080BEc

Wallet: 0x60889af507ec3f9da136f38b1c58080bec32f361 Window: 2026-04-08 to 2026-05-06 (29 calendar days, all 29 active) Universe: 142,621 trades across 7,549 unique markets (7,548 unique events) - $1,747,867 gross BUY notional Net P/L: +$25,483 on $1,747,867 deployed = +1.458% ROI

The Punchline

This wallet is a pure automated spread-capture market maker on Polymarket's BTC 5-minute Up/Down markets. The bot buys both sides of virtually every market it enters, holds to resolution, and collects the guaranteed spread embedded in the combined purchase cost falling below $1.00. There is no SELL engine, no exit management, no directional thesis. The 50.34% win rate on resolved BUYs is not the signal. It is a mechanical consequence of buying both sides: exactly half the capital resolves as a winner, which is precisely what a 50/50 binary market produces when you've bought both outcomes.

The strategy has a secondary layer that becomes visible in the dominance ratio analysis: in a meaningful subset of markets the bot tilts heavily to one side (dominance ratio above 3x), and those lopsided markets show a dominant-side win rate of 95.8%. This implies a directional signal of some kind is driving asymmetric allocation on selected markets, likely a spot-tape or momentum read similar to what pure directional traders run. But spread capture is the primary business. The directional tilt is a performance multiplier layered on top.

The economics are simple and mechanical. Median paired cost of $0.9907 means the bot locks in an average $0.0093 per unit of paired shares on every resolved both-sides market. Across $1.75 million of deployed capital, that structural spread generates the majority of the +$25,483 net P/L. The remaining P/L comes from the directional tilts in high-dominance markets, which are small in count but outsized in per-market return.

What He Trades

The universe is singular: BTC 5-minute Up/Down markets on Polymarket, slug pattern btc-updown-5m-*. Every single market in the top-10 by volume, every slug in the CSV sample, every entry in the best and worst P/L lists carries this slug. There is no ETH, no SOL, no 15-minute duration, no 1-hour duration, no sports, no politics. One asset, one duration.

btc-updown-5m-*   7,549 markets  $1,747,867 BUY notional  100% of book
All other slugs:  0 markets      $0 BUY notional

The 7 remaining trades (0.005%) outside Crypto are "Other" with $32.88 volume and -$12 P/L. Rounding error.

Within BTC 5-minute markets, the bot does not discriminate by time of day or day of week in any meaningful way. Trade volume runs from 3,801 fills (21:00 UTC) to 7,436 fills (13:00 UTC) across all 24 hours, with a coefficient of variation of roughly 23% across hours. This is essentially flat 24/7 coverage with a mild daytime bias, not the hard session-specific scheduling seen in directional traders.

Trade size distribution is deliberately near-uniform:

The top-5% share at 13.7% is the lowest concentration ratio in any trader we have profiled. For comparison, SirMartingale's top-5% share is 37%. This wallet deliberately caps fill sizes to keep per-market exposure bounded and variance low. There is no Kelly scaling, no conviction sizing, no large outlier fills.

The Order of Operations: One Market, Trade by Trade

The clearest both-sides illustration from the CSV is Bitcoin Up or Down, April 7, 8:00PM-8:05PM ET (slug btc-updown-5m-1775606400, resolved: Down).

Resolution: Down wins. All "Down" shares pay $1.00. All "Up" shares pay $0.00.

Walk-through:

1. 00:00:07 to 00:01:31. The bot opens the window buying both Up and Down in a near-symmetrical scatter. It buys Up at $0.47-$0.66 and Down at $0.44-$0.56. Combined cost on symmetric pairs is ~$0.97-$1.01. The bot is harvesting whatever spread exists in the initial orderbook.

2. 00:02:17 to 00:03:49. Midway through the 5-minute window something changes. The bot buys Up at extremely high prices ($0.79-$0.95) and Down at extremely low prices ($0.06-$0.19). This is the dominance tilt in action: the bot has concluded "Down" is the strong probability and it buys heavily asymmetric. It still buys both sides, but the allocation is 5:1 or more in favor of Down.

3. Outcome. Down resolves winner. The Down shares bought at $0.06-$0.56 pay $1.00 each. The Up shares bought at $0.47-$0.95 pay $0.00. The net result on the entire market depends on the ratio of Up to Down USDC deployed. In the late phase, the Down-heavy tilt plus resolution in Down's favor generates strong positive P/L for that market.

