FlippingSharks

Wallet: 0xc387c2a40d389f17b723b6bba9b18b7dbd2de4f4 Window: 2026-05-20 to 2026-05-31 (12 calendar days, 12 active) Universe: 89,684 trades across 2,883 markets, $1,359,832 gross turnover Net P/L (computed): +$4,121 on $1,359,559 BUY notional = +0.30% ROI in 12 days

The Punchline

FlippingSharks is a fully automated bilateral spread-capture market maker operating exclusively on Polymarket's 5-minute Bitcoin Up/Down series. It buys both sides of 98.3% of all markets it touches, with the two legs entering within a median of 7 seconds of each other, and holds every position to settlement. There is no early exit, no stop loss, and no directional bet in the conventional sense. The strategy makes money because the combined price of Up and Down shares in these markets is frequently below $1.00 at the time of entry, which guarantees a profit on paired shares regardless of which outcome resolves.

The secondary alpha source is asymmetric allocation: the wallet allocates more capital to the side that is priced cheaper (lower implied probability) at entry. Across the 902 markets where one side received 3.0x or more capital than the other, that cheaper side resolved correctly 95.8% of the time. This is not luck. The market is pricing the lower-probability side correctly on average, and the wallet is expressing a directional view through its allocation asymmetry -- it just never makes a single-sided bet.

The economics are thin per market but aggregate across volume. +$4,121 on $1.36M deployed is a 0.30% return over 12 days, which annualizes to roughly +9% on the BUY notional deployed. The actual working capital required is far less than $1.36M because positions cycle every 5 minutes and the wallet is recycling the same dollars across sequential markets. The real yield on peak instantaneous exposure is substantially higher.

What He Trades

The entire book is one instrument:

btc-updown-5m-*    89,678 BUYs    $1,359,559 notional    2,883 markets

Every btc-updown-5m- market slug in the observation window is touched. This is a blanket market-maker, not a selective trader. No other slugs appear. The 2,883 markets correspond closely to the number of 5-minute windows in 12 days (12 days x 288 windows/day = 3,456 theoretical maximum), so the wallet enters roughly 84% of available windows, skipping a portion due to the active-hours scheduling described in Phase 6.

The CSV confirms the pattern. Looking at consecutive rows, the timestamp sequence jumps from one market (e.g., btc-updown-5m-1780271400) to the next (btc-updown-5m-1780271100) as the wallet moves through sequential 5-minute windows. Within each market, both Up and Down fills appear within seconds.

There is no ETH, no BTC hourly/daily, no politics, no sports, no other asset. The universe restriction is as hard as it gets.

The Order of Operations -- One Market, Trade by Trade

This example traces the wallet's activity in Bitcoin Up or Down -- May 19, 8:20PM-8:25PM ET (btc-updown-5m-1779236400), which resolved Down.

Walk-through:

The wallet enters this market around 3 minutes before close. It fires both Up and Down buys nearly simultaneously, with the paired-entry lag visible throughout. The first fills (00:21:02 and 00:21:05) are a Down at $0.28 and an Up at $0.68 -- combined cost = $0.96, which guarantees a 4-cent profit per paired share. The wallet then continues to add both sides across the remaining window, walking the book in chunks.

By 00:23:22 the market has drifted dramatically: Down is now $0.85 and Up is $0.14. The combined cost on those fills is $0.99 -- still sub-$1.00, barely profitable on the paired shares, but the allocation has become extremely asymmetric. Down gets 60.21 shares at $0.85 ($51.18) while Up gets 60.21 shares at $0.141 ($9.01). The wallet has a 6:1 allocation skew toward Down.

The market resolves Down at 01:20 ET. The Down shares settle at $1.00 each; the Up shares settle at $0.00. Net: the wallet collects on the heavy Down allocation, loses the small Up allocation, and retains the guaranteed spread on the truly paired (equal-shares) portion. This market was a winner because (a) the spread was locked in on early paired fills, and (b) the directional skew toward Down was correct.

