1.00: 27.5%
• Paired cost % under $0.97: 6.4%
• Median 2nd-side hedge lag: 4s

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: 0 of 29 (0.0%)
• Rolling 7-day P/L range: -$13,876 → -$454
• Rolling 15-day windows green: 0 of 29 (0.0%)
• Rolling 15-day P/L range: -$23,399 → -$454

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 2026-07-05 06:26 UTC.

Wallet: 0x0484e64092ba4108c2786b61e6fc052d3bf41b1a Window: 2026-06-04 to 2026-07-02 Baseline: 172,530 BUYs, 49.8% WR, $3,480,124 deployed, -$34,877 trading P&L, -1.00% trading ROI Account total (verified): +$58,791 (includes +$93,668 in liquidity-mining rewards)

The headline result

Every filter either makes no difference, degrades trading ROI, or destroys the strategy entirely by reducing volume. This is not a failure of the filter framework. It is the correct finding for a volume-farming strategy. The trader's alpha does not live in a subset of high-quality trades that filters could isolate - it lives in the aggregate volume that generates the reward income. Every trade removed from the book, regardless of its individual trading quality, reduces the reward allocation.

The single most important insight from this filter analysis: standard directional-betting filters (price band, dominance, hour exclusion) are structurally misaligned with the farming model. They are designed to find the signal within a noisy book. Here there is no signal to find in the trading P&L; the signal is in the reward yield on total volume.

---

Filter results table

---

Filter-by-filter commentary

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

Applying the standard sweet-spot filter cuts the trade count from 172,530 to 95,396 (a 44.7% reduction) and the deployed capital from $3.48M to $2.01M. The trading ROI degrades from -1.00% to -1.08%. That is both a larger absolute trading loss per dollar deployed AND a smaller total volume. Both directions are wrong.

The degradation makes mechanical sense: the $0.30-$0.70 zone is where paired costs are highest (the market is thickest near coin-flip pricing, creating more slippage). Concentrating in this band removes the sub-$0.10 and sub-$0.20 fills where the bot's directional signal is sharpest (buying near-certain outcomes creates very low spread situations). It also removes the $0.80-$0.99 fills where the bot is buying near-certainties cheaply.

Beyond the trading ROI impact, the reward-income consequence is catastrophic: losing 44.7% of volume likely means losing roughly 44.7% of reward allocation, or approximately $41,900 in lost rewards. Against the $13,097 of saved trading losses, this is a net account P&L reduction of about $28,800. Do not apply this filter.

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

The high-conviction filter keeps only trades on the dominant side of markets where the bot allocated at 2x or greater dominance ratio. This produces 37,436 trades with a 91.7% win rate and -0.86% trading ROI. The trading ROI improvement is real (14 bps) but the volume reduction is massive: 78.3% of trades and 68.5% of capital are excluded.

The 91.7% win rate is genuinely impressive, confirming that the bot's directional signal is real. But applied as a filter in isolation, this approach saves approximately $25,391 in trading losses while likely sacrificing about $64,200 in reward income (68% of $93,668). Net account P&L change: approximately -$38,800.

The high-conviction filter would only make sense if reward income were zero. As a purely directional strategy, it actually has merit. As a filter on top of a farming strategy, it is deeply counterproductive.

3. Category filter (Crypto only) NO-OP

100% of all trades are already Crypto category. This filter is identity-equivalent to unfiltered. No change in any metric.

4. Hour exclusion filter (exclude worst 4 hours: UTC 6, 11, 18, 21) DESTRUCTIVE

The four worst trading hours by P&L are 06:00 UTC (-$2,162), 11:00 UTC (-$1,763), 18:00 UTC (-$869), and 21:00 UTC (-$89). Excluding them removes 29,125 trades (16.9% of volume) and reduces trading P&L loss from -$34,877 to -$29,994, saving $4,883.