This is the full strategy in one market. Early-window symmetric spread capture, late-window directional tilt when the bot sees a signal, hold to resolution.

Why It Works: The Math

Spread capture base case (symmetric allocation):

Paired cost example:
  Buy Up  at $0.52, Buy Down at $0.48 → paired cost = $1.00 → spread = $0.00
  Buy Up  at $0.50, Buy Down at $0.47 → paired cost = $0.97 → spread = $0.03
  Buy Up  at $0.55, Buy Down at $0.45 → paired cost = $1.00 → spread = $0.00

Median observed paired cost: $0.9907
  → Expected spread per paired unit: $0.0093
  → On $1,747,867 gross notional at ~50% paired:
    Structural spread P/L ≈ $15,679 (matches pnl_decomp.spread_pnl = $15,679)

High-conviction directional tilt (2x+ dominance):
  33,398 fills in qualifying markets
  Win rate on dominant side: 88.8%
  At $626,162 deployed, +$122,455 P/L = +19.6% ROI
  (This is the filter result; the base book books less because the dominant
   allocation is partially offset by the losing non-dominant side)

The spread P/L of $15,679 is durable and structural. The directional tilt P/L is the source of performance variance. When the tilt signal works (as it did in weeks 1-3), aggregate P/L is strong. When the tilt signal degrades or competes against other market makers with the same edge (weeks 4-5), aggregate P/L collapses.

Phase 1: Trader Profile

Scale and activity: - 142,621 BUYs, 0 SELLs, 29 active days of 29 calendar days - $1,747,867 total BUY notional deployed - 7,549 unique markets, 7,548 unique events (one market per event) - 4,918 trades per day, 205 per hour, approximately 3.4 fills per minute

Execution signature: - Median inter-fill gap: 10.0 seconds - 45.7% of fills under 10 seconds apart - 92.2% under 60 seconds - 100% under 1 hour

The sub-10-second inter-fill gap at 45.7% confirms automated execution. The median of 10 seconds is not as fast as a pure latency-arb bot (SirMartingale shows a 0-second median) but is consistent with a market maker looping through available markets and posting fills at the top of the orderbook every 10-30 seconds. The fills at exactly the same second (visible in the CSV: e.g., 00:05:17 shows two fills simultaneously) confirm fan-out ordering where the bot submits multiple orders in a single block.

Buy vs. sell: 100% BUY, 0% SELL. Exceptional even among hold-to-settlement traders. Passive market maker that relies entirely on resolution payouts.

Active hours: All 24 hours have meaningful volume. No sleep window, no scheduling artifact. This is a genuinely 24/7 automated system.

Archetype: SPREAD CAPTURE / MARKET MAKER with secondary DIRECTIONAL TILT on high-dominance markets.

Phase 2: Core Strategy Identification

Both-sides participation rate: 99.0% (7,474 of 7,549 markets). This is unambiguous. No trader who runs a directional strategy accidentally buys both sides 99% of the time. The 0.99% of markets with only one side purchased are either execution failures, markets that expired before the second leg could be placed, or very early-window opportunistic buys.

Classification: Strategy A (Both-Sides Spread Capture / Market Making) with a strong secondary element of Strategy B (Directional Betting) visible exclusively in the high-dominance-ratio subset.

Not a directional bettor. Not a copy trader. Not a latency arb in the pure sense. The overall win rate of 50.34% is exactly what you get when you buy both sides of a binary market and the winning side pays $1.00.

Second-side lag median: 10 seconds. The bot enters both sides of a market within 10 seconds of each other on the median, confirming intentional pairing rather than opportunistic hedging.

Phase 3: Dominance Ratio Analysis

This is where the strategy's dual nature becomes quantitatively clear:

The dominant-side win rate progression from 61.9% to 95.8% is a real directional signal, not noise. If the bot were allocating randomly, every bucket would show approximately 50-55% dominant-side wins (since the dominant side has more capital exposed in a balanced binary market). The 95.8% win rate at 3.0x+ dominance means the bot is correctly identifying the winning side nearly 19 out of 20 times when it tilts heavily.