The 7-second median second-side lag is visible here: every time the wallet enters a large Down position, a corresponding (smaller) Up position follows within seconds. The two fills are dispatched nearly simultaneously by the bot; the offset reflects CLOB processing and gas confirmation timing, not decision-making delay.

Why It Works -- The Math

The strategy has two independent P/L sources that stack:

<pre><code>Source 1: Spread capture on sub-$1.00 paired markets pct_markets_sub_1.00 = 50.1% (from both_sides.pct_sub_1 = 0.501) median_paired_cost = $0.9996 profit per paired share = 1.0 - 0.9996 = $0.0004/share

At $4.21 median clip and ~2,835 two-sided markets: Guaranteed spread P/L estimate = (0.501 x 2835) x $0.0004/share x ~30 shares/clip = ~1,421 markets x $0.012 = ~$17 guaranteed / market in the sub-$1.00 subset (rough; actual depends on paired_shares, not trade count)

Source 2: Directional accuracy from allocation skew Dom 1.0-1.5x: 62.7% dominant-side WR (785 markets) Dom 1.5-2.0x: 74.0% dominant-side WR (496 markets) Dom 2.0-3.0x: 88.2% dominant-side WR (652 markets) Dom 3.0x+: 95.8% dominant-side WR (902 markets)

In the 3.0x+ bucket: 864 wins / 902 markets The smaller (hedge) leg costs ~25% of the dominant-leg outlay Expected net per market in this bucket is strongly positive</code></pre>

The two sources are complementary: spread capture works best when both sides are priced near $0.50 (balanced book), while directional skew works best when one side is heavily mispriced. The wallet captures spread when the book is balanced and captures directional wins when it is not -- which means there is a profitable play in essentially every market state.

The strategy's vulnerability is the 49.9% of markets where the combined VWAP exceeds $1.00. On those markets, the wallet is paying a guaranteed loss on the paired shares and must rely entirely on the directional skew to be net profitable. If the directional signal degrades (which would show up as lower dominant-side WR in the 2.0x+ buckets), those markets become consistently negative.

Phase 1 -- Trader Profile

Scale and Activity

Six SELL trades in 89,684 total is effectively zero. This is a settlement-only exit strategy.

Trade Size Distribution

The median of $4.21 vs mean of $15.16 reflects a highly right-skewed distribution -- many small clips and occasional large ones. The max of $222.75 is only 53x the median, which is moderate concentration by Polymarket standards. The top 5% of trades carry 38.7% of the capital.

Execution Signature

75.9% of consecutive fills arrive within 10 seconds. This is a fully automated bot. The 0.0-second median means many fills occur in the same timestamp block (same-second multi-leg fan-out is visible throughout the CSV). The longer tail (P90 = 39s) reflects the inter-market transition time between sequential 5-minute windows.

Second-side lag median: 7 seconds. When the wallet enters a market, the complementary side follows within 7 seconds at the median. This is the paired-entry signature of a spread-capture bot.

Phase 2 -- Core Strategy Identification

Both-sides participation rate: 98.3% (2,835 of 2,883 markets)

This is the defining statistic. A rate above 60% indicates spread-capture as the core mechanism. At 98.3%, this is an essentially pure bilateral market-maker. The 48 single-sided markets (1.7%) likely reflect edge cases where the second leg failed to fill before expiry or orderbook conditions prevented entry.

Classification: SPREAD CAPTURE + DIRECTIONAL SKEW

This wallet is Archetype A (Both-Sides Spread Capture) with a meaningful Archetype B (Directional) overlay via asymmetric allocation.