However, the trading ROI worsens slightly (-1.04% vs -1.00%) because the worst-P&L hours in absolute dollars also had the highest trade counts. Removing high-volume hours with average ROI does not improve the per-dollar loss rate.

Unlike SirMartingale (where the hour filter was a no-op because he already wasn't trading bad hours), here the filter actively removes volume that was generating reward income. Losing 16.9% of trades to save 14% of trading losses means reward income drops by approximately $15,850 while trading losses fall by only $4,883. Net account impact: approximately -$10,967.

The bot's 24/7 schedule is not inefficient - every active hour contributes to the reward allocation at a roughly similar rate. There is no "best hours" to concentrate into.

5. Combined stack filter MOST DESTRUCTIVE

The combined stack (price $0.30-$0.70 + dom 2x+) produces 79,430 trades, 50.8% win rate, -$18,752 P&L, -1.12% ROI. This is the worst trading ROI of any configuration tested. The stacking compounds the negative effects without adding synergy: high-dominance markets with coin-flip pricing are exactly the wrong combination (narrow directional edge, widest spreads).

Volume loss is 54% of baseline. Estimated reward income loss: approximately $50,600. Estimated trading P&L savings: $16,125. Net account impact: approximately -$34,475. This filter combination would convert a profitable operation into a losing one.

---

What filters *would* make sense for this strategy

The standard filter framework is designed for directional bettors. For a volume farmer, the relevant optimization dimensions are entirely different:

The one genuinely actionable optimization: preferentially size up in markets where the dominance signal is 3x+ and the expected entry price for the dominant side is still favorable. At 96.8% win rate, these markets contribute positive expected value even on the trading leg. More capital in these markets would improve trading ROI without sacrificing volume.

But this is a sizing optimization, not a filter. It does not remove trades; it reweights them.

---

Bottom line for replication

Do not apply standard PR&R directional-betting filters to this wallet. The strategy's profitability is volume-dependent. Any filter that reduces trade count also reduces reward income by roughly the same proportion, and reward income is 2.69x the trading loss that filters are trying to reduce.

The only actionable insight from filter analysis: if you cannot confirm a reward yield above ~1.5% of volume (leaving a 50 bps cushion above the structural ~1.0% trading drag), do not run this strategy at all. The filter question becomes moot if the reward rate is insufficient.

Source wallet: 0x0484e64092ba4108c2786b61e6fc052d3bf41b1a Strategy: Both-sides BTC Up/Down liquidity farming with directional overlay Reference book: $3,480,124 BUY notional, -$34,877 trading P&L, +$93,668 rewards, +$58,791 account P&L in 29 days

---

One-paragraph operator brief

Build a 24/7 automated bot that opens every new Bitcoin 5-minute Up/Down market on Polymarket, purchases both Up and Down outcomes at current orderbook prices, and holds all positions to resolution. Apply a directional signal (BTC spot price or orderbook mid) to tilt allocation toward the more likely outcome within each market. Size each fill at $15-$50, cap per-market total at $200. Never sell before resolution. Run continuously with no sleep window. The trading book will bleed approximately -1.0% of deployed volume per day by design. The strategy is profitable only if Polymarket's liquidity-mining rewards exceed that drag. At the reference wallet's reward rate of ~2.69% of volume, the net yield is approximately +$2,000/day on $120,000/day of deployed capital. Before building this bot, confirm the current reward rate explicitly. If rewards fall below 1.0% of volume, the strategy loses money.

---

1. Market selection

Why BTC 5-minute only: The reference wallet generates 99%+ of its trade count and volume in BTC 5-minute windows. These markets exist continuously (one new window every 5 minutes, 288 per day), are actively traded by multiple participants providing liquidity, and appear to generate the largest reward allocation. The 15-minute windows add minor supplemental volume.

Why cover every window: The reward allocation likely depends on total volume across all windows over a period, not on selective participation. Skipping windows reduces volume and reduces reward income without a corresponding improvement in trading P&L.