The mean paired cost dropping to $0.9736 in the 3.0x+ bucket is counterintuitive: it means the bot is spending disproportionately on the dominant side at prices close to $1.00 (near-certain outcomes) while the non-dominant side is bought at very low prices near $0.00. The extreme paired costs visible in the 3.0x+ list (e.g., 0.433, 0.55, 0.593, 0.591) confirm this pattern: dominant side at $0.95, non-dominant at $0.05, paired cost = $1.00 minus the spread on the non-dominant wing.

What this implies: The high-conviction signal likely fires when the market already reflects a strong directional probability, and the bot positions asymmetrically to capture the remaining spread on the near-certain side while taking a minimal hedge on the losing side for optionality.

Phase 4: Entry Price Analysis

The win-rate column is a near-perfect probability calibration curve: 6.1% wins at $0.00-$0.10 entry (correct for ~5% implied probability), 94.7% wins at $0.90-$1.00 (correct for ~95%). This confirms the market is efficiently pricing these markets and the bot is buying at prices that reflect real probabilities.

The ROI column is the more interesting signal. The highest ROI is in the $0.20-$0.30 band (+4.9%) and the $0.40-$0.50 band (+3.7%). These are the near-midpoint bands where spread capture is most available (paired cost cheapest to achieve below $1.00). The $0.60-$0.80 bands show near-zero ROI despite correct win rates, because buying near-favorites at $0.70-$0.80 and pairing with the near-underdog at $0.20-$0.30 still produces thin spreads.

Sub-bucket concentration: Unlike tick-specific bots (floor bidders, or near-certainty snipers), this wallet's entry prices are genuinely distributed across all 101 cents. No single price point dominates. This is consistent with a bot that lifts whatever the orderbook offers on both sides without anchoring to a specific price level.

The two loss bands ($0.00-$0.10 and $0.10-$0.20) represent the non-dominant hedge legs in high-conviction markets: the bot buys "Down" at $0.06 when it thinks "Up" is near-certain, those cheap Down shares almost never win, producing a structural drag. This drag is the cost of maintaining the both-sides framework.

Phase 5: Category and Market-Type Breakdown

There is no useful cross-category analysis. The wallet is 99.997% BTC 5-minute Up/Down by notional. The Crypto ROI of +1.46% is the whole story.

Phase 6: Timing and Execution Analysis

Hourly distribution (UTC):

Every single hour is profitable. The "worst" hours still print positive P/L. The spread-capture mechanism is so consistent that there is no genuinely bad hour. The variation in hourly P/L (from +$327 at 05:00 UTC to +$1,980 at 21:00 UTC) is more a function of trade volume than per-trade edge.

Win rates by hour: Tightly clustered between 50.0% and 50.7% across all 24 hours. No systematic hour-of-day directional edge. The slight variance around 50.3% baseline is statistical noise.

Day-of-week P/L:

Saturday has the highest ROI (+2.59%) and the second-highest win rate (50.8%). Weekends have lower competition from other market makers, which translates to wider available spreads and better paired costs. Thursday is the best absolute-P/L weekday at +$5,684 on 23,758 trades. Tuesday is the weakest day across both ROI and absolute P/L.

Burst patterns: The CSV shows repeated multi-fill same-second bursts. In the April 7, 8:00PM-8:05PM market, multiple fills land at 00:00:07, 00:00:31, 00:01:07, 00:01:31, 00:02:55, 00:03:17 UTC. Same-second multi-leg ordering is the entry signature of an automated system routing multiple orders through the CLOB simultaneously.

Accumulation pattern per market: The bot typically places 10-40 fills per 5-minute market window, spread across the full 5 minutes, with fill cadence of every 15-60 seconds. It does not rush to establish a position in the first seconds; it builds gradually across the window. This is different from SirMartingale's entry-burst-then-exit pattern.

Phase 7: Filter Experiments

The finding that matters most: The high-conviction filter is explosively productive. Restricting to markets where dominance ratio is 2x or higher, dominant leg only, returns +$122,455 at +19.6% ROI on $626,162 of capital. That is 4.8x the ROI of the base strategy with 64% less capital deployed. This is not a noise result.

The price filter ($0.30-$0.70) is mildly helpful (+1.82% vs +1.46% ROI) but destroys absolute dollars (-$6,932). It removes the very-cheap hedge-leg buys (which are structurally slightly negative) but also removes some of the strong near-certainty dominant fills.