Not in scope:

• Not a copy-trader (no lag pattern, every market is simultaneous both-sides)
• Not a latency arbitrageur (no SELL engine, no early exit)
• Not a DCA accumulator (enters each market fresh, never returns to a prior market)
• Not a pure spread-only market maker (the 3.0x+ dominance bucket shows real directional signal, not just hedged exposure)

Phase 3 -- Dominance Ratio Analysis

The dominant-side win rate escalates sharply with dominance ratio. At 1.0-1.5x (near-equal allocation), the WR is 62.7% -- above random but not dramatically so. By 3.0x+, it reaches 95.8%. This is the cleanest possible evidence of a real directional signal embedded in the allocation decision.

The mean paired cost tracks the dominance ratio inversely: when the wallet tilts heavily (3.0x+), the combined VWAP averages $0.9954, meaning the spread is more reliably locked in on high-conviction markets. The 1.0-1.5x bucket has a mean paired cost of $1.0026 -- slightly above $1.00, meaning these lower-conviction markets are on average slightly lossy on the spread alone. The directional precision at high dominance more than compensates.

The practical interpretation: the wallet reads the CLOB, identifies the side that is underpriced relative to some fair-value model, and loads that side heavily while putting a small hedge on the other side. The "small hedge" side nearly always loses -- which is the cost of running a delta-hedged position -- but the dominant side wins at 96% when the conviction is high.

Phase 4 -- Entry Price Analysis

The win-rate column is a near-perfect calibration curve. At $0.00-$0.10, the WR is 5.0% (consistent with ~5% implied probability). At $0.90-$1.00, WR is 94.0%. This confirms the market is pricing outcomes correctly on average -- the strategy is not exploiting systematic mispricings in the implied probability surface.

The ROI profile is the more telling view. ROI is highest at sub-$0.20 entries (1.6%) and declines monotonically as price increases, hitting near-zero or negative at the $0.90+ zone. This is the spread-capture signature: buying the cheap side of a paired position has the highest expected dollar return per dollar deployed because those positions are small (cheap) and the paired spread is wide, while buying near-$1.00 favorites returns very little even when correct.

Sub-bucket inspection: The capital distribution is relatively smooth across all 10 bands, with the $0.50-$0.60 band as the single largest (15.6% of capital at $211,872). No single $0.01 price tick dominates. This is the opposite of a single-tick floor-bidder strategy -- the wallet is walking the entire book on every market.

Phase 5 -- Category and Market-Type Breakdown

The category breakdown is trivially uninformative: 100% of trades are Crypto. The interesting decomposition is by market within Crypto, and here the data is also thin because there is only one slug (btc-updown-5m-*). Market-to-market P/L variation is high (top market: +$235, worst market: -$178), but the variation is random noise around the mean, not a systematic sub-category signal.

Best markets by P/L:

Worst markets by P/L:

The range of -$178 to +$235 per market is small relative to the volume. The worst single market loses $178 on $1,152 of volume (-15.4%). Even the biggest single-market loss is survivable within a session that generates $4,121 net.

Phase 6 -- Timing and Execution Analysis

Hourly Activity Distribution

The wallet trades every hour of the day, but volume is notably lower in the 12:00-15:00 UTC window (the lightest 4-hour block) and peaks in the 03:00-06:00 UTC window. The wallet is not a US-hours-only operator -- it runs nearly around the clock.

The four worst hours by P/L are 4 (-$434), 20 (-$533), 11 (-$347), and 0 (-$71). Hours 4 and 20 stand out as the primary drags. The filter analysis identifies 4, 16, 19, 20 as the worst hours to exclude.

Day-of-week patterns:

Weekdays are consistently profitable; weekends (especially Sunday) are negative. Saturday and Sunday together drain -$1,612 from a book that makes +$5,733 on weekdays. This is a real and actionable pattern: the spread-capture edge is weaker on weekends, likely because BTC volatility profiles and market-maker competition patterns differ on Saturdays and Sundays.

Phase 7 -- Filter Experiments

The high-conviction filter and the hour-exclusion filter are the only two that do real work. The price-band and category filters are structurally inapplicable or destructive.

See the Filters tab for the full per-filter analysis.