---

2. Entry logic

def should_enter_market(market):
    # Market type whitelist
    if "btc-updown" not in market.slug:
        return False
    
    # Timing: do not enter in final 30 seconds of window
    sec_remaining = market.close_time - now()
    if sec_remaining < 30:
        return False
    
    # Coverage: always enter if within the window
    return True

def compute_allocation(market, btc_signal):
    """
    btc_signal: probability estimate for BTC Up in this window
    Returns (up_usdc, down_usdc) allocation
    """
    base_clip = 30.0  # USD per side, baseline
    
    if btc_signal is None:
        # No signal: symmetric allocation
        return base_clip, base_clip
    
    # Directional tilt based on signal strength
    gap = abs(btc_signal - 0.50)
    
    if gap < 0.05:
        # Near coin-flip: symmetric
        return base_clip, base_clip
    elif gap < 0.15:
        # Mild conviction (1.5-2x dominance equivalent)
        favored = base_clip * 1.5
        hedge = base_clip * 0.7
    elif gap < 0.25:
        # Moderate conviction (2-3x dominance equivalent)
        favored = base_clip * 2.0
        hedge = base_clip * 0.6
    else:
        # High conviction (3x+ dominance equivalent)
        favored = base_clip * 3.0
        hedge = base_clip * 0.5
    
    if btc_signal > 0.5:
        return favored, hedge
    else:
        return hedge, favored

---

3. Sizing model

The reference wallet uses near-uniform sizing with no power-law. Replicate this exactly:

Do not scale clips with bankroll beyond these caps. The reference book maintains the same $17-$50 median regardless of the wallet's total capital. The cap is structural, not conservatism - larger fills move prices on thin BTC 5-minute markets, increasing paired costs and destroying the economics.

Multi-fill pattern: The reference executes 15-40 fills per market window across a 60-180 second entry period. Do not submit one large order. Walk the orderbook with multiple smaller clips to minimize market impact.

---

4. Both-sides allocation mechanics

Every market entry requires BOTH sides:

def execute_market_entry(market, up_usdc, down_usdc):
    # Submit both sides nearly simultaneously (< 5 second lag)
    up_fills = walk_book_buy(market, "Up", max_usdc=up_usdc,
                             clip_size=15.0, max_price=None)
    down_fills = walk_book_buy(market, "Down", max_usdc=down_usdc,
                               clip_size=15.0, max_price=None)
    
    # Continue adding across the window (spread entries over 60-180 seconds)
    # Do not front-load the entire allocation in first fill
    schedule_additional_fills(market, remaining_budget, window_close_time)
    
    return up_fills + down_fills

Key constraints:

• Never skip the hedge side, even for high-conviction markets. The reference wallet buys the "wrong" side on every single market (the losing side always gets something). This is intentional: the reward program likely measures two-sided volume.
• Maintain the second-side lag under 5 seconds. Same-second pairing is common in the reference book.
• Spread fills across the full window rather than entering in one burst. The reference shows continuous adding from window open to ~3 minutes in.

---

5. Exit strategy

No exits. Zero SELL trades in the 29-day reference window. Hold every position to resolution.

def manage_positions():
    for position in open_positions:
        # Never sell before resolution
        # Market will settle at $1.00 (win) or $0.00 (loss)
        # Do not post sell orders
        # Do not panic-exit if position is underwater mid-window
        pass
    # Settlement happens automatically at window close

The hold-to-resolution rule is not laziness - it is structural. The strategy relies on resolution payouts for its full accounting. Selling early at a premium would capture some unrealized gain but would also eliminate the paired hedge's offsetting effect on the accounting and potentially reduce the "volume held to resolution" metric that rewards may track.

---

6. Directional signal construction

The reference bot's 96.8% dominant-side win rate at 3x+ dominance across 1,486 markets confirms it has a genuine predictive signal. Based on the market type (BTC 5-minute Up/Down: did BTC go up or down in this 5-minute window?) the most plausible signals are:

def estimate_btc_up_probability(market, btc_feed):
    """
    Estimate probability that BTC price will be higher at market close
    than at market open, for a given 5-minute window.
    