The hour filter does nothing meaningful. Every hour is positive and no hour is far enough below baseline to justify exclusion.

The combined filter improves ROI to +2.03% at the cost of halving the capital base and reducing absolute P/L by $8,700. For a capacity-constrained replicator who cannot deploy $1.75 million per month, this focuses the book on the most efficient markets.

Phase 8: Rolling Window Consistency

The deterioration at the end of the window is real and significant. The rolling 7-day P/L peaked at +$13,096 on April 24, then degraded monotonically: April 27 (+$9,409), April 30 (+$4,466), May 1 (+$1,247), May 3 (+$152), May 4 (-$1,344), May 5 (-$2,877), May 6 (-$3,276).

Week-by-week summary:

The strategy peaked in Week 17 and has been in controlled retreat since. The most likely explanation is competitive spread compression: other market makers entering the BTC 5-minute market reduced the frequency of sub-$1.00 paired cost opportunities, squeezing the structural spread from ~$0.009 toward zero. The directional tilt signal may also have had a rough patch in weeks 4-5.

Phase 9: P/L Decomposition

The hedge tax of $672,224 is the cost of the both-sides discipline. Of the $1,747,867 deployed, roughly $672K went to losing outcomes. This is not "wasteful" in the framework's terms because the spread capture on the winning leg more than compensates. The net is +$25,483.

The structural spread P/L of $15,679 represents the base, guaranteed component. The remaining $9,804 comes from the directional tilt signal performing on net. In bad weeks, the tilt can subtract from the spread P/L, explaining the negative rolling windows in the final days.

Why no SELL engine: The spread capture model works because the bot's edge is locked in at purchase time via the combined cost being below $1.00. There is no need to manage exit. The SELL leg would require tracking individual lots, posting asks, and competing for exit liquidity, all for no additional benefit since the market will resolve to $1.00 or $0.00 in at most 5 minutes. The hold-to-resolution approach is structurally correct for this strategy.

Phase 10: Strategy Specification

One-sentence summary: An automated 24/7 market maker that buys both sides of every btc-updown-5m-* market at a combined cost below $1.00, holds to resolution for guaranteed spread capture, and applies an asymmetric directional tilt on a subset of markets where a secondary signal fires.

Edge sources (two stacked):

1. Structural spread capture: paired cost of $0.9907 median locks in $0.0093 per unit of paired shares, guaranteed regardless of direction.
2. Directional tilt alpha: in 3,483 markets at 2x+ dominance, the dominant side wins 84-96% of the time, generating disproportionate P/L above the spread baseline.

What works: BTC 5-minute markets exclusively. Symmetric both-sides allocation as base. Asymmetric allocation when directional signal fires (dominance ratio 2x-10x). Small bounded clips ($5-$20 typical). 24/7 coverage. Hold to resolution.

What drags: Hedge-leg buys at very low prices ($0.01-$0.10) that almost never win, representing a structural drag of $1,921 in the $0.00-$0.10 bucket alone. Late-window competitive spread compression visible in weeks 4-5. Tuesday underperformance.

Rebuild parameters: See the Replication Playbook for the full implementable spec.

Wallet: 0x60889af507ec3f9da136f38b1c58080bec32f361 Window: 2026-04-08 → 2026-05-06 (29 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: 99.01% (7,474 of 7,549 markets)
• Median paired cost: $0.9907
• Mean paired cost: $0.9880
• Paired cost % under $1.00: 58.8%
• Paired cost % under $0.97: 33.5%
• Median 2nd-side hedge lag: 10s

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: 26 of 29 (89.7%)
• Rolling 7-day P/L range: -$3,276 → +$13,096
• Rolling 15-day windows green: 29 of 29 (100.0%)
• Rolling 15-day P/L range: +$1,638 → +$23,141

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 2026-05-08 10:32 UTC.

Wallet: 0x60889af507ec3f9da136f38b1c58080bec32f361 Window: 2026-04-08 to 2026-05-06 Baseline: 142,615 resolved BUYs · 50.34% WR · $1,747,843 deployed · +$25,483 P/L · +1.46% ROI

The headline result

One filter is transformative. Two are mildly useful. Three are structural no-ops.