Phase 8 -- Rolling Window Consistency

Weekly performance:

Rolling 7-day windows:

The rolling 7-day picture shows a notable late fade: from a peak of +$5,109 (May 27) to +$238 (May 30). The final week (May 25-31) generated only +$631 vs +$3,488 in the prior week. This could reflect increased competition (paired costs rising toward $1.00) or a period of lower BTC volatility that compressed the spread available. The strategy is not showing decay across its entire history, but the second-week weakness is worth monitoring.

All rolling 7-day windows are positive (minimum +$238), confirming the strategy is generating consistent small profits without a meaningfully negative sustained window.

Phase 9 -- P/L Decomposition

The spread P/L of -$1,869 is the key decomposition finding. Taken alone, the spread mechanism (buying both sides and holding to settlement) generates a slight net negative across the whole book -- because 49.9% of markets have combined VWAP above $1.00, and those markets lose more than the sub-$1.00 markets gain. The strategy's net profitability (+$4,121) comes entirely from the directional accuracy of the asymmetric allocation: the dominant-side wins enough of the time at high-dominance ratios to overcome the spread losses and the hedge tax.

Put differently: if the wallet lost all directional signal and split perfectly 50/50 on every market, it would be a net -$1,869 strategy. The +$5,990 gap between -$1,869 and +$4,121 is the value of the allocation signal.

Phase 10 -- Strategy Specification

One-sentence summary: A 24/7 bilateral market-making bot that enters both sides of every sequential 5-minute Bitcoin Up/Down market on Polymarket, allocates more capital to the side with lower implied probability, and holds all positions to settlement to capture paired spread and directional wins.

What works: Both-sides entry with 7-second lag. Asymmetric allocation driven by live CLOB prices. Settlement-only exits. 3.0x+ dominance markets (95.8% dominant-WR). Weekdays. Hours 12, 15, 21 UTC (top performers).

What drags: Markets where combined VWAP exceeds $1.00 (49.9% of markets). Weekends (especially Sunday, -0.63% ROI). Hours 4, 11, 20 UTC.

What replicators must build: A persistent CLOB connection to Polymarket's 5-minute BTC series. A paired-entry logic that fires both sides within seconds at entry. An allocation model that tilts more toward the lower-priced side. A settlement-only exit (no stop-loss, no early sell). A scheduling layer that caps activity during the identified weak-ROI hours and weekends.

See the Playbook tab for the complete implementable spec.

Wallet: 0xc387c2a40d389f17b723b6bba9b18b7dbd2de4f4 Window: 2026-05-20 → 2026-05-31 (12 active / 12 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: 98.34% (2,835 of 2,883 markets)
• Median paired cost: $0.9996
• Mean paired cost: $0.9981
• Paired cost % under $1.00: 50.1%
• Paired cost % under $0.97: 25.6%
• Median 2nd-side hedge lag: 7s

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: 12 of 12 (100.0%)
• Rolling 7-day P/L range: +$238 → +$5,109
• Rolling 15-day windows green: 12 of 12 (100.0%)
• Rolling 15-day P/L range: +$360 → +$5,881

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 2026-06-03 05:10 UTC.

Wallet: 0xc387c2a40d389f17b723b6bba9b18b7dbd2de4f4 Window: 2026-05-20 to 2026-05-31 Baseline: 89,678 resolved BUYs, 47.9% WR, $1,359,559 deployed, +$4,120 P/L, +0.30% ROI

The headline result

Two filters do real positive work. Three are structural no-ops or worse. The most important finding is a negative finding: the price-band filter that is most commonly applied to Polymarket directional traders destroys P/L in this book because it removes the high-ROI cheap-side fills that are load-bearing components of the spread-capture mechanism.

The single most useful refinement is scheduling: excluding the four worst-performing hours boosts ROI from 0.30% to 0.40% with no change to the underlying strategy. The high-conviction dominance filter delivers a nearly identical absolute P/L (+$4,164 vs baseline +$4,120) on only 22% of the capital, which sounds impressive but requires running the other 78% of capital somewhere else -- it doesn't produce lift in isolation.