    Signal options (in order of likely effectiveness):
    """
    # Option 1: Orderbook imbalance
    up_mid = market.up_side.best_ask
    down_mid = market.down_side.best_ask
    clob_prob_up = up_mid / (up_mid + down_mid)  # rough fair value
    
    # Option 2: BTC spot momentum (current price vs window open)
    btc_now = btc_feed.latest_price()
    btc_open = btc_feed.price_at(market.open_time)
    momentum = (btc_now - btc_open) / btc_open  # % change so far
    
    # Option 3: BTC realized volatility + momentum combination
    vol = btc_feed.realized_vol_5min()
    zscore = momentum / vol if vol > 0 else 0
    prob_up = norm.cdf(zscore)  # convert to probability
    
    return prob_up

The signal must be fast (sub-second) and persistent (available throughout the 5-minute window as fills are staggered). The reference bot shows a 4-second second-side lag, suggesting the signal fires before order submission begins.

---

7. Operational requirements

Working capital sizing:

Reference: $3,480,124 BUY volume over 28 days = $124,290/day
Peak concurrent exposure: ~15 open markets × $150/market = ~$2,250
Required liquid balance: $15,000-$30,000 (provides 7-13x buffer for
  settlement timing gaps)

The working capital requirement is surprisingly modest because positions turn over every 5 minutes. The wallet never has more than ~30-40 markets open simultaneously (the ones that opened in the last 5-15 minutes). At $150 average per market, peak instantaneous exposure is $4,500-$6,000. A $15,000 USDC balance provides comfortable headroom.

---

8. Risk profile

Maximum per-day trading loss: At -1.0% ROI on ~$120K of daily volume, the expected daily trading drag is ~$1,200. The absolute worst observed day in the reference window was approximately -$2,700, roughly 2.2x the average. A $30,000 balance can absorb 11+ worst-case trading days before rewards need to be withdrawn to cover losses.

---

9. Diagnostic checklist - "is the bot still working?"

Run daily:

Run weekly:

• Compute net account yield (trading P&L + estimated rewards) as % of volume. Target: >1.5%.
• Compare your BTC 5m volume to apparent total market volume. If your share has dropped significantly, competition may be diluting rewards.
• Review paired cost distribution. Check if new high-cost outliers are emerging.

---

10. What this playbook deliberately does NOT include

No sleep window. Unlike directional bettors who benefit from scheduling around peak-signal hours, this strategy's value comes from covering every window. Gaps in coverage reduce volume and reward income. The reference bot runs 24/7 without exception.

No price-band filtering. All price bands produce negative trading ROI, and filtering removes volume that earns rewards. Do not attempt to "trade smarter" by restricting entry prices.

No SELL engine. Unlike SirMartingale's aggressive exit-into-strength mechanic, this strategy holds everything to resolution. Adding a SELL engine would introduce execution complexity and would likely reduce the "volume held to settlement" metric that reward programs typically track.

No directional-only mode. Even at 96.8% accurate at 3x+ conviction, running the dominant side only and dropping the hedge would reduce volume by roughly 25-30%, sacrificing more in rewards than the hedge costs in trading P&L.

No leveraged sizing. The strategy's economics work at $15-$50 per fill because paired costs stay near $1.013. Scaling clip sizes beyond $84 starts moving prices on these thin markets, increasing paired costs and degrading the trading P&L faster than rewards scale.

No expansion to ETH, SOL, or other categories. The reward program appears optimized around BTC Up/Down. Other categories have different reward structures, different paired cost dynamics, and different liquidity profiles. Do not dilute the core BTC 5m coverage to branch into other markets without separately validating the reward economics for each category.

The strategy works because it is relentlessly simple: buy every window, both sides, hold to resolution, collect rewards. Every "improvement" introduces friction, reduces volume, or adds cost. Trust the mechanics.