The transformative one is the high-conviction dominance filter: restricting to markets with 2x+ dominance ratio, dominant leg only, multiplies ROI from +1.46% to +19.6% and extracts +$122,455 of P/L from $626,162 of capital. That is 4.8x the base strategy's ROI and represents the directional alpha buried inside the spread-capture book.

The price filter is mildly useful for improving ROI per dollar (+1.82% vs +1.46%) but destroys absolute P/L by cutting below-$0.30 and above-$0.70 entries, which include many of the dominant-side fills in high-conviction markets. The hour filter is functionally a no-op; no hour loses money.

The category filter is identity-equivalent to baseline (99.997% Crypto). The combined filter improves ROI to +2.03% but at the cost of halving deployed capital and losing $8,700 of absolute P/L.

Filter results table

Filter-by-filter commentary

1. Price band filter ($0.30-$0.70) - MILD LIFT ON ROI, DESTRUCTIVE ON ABSOLUTE P/L

Applying the standard $0.30-$0.70 sweet-spot filter improves ROI slightly from +1.46% to +1.82% but cuts absolute P/L from +$25,483 to +$18,551, a loss of $6,932.

The mechanism: the filter removes the very cheap non-dominant hedge legs (the $0.05-$0.15 buys on the losing side of high-conviction markets), which carry negative ROI as standalone positions. But it also removes many high-dominance dominant-side buys that land above $0.70, specifically the near-certainty buys in the 3x+ dominance bucket at $0.80-$0.95. Those dominant buys are the highest-expected-value fills in the book.

Verdict: Do not apply this filter to a spread-capture market maker without first stratifying by dominance ratio. Applied blindly it sacrifices the best directional fills along with the marginal hedge legs. If you must use a price filter, apply it only to the symmetric (1x-1.5x dominance) subset of markets, not to high-conviction tilted markets.

The win rate improvement from 50.34% to 51.26% under this filter confirms the eliminated trades were below-50% wins, consistent with removing the non-dominant hedge legs.

2. High-conviction filter (dominance 2x+, dominant leg only) - MEANINGFUL LIFT

This is the most important filter result in the entire analysis. Keeping only the dominant-side fill in markets where the bot allocated at least 2x more capital to one side than the other produces +$122,455 P/L at +19.56% ROI on $626,162 of capital.

The mechanism is straightforward: the bot's secondary directional signal has genuine predictive value. When it fires strongly enough to produce a 2x dominance ratio, the dominant side wins 88.8% of the time. At 3x+ dominance, it wins 95.8% of the time. These are extraordinary win rates for a binary market where the prior is 50/50.

What the filter discards is the hedge leg, the non-dominant buys that almost never win in high-conviction markets. These hedge legs are a structural cost of the both-sides framework. An operator who could identify the high-dominance markets in advance and skip the non-dominant fills would capture the directional alpha without paying the hedge tax.

Practical implication: If you're replicating this strategy, the high-conviction signal is where you want to concentrate. The question is how the bot identifies which markets warrant 2x+ asymmetric allocation. The signal is not disclosed in the trade data but the result is measurable. Market price movement in the final seconds of a 5-minute window, divergence from a spot-tape model, or order flow imbalance on the CLOB are the likely candidates.

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

The category filter is structurally identity-equivalent to baseline. The wallet is 99.997% Crypto by notional ($1,747,831 of $1,747,843). The $12 difference is rounding from the 7 "Other" resolved trades. This filter adds no information and should not be applied.

4. Hour exclusion filter (remove hours 1, 13, 15, 22) - WEAKLY DESTRUCTIVE

The filter removes the four lowest-P/L hourly buckets as identified by the data (UTC hours 01, 13, 15, 22). The result: 116,193 trades, $1,415,598 capital, +$21,903 P/L, +1.55% ROI.

ROI improves from +1.46% to +1.55% but absolute P/L falls from +$25,483 to +$21,903, a loss of $3,580. The "worst" hours are still profitable. Hour 01 prints +$1,003, hour 13 prints +$928, hour 15 prints +$823, hour 22 prints +$825. There is no genuinely bad hour in this strategy. The spread capture mechanism works around the clock.

Unlike a directional trader with a hard sleep window (SirMartingale), this bot has no hour that structurally bleeds. The slight variation in hourly ROI is noise around a stable 1.4-1.6% baseline. Applying an hour filter to this strategy reduces capacity without meaningfully improving edge per dollar. Not recommended.