Filter results table

Filter-by-filter commentary

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

Applying the standard sweet-spot filter cuts 46% of trade volume and reduces P/L by $1,763 (from +$4,120 to +$2,357). The WR inside the band improves marginally from 47.9% to 50.2%, but the absolute profit falls substantially.

The reason is structural. The spread-capture strategy requires both the cheap side (which prices below $0.30 or above $0.70) and the expensive side (which sits near the opposite pole) to close out the paired position. When the wallet buys Down at $0.12 and Up at $0.85, it needs both fills to capture the $0.03 locked spread. Filtering to $0.30-$0.70 removes the cheap-side fills entirely, leaving the strategy without its paired counterpart. The result is not a directional book with higher win rates -- it is a broken paired strategy with half its positions missing.

More specifically: the sub-$0.10 and $0.10-$0.20 entry bands deliver +$179 and +$578 respectively on low capital, with ROI of 1.66% and 1.62%. These are the hedge legs of high-dominance paired positions. Removing them doesn't improve the book; it destroys the P/L contribution of the cheap-side fills while leaving their paired counterparts intact and unable to benefit from the spread guarantee.

Do not apply this filter to any spread-capture book. It is structurally incompatible.

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

This filter delivers +$4,164 P/L on $524,299 of capital (0.79% ROI) vs the baseline of +$4,120 on $1,359,559 (0.30%). In absolute P/L terms, it matches the entire unfiltered book on less than 40% of the capital deployed.

The mechanism is the 3.0x+ dominance signal described in Phase 3. When the wallet tilts more than 2x toward one side, that side wins at 88-96%. The dominant-side-only filter isolates those high-accuracy bets and discards the hedge legs. The resulting 91.6% win rate is a real and meaningful figure -- not a statistical artifact.

The catch: this filter is not a standalone strategy. The hedge legs that are excluded are necessary for the spread guarantee. If you run only the dominant-side fill, you are running a directional bet without a hedge -- and you need to pair it with something (either the full spread-capture framework or a different risk management layer). The filter cannot be applied in isolation without fundamentally restructuring the strategy.

As a capital-efficiency diagnostic, this is the most useful filter in the battery. It shows where the real edge is: high-dominance allocations, not the near-50/50 balanced markets.

3. Category filter (Crypto only) -- NO-OP

100% of trades are Crypto. The category filter is identity-equivalent to baseline. The output is unchanged: 89,678 trades, 47.9% WR, +$4,120 P/L, +0.30% ROI.

This is not a design failure of the filter battery -- it correctly reflects that the strategy has already applied the most aggressive possible category filter by operating in a single slug family. There is nothing to improve here.

4. Hour filter (exclude worst 4 hours: UTC 4, 16, 19, 20) -- MEANINGFUL_LIFT

Excluding hours 4, 16, 19, and 20 UTC from trading removes 15,382 trades (17% of the book) and produces the only filter that improves both absolute P/L and ROI simultaneously:

• Baseline: +$4,120 on $1,359,559 = 0.30% ROI
• Filtered: +$4,487 on $1,124,932 = 0.40% ROI
• Delta: +$367 additional P/L while deploying $234,627 less capital

The four excluded hours have the following individual P/L contributions: hour 4 (-$434), hour 20 (-$533), hour 11 (-$347), hour 19 (+$442). Wait -- hour 19 is actually positive (+$442). The filter is using the computed worst-hour list from the stats blob, which identifies 4, 16, 19, 20 as worst; this may be on a different metric than raw P/L. On raw P/L alone, the four worst are: hour 20 (-$533), hour 4 (-$434), hour 11 (-$347), hour 0 (-$71). Either way, the filter as computed produces genuine lift.

The underlying reason: BTC volatility and spread-capture opportunity are not uniform across the 24-hour clock. Hours 4 UTC (roughly midnight ET) and 20 UTC (4pm ET) may represent transition periods in global market participant activity where the CLOB becomes thinner, paired costs rise above $1.00 more frequently, and the directional signal degrades.

This is the single most actionable filter finding. Implement a scheduling layer that skips these hours with no other changes to the strategy.

5. Combined filter (dom 2x+ and excl worst hours) -- NO-OP (net)

The combination filter produces +$2,606 P/L on $589,184 capital = 0.44% ROI. ROI is slightly better than the hour-filter alone (0.40%), but absolute P/L drops to $2,606 -- well below both the baseline ($4,120) and the hour-filter alone ($4,487).

The combined filter is dominated by the dominance component's structural problem: it strips the hedge legs, leaving a directional-only book at reduced notional. The ROI improvement is real but the absolute P/L reduction makes it a net downgrade for an operator who cares about total dollars.

The stacked filter is not recommended. Run the hour filter alone; let the full book (including hedge legs) run during the approved hours.

What filters would add value if measurable

The standard battery misses the genuinely exploitable dimensions of this strategy. The following would require data beyond the trade CSV:

The weekend exclusion is fully computable from available data and is the most actionable refinement beyond the hour filter. Saturday contributes -$469 and Sunday contributes -$1,143 in the 12-day window. Turning the bot off on weekends would retain approximately +$5,733 of weekday P/L against only +$631 of weekend P/L currently being generated -- and those weekend dollars come at negative ROI (-0.20% Saturday, -0.63% Sunday). This is a strictly better outcome: less capital at risk, higher ROI, same absolute P/L from the profitable portion of the schedule.

Bottom line for replication

Three concrete recommendations:

1. Implement the hour-exclusion schedule. Skip UTC hours 4, 16, 19, 20. This adds $367 of P/L and reduces capital deployed, improving ROI from 0.30% to 0.40%. Straightforward to implement in any bot framework.

1. Consider weekend suspension. Saturday and Sunday together cost $1,612 in the observation window while providing only -0.4% average ROI. Suspending on weekends trades $631 of weekend P/L (negative returns) for reduced risk and simplified operations.

1. Do not apply any price-band filter. The paired cost of this strategy depends on fills across the full price range. The $0.30-$0.70 sweet-spot destroys $1,763 of P/L by removing the cheap-side hedge legs that are structural to the paired-entry mechanism.

Source wallet: 0xc387c2a40d389f17b723b6bba9b18b7dbd2de4f4 Strategy: Bilateral spread-capture market maker on 5-minute BTC Up/Down markets with asymmetric allocation signal Reference book: $1,359,559 BUY notional, +$4,121 net computed P/L, +0.30% ROI in 12 days

One-paragraph operator brief

Build a Polymarket bot that enters both sides of every available btc-updown-5m-* market, firing both legs within a 7-second window at whatever CLOB prices are available. Allocate more capital to the side with the lower mid-price (the underpriced side). Hold every position to settlement -- no early exits. Skip UTC hours 4, 16, 19, and 20. Consider suspending on weekends (Saturday and Sunday drain $1,612 vs weekday production of +$5,733 in the reference window). Cap per-market exposure at ~$220 total notional across both sides. The strategy's edge comes from two stacking mechanisms: locked spread on the ~50% of markets where the combined VWAP is below $1.00, and directional accuracy from the allocation skew (dominant side wins 95.8% of the time at 3.0x+ allocation asymmetry). Expect +0.30-0.40% ROI per 12-day cycle on the deployed notional, with higher yield on instantaneous working capital because positions cycle every 5 minutes.

1. Market selection

Why 5-minute only: The 5-minute series is the primary Polymarket BTC Up/Down product with the deepest CLOB and the tightest spread dynamics. The 15-minute and hourly series have fewer fills and wider variance, and the reference wallet shows no activity there. The edge in this strategy comes from frequency and volume, not per-trade size -- maximize the number of markets entered, not the per-market clip.