5. Combined filter (price $0.30-$0.70 + exclude worst 4 hours) - NO-OP NET, SLIGHT ROI IMPROVEMENT

67,460 trades, $826,892 capital, +$16,783 P/L, +2.03% ROI. Absolute P/L is -$8,700 vs baseline. ROI improves by 57 basis points.

The combined filter helps a capital-constrained operator run a higher-ROI book on less capital. For a replicator with $50,000 of bankroll rather than $1.75 million in turnover capacity, this focused book may be preferable. But for an operator who can deploy the full capacity, the combined filter is net-negative in dollar terms.

What filters would genuinely add value

The standard filter battery is not well-suited to this trader. The genuinely useful refinements require data points outside the trade CSV:

Bottom line for replication

The most actionable filter insight from this analysis:

1. Run the high-conviction dominance filter. Markets where your bot allocates 2x+ on one side should be tracked separately. The dominant-leg-only P/L in those markets (+19.6% ROI) is where the real alpha lives. Consider increasing dominant-side size and reducing or eliminating the non-dominant hedge leg in those markets.

2. Do not apply the price filter to the whole book. It removes high-value dominant fills above $0.70 along with the hedge legs below $0.30. If you use a price filter, apply it only to the symmetric (1x-1.5x dominance) markets.

3. Do not exclude hours. Every hour is profitable. The hour filter reduces capacity without improving per-dollar edge.

4. Monitor the rolling 7-day P/L weekly. The spread-capture base is too thin to sustain extended competitive pressure. Three consecutive negative rolling windows is a pause-and-audit trigger.

Source wallet: 0x60889af507ec3f9da136f38b1c58080bec32f361 Strategy: Automated both-sides spread-capture market maker on BTC 5-minute Up/Down markets with high-conviction directional tilt Reference book: $1,747,867 BUY notional deployed over 29 days · +$25,483 net P/L · +1.46% ROI High-conviction subset: $626,162 deployed · +$122,455 P/L · +19.56% ROI

One-paragraph operator brief

Build a Polymarket bot that opens every fresh btc-updown-5m-* market, immediately buys both "Up" and "Down" at roughly equal USDC clips targeting a combined (paired) cost below $1.00, and holds both legs to resolution. This is pure spread capture: one side always pays $1.00 per share, the other pays $0.00, and if the combined cost is below $1.00 the bot made money regardless of direction. Layered on top: when a secondary signal fires (BTC spot divergence from CLOB mid, momentum, or order flow imbalance), allocate 2x-5x more USDC to the favored side. Track dominance ratio; at 2x+ the historical win rate is 88.8% and at 3x+ it is 95.8%. Keep fills small ($5-$20 each), run 24/7, hold everything to resolution, never SELL. Expect ~$850/day in spread P/L at the reference scale of $60K/day deployed; the directional tilt adds variance and, when working, disproportionate upside.

1. Market selection

Prioritization: Enter every qualifying market. There is no selectivity among markets. The strategy is high-volume, low-margin. Volume is essential. Selectivity reduces capacity faster than it improves per-trade edge.

Capacity and competition: The reference book deploys ~$60,000 per day across ~7,549 markets per 29-day window (roughly 260 markets per day, or ~26 concurrent at any moment given the 5-minute intervals). Each BTC 5-minute market expires and a new one opens every 5 minutes, 288 markets per day. The bot should be entering approximately 90% of all available markets.

2. Entry logic

def should_enter(market, clob_up_mid, clob_down_mid):
    """Base spread-capture gate."""
    # Eligibility: active window, not in final 30s
    sec_elapsed = now() - market.open_time
    sec_left    = market.close_time - now()
    if sec_elapsed < 15 or sec_left < 30:
        return False

    # Slug whitelist: BTC 5-minute only
    if not market.slug.startswith("btc-updown-5m-"):
        return False

    # Paired cost gate: only enter when spread is positive
    paired_cost = clob_up_mid + clob_down_mid
    if paired_cost >= 1.00:
        return False  # No guaranteed spread available

    return True

def compute_allocation(market, clob_up_mid, clob_down_mid,
                        spot_fair_prob_up, base_clip_usdc):
    """
    Returns (up_usdc, down_usdc) to buy.
    Applies symmetric base allocation plus directional tilt if signal fires.
    """
    up_usdc   = base_clip_usdc
    down_usdc = base_clip_usdc