ETH exclusion is observed behavior, not a design choice. The reference wallet strictly avoids ETH despite ETH 5m having interesting characteristics in other wallets we have analyzed. The cause may be competition dynamics or model calibration; stick to BTC until the ETH book is separately validated.

2. Entry logic

def should_enter(market, clob):
    # Duration check
    if market.slug_prefix != "btc-updown-5m":
        return False
    
    # Window timing -- need at least 60s left
    sec_left = seconds_until_close(market)
    if sec_left < 60 or sec_left > 295:  # don't enter in first 5s either
        return False
    
    # Schedule check
    if utc_hour(now()) in (4, 16, 19, 20):
        return False
    
    # Optional: weekend suspension
    if day_of_week(now()) in ("Saturday", "Sunday"):
        return False  # toggle based on operator preference
    
    # Both sides must have quotes
    up_mid  = clob.mid_price(market.up_token)
    down_mid = clob.mid_price(market.down_token)
    if up_mid is None or down_mid is None:
        return False
    
    combined = up_mid + down_mid
    
    # Spread quality check (optional but recommended)
    # Enter always, but flag sub-1.00 markets for larger clips
    is_spread_positive = combined < 1.00
    
    return True, up_mid, down_mid, combined, is_spread_positive

Entry price discipline: The reference wallet uses no price anchor. It buys wherever the CLOB offers at the time the signal fires. The sub-bucket analysis shows fills spread across all 10 price bands with no single-cent concentration. Do not pin a specific price; lift the best available ask on both sides.

3. Allocation logic (the directional signal)

This is the most important piece of the strategy to implement correctly. The wallet does not split 50/50. It reads which side is underpriced and allocates more there.

def compute_allocation(up_mid, down_mid, base_clip_usdc):
    # The cheaper side has the higher upside -- allocate more there
    if up_mid < down_mid:
        # Up is cheaper (lower implied probability)
        # Standard formula: allocate by inverse of price (more to cheaper side)
        up_share = (1 - up_mid) / 1.0       # proportional to expected upside
        down_share = (1 - down_mid) / 1.0
        total_share = up_share + down_share
        
        up_allocation   = base_clip_usdc * (up_share / total_share)
        down_allocation = base_clip_usdc * (down_share / total_share)
    else:
        # Down is cheaper
        down_share = (1 - down_mid) / 1.0
        up_share   = (1 - up_mid) / 1.0
        total_share = up_share + down_share
        
        down_allocation = base_clip_usdc * (down_share / total_share)
        up_allocation   = base_clip_usdc * (up_share / total_share)
    
    # Compute resulting dominance ratio
    dominant   = max(up_allocation, down_allocation)
    subordinate = min(up_allocation, down_allocation)
    dom_ratio  = dominant / subordinate  # target 2.0-4.0x at full signal
    
    return up_allocation, down_allocation, dom_ratio

The allocation formula above weights by inverse price (or equivalently, by expected payout per dollar invested). This is an approximation of the reference wallet's actual logic; the exact signal is not observable from the trade data, but the resulting allocations consistently favor the lower-priced side.

Dominance ratio targets by market state:

4. Sizing model

The reference wallet operates with a median clip of $4.21 per fill and a maximum of $222.75. The per-market total exposure (across all fills on both sides) ranges from roughly $10 to $2,000, with the top 10 markets by volume ranging from $1,655 to $2,161.

Recommended sizing for replication:

The critical insight is that the working capital requirement is much less than the daily notional because positions cycle every 5 minutes. On a $10,000 bankroll with 220 markets per day at 5-minute cycles, you have roughly 12 concurrent open positions at any moment (averaging $350 total exposure each = $4,200 instantaneous peak exposure), well within the bankroll.

The strategy's natural per-wallet capacity ceiling is around $25,000-$50,000 based on the reference book's scale. Above that, your fills start to move prices in the 5-minute windows.

5. Both-sides allocation and pairing rules

The 7-second second-side lag is the operational target. The reference wallet achieves this through a concurrent order dispatch -- both legs are submitted nearly simultaneously and the confirmation lag accounts for the 7-second gap. A sequential-entry approach (first leg confirmed, then second leg dispatched) would introduce 15-30 seconds of delay on Polygon and risk price drift between legs.

Implement the two legs as parallel async calls with a shared nonce manager:

async def enter_market(market, up_usdc, down_usdc):
    async with asyncio.TaskGroup() as tg:
        up_task   = tg.create_task(submit_buy(market.up_token,   up_usdc))
        down_task = tg.create_task(submit_buy(market.down_token, down_usdc))
    up_fill, down_fill = up_task.result(), down_task.result()
    return up_fill, down_fill

6. Exit strategy

The exit strategy is: hold to settlement. There is no early exit.

The reference wallet has 89,678 BUY trades and 6 SELL trades. The 6 SELLs are effectively rounding errors. Every position is held until the market resolves and the winning side is paid $1.00/share.

Why no early exit: The strategy's P/L is maximized by holding to resolution because the spread is guaranteed at entry (for sub-$1.00 paired cost markets) and directional wins resolve at $1.00 (the maximum possible payout). Early selling would sacrifice the guaranteed terminal value. The strategy has no SELL engine because the edge is in the entry, not the exit.

This means the strategy requires a USDC reserve to fund ongoing entries even while waiting for prior positions to settle. At 5-minute resolution windows, capital recycles relatively quickly (losing positions settle in 5 minutes; winning positions in 5 minutes), but there is always some concurrent exposure.

7. Operational schedule

The operating schedule that maximizes risk-adjusted return: run UTC hours 0-3, 5-15, 17-18, 21-23 on Monday through Friday. That covers 20 active hours per day, 5 days per week. Consider running a reduced-size weekend test to validate whether the weekend drag is structural or specific to the observation window.

8. Operational requirements

9. Risk profile

Monitoring the competition signal is the most important ongoing risk management task. Track pct_sub_1.00 (the fraction of markets where your combined fill VWAP is below $1.00) weekly. The reference level is 50.1%. A sustained decline below 40% indicates competitors are filling the same side before you can lock the spread, and you should reassess clip sizes and entry timing.

10. Diagnostic checklist

Run weekly:

What this playbook deliberately does NOT include

• No price-band filter. The $0.30-$0.70 sweet-spot destroys $1,763 of P/L by removing cheap-side hedge legs that are structural to the paired-entry mechanism. Never apply this filter.
• No directional view on BTC. The strategy is not a BTC bull or bear position. Do not add a layer that says "skip Down markets when BTC is trending up." The allocation skew handles directionality automatically from live CLOB prices; adding an external view will bias the signal and degrade calibration.
• No stop-loss or early exit. The spread is locked at entry. Exiting early sacrifices the guaranteed terminal value on sub-$1.00 markets and the full payout on correct directional calls. The strategy's risk is bounded by clip size, not by a stop-loss trigger.
• No SELL engine. This is explicitly not the SirMartingale model. The two strategies differ completely in their exit mechanism. FlippingSharks derives zero value from active price discovery in the CLOB after entry -- its value is in the entry price, not the exit price.
• No ETH or other assets. The reference wallet is strictly BTC 5m. Until a separate validation run shows comparable economics on ETH 5m in this wallet's operating style, do not extend the universe.
• No leverage. The per-market returns are thin (approximately 0.04 cents per paired share on a $1.00 position). Adding leverage amplifies the tail losses (the super-$1.00 combined VWAP markets) without proportionally lifting the expected gains. The strategy's edge is frequency and volume, not per-market leverage.

The strategy is intentionally simple and mechanical. Its robustness comes from systematic coverage of every available market, not from clever market timing. Any "improvement" that reduces market coverage (price filters, directional skips, early exits) is almost certainly counterproductive at this scale. Build the bot to enter every window during approved hours and hold to resolution, and the aggregate statistics will do the rest.