    # Compute fair-value gap
    gap = spot_fair_prob_up - clob_up_mid

    # Tilt logic: if spot diverges from CLOB by > 5%, increase favored side
    if abs(gap) >= 0.05:
        if gap > 0:  # Up is underpriced vs fair value
            up_usdc   = base_clip_usdc * tilt_multiplier(gap)
            down_usdc = base_clip_usdc  # Keep hedge at base size
        else:        # Down is underpriced vs fair value
            down_usdc = base_clip_usdc * tilt_multiplier(abs(gap))
            up_usdc   = base_clip_usdc  # Keep hedge at base size

    return up_usdc, down_usdc

def tilt_multiplier(gap):
    """Scale from 1x (no tilt) to 5x (strong signal)."""
    if gap < 0.05:  return 1.0
    if gap < 0.10:  return 1.5
    if gap < 0.15:  return 2.5
    if gap < 0.20:  return 4.0
    return 5.0

Multi-leg fan-out: The reference bot fires 10-40 fills per 5-minute window, spaced roughly every 15-60 seconds. This reduces market impact and allows the bot to accumulate position across the window as the orderbook updates. Do not place a single large order; walk the book with repeated small clips.

3. Exit logic (hold to resolution)

There is no exit logic. Every position is held to resolution. The BTC 5-minute window closes in at most 300 seconds, at which point one outcome pays $1.00 per share and the other pays $0.00. The bot collects its resolution payouts automatically through the Polymarket CLOB settlement mechanism.

def manage_position(position):
    """No active management. Hold to resolution."""
    # Do not post SELL orders
    # Do not cancel BUY orders mid-window
    # Wait for market.resolved == True
    # Collect settlement payout via API
    pass

Why no SELL engine: Unlike directional traders who need to convert unrealized to realized before markets reverse, this bot's edge is locked in at entry time. If paired cost is $0.97, the expected value of $0.03 per unit is guaranteed by the binary resolution mechanic. Selling early would require posting asks, competing for fill, and potentially legging out of a paired position and creating directional exposure. The hold-to-settlement approach is structurally optimal for this strategy.

4. Sizing model

The reference book uses near-uniform sizing across all fills:

Tilt sizing for high-conviction markets: When the directional signal fires at 2x+ tilt, increase the favored-side clip to 2x-5x the base clip while keeping the hedge-side clip at base. At the reference scale with a $10 base clip, a 4x tilt means $40 on the favored side and $10 on the hedge side. This matches the observed 3x+ dominance bucket behavior.

Sizing discipline: Do NOT Kelly-size. Do NOT scale by expected spread. The spreads are too thin and too similar across markets for conviction-based sizing to add value on the symmetric spread-capture component. Uniform small clips minimize variance and prevent any single market loss from mattering.

Natural capacity ceiling: At approximately $100K of working capital per wallet, the bot begins to move CLOB prices on its own fills, degrading the paired cost it achieves. Above that threshold, fragment into 2-3 wallets or accept that per-dollar ROI will compress.

5. Both-sides allocation

Both sides of every market are purchased. This is non-negotiable for the base strategy. The both-sides requirement:

1. Guarantees positive P/L when paired cost < $1.00, regardless of direction.
2. Eliminates directional risk from the spread-capture component.
3. Maintains a natural hedge against the directional tilt being wrong.

Allocation ratio guide:

At 5x tilt, the hedge leg is 17% of total capital deployed in the market. This matches the observed 3x+ dominance markets where the non-dominant side is bought at very low prices (0.05-0.15) as a small hedge against the primary directional position.

Paired cost discipline: Before entering any market, confirm that the target allocation produces a paired cost below $1.00. In tilted markets where you're buying the favored side at $0.85 and the hedge at $0.15, paired cost = $1.00 exactly. In those cases you are no longer capturing spread; you are making a directional bet with a cheap hedge. Be explicit about which mode you're in.

6. Scheduling and coverage

Recommended schedule: 24/7 continuous coverage.

Unlike directional traders with session-specific edges, this strategy has no structural hour-of-day edge variation. Every UTC hour is profitable in the reference data. Halting the bot overnight wastes capacity and reduces the total spread capture.

BTC 5-minute market schedule: Polymarket opens a new btc-updown-5m-* market every 5 minutes, 24/7. That is 288 markets per day. The reference book touches approximately 260 per day (90% coverage). Full coverage is achievable with a well-tuned bot; the 10% miss rate in the reference data may be latency or fee-related passes.

7. Operational requirements

8. Risk profile

The primary risk is competitive spread compression. The reference data already shows this: the rolling 7-day P/L degraded from a peak of +$13,096 (week ending April 24) to -$3,276 (week ending May 6). The likely cause is other market makers entering the BTC 5-minute space and bidding up the paired cost toward $1.00. When the paired cost median rises above $0.995 for more than 2-3 days, the spread-capture base becomes economically marginal and the strategy's P/L becomes entirely dependent on the tilt signal performing.

9. Diagnostic checklist: is the bot still working?

Run weekly:

10. What this playbook deliberately does not include

No SELL orders. The spread-capture model works because the edge is locked in at entry. Adding a SELL engine would introduce exit slippage, create the possibility of legging out of a paired position, and add operational complexity for zero structural benefit. Every position is held to the 5-minute resolution.

No market selection within BTC 5-minute. Do not add "skip markets where realized vol is low" or "prefer markets with higher VIX" or any other market selection filter within the BTC 5-minute universe. The strategy's edge is volume-dependent and any selectivity reduces capacity faster than it improves per-trade ROI. The exception: skip markets where paired cost is at or above $1.00, which is already hardcoded in the entry gate.

No other asset classes or durations. The reference book had one each of Soccer, Politics, and several "Other" trades. They all lost money or broke even. BTC 5-minute is the only slug. ETH 5-minute and BTC 15-minute are not included in this trader's book and their liquidity profiles and spread characteristics are different enough that you should not assume they carry the same edge without fresh analysis.

No position re-entry or averaging down. Once the bot has placed its fills in a 5-minute window, it does not go back and add more to a position because the price moved favorably. Each market window is a single allocation decision. Adding within a window increases position size beyond the planned cap and introduces timing risk.

No manual override of the dominance tilt. The tilt signal must be quantitative and rules-based. Manual judgments about whether Bitcoin "looks bullish" are not part of this strategy and historically underperform systematic signals on short time horizons.

No attempt to recover from spread compression by widening clips. If the weekly diagnostic shows paired cost rising toward $1.00, the correct response is to reduce sizing and wait for competition to normalize, not to increase clip sizes hoping to "make it up on volume." The spread is what it is; you cannot manufacture a positive spread by deploying more capital.

TL;DR - implementable in ~100 lines of Python

async def run_bot(wallet, usdc_balance):
    clob_feed  = await connect_polymarket_clob_ws()
    spot_feed  = await connect_spot_websocket()
    base_clip  = min(10.0, usdc_balance * 0.0002)  # ~0.02% per fill

    async for market_event in clob_feed:
        market = market_event.market
        if not market.slug.startswith("btc-updown-5m-"):
            continue

        sec_left = (market.close_time - now()).seconds
        if sec_left < 30 or (300 - sec_left) < 15:
            continue

        up_mid   = market.up_side.mid_price
        down_mid = market.down_side.mid_price

        # Gate 1: paired cost must be below $1.00
        if up_mid + down_mid >= 1.00:
            continue

        # Gate 2: skip if spread is microscopic
        if (1.00 - up_mid - down_mid) < 0.005:
            continue

        # Compute directional tilt
        spot_btc    = spot_feed.latest_btc_mid()
        fair_prob_up = compute_fair_prob(market, spot_btc)
        gap          = fair_prob_up - up_mid

        up_clip   = base_clip
        down_clip = base_clip

        if abs(gap) >= 0.05:
            multiplier = tilt_multiplier(abs(gap))
            if gap > 0:    up_clip   = base_clip * multiplier
            else:          down_clip = base_clip * multiplier

        # Submit both legs
        await buy(market, "Up", usdc=up_clip)
        await buy(market, "Down", usdc=down_clip)
        # Holds to resolution automatically - no sell logic

    # Reconcile after each resolved market
    # Collect settlement payouts via /positions endpoint

Expect ~$850/day in base spread P/L at a $60K/day deployment rate. The directional tilt adds meaningful upside when the signal is fresh and the competition is thin. Monitor paired cost weekly. Pause when the rolling 7-day window goes negative for three consecutive periods.