ohanism

Wallet: 0x89b5cdaaa4866c1e738406712012a630b4078beb Window: 2026-04-09 to 2026-05-05 (27 calendar days, 27 active) Universe: 694,439 trades across 37,982 markets | $8.57M BUY notional | $1.13M SELL notional | $9.70M gross turnover Net cash-flow P/L: +$258,687 on $8.57M deployed = +3.02% ROI in 27 days

The Punchline

This is the highest-volume wallet in the PR&R dataset by trade count. 694,439 fills in 27 days, or roughly 21,800 per day, across 37,982 unique markets spanning BTC, ETH, SOL, and XRP in 5-minute and 15-minute Up/Down windows. The gross BUY notional of $8.57 million is not a typo.

The 3.02% ROI sounds small but is structurally sound: +$258,687 net realized over 27 days on a perpetually cycling working capital of roughly $300-$400K. The capital cycles fast (each 5-15 minute market closes before the next opens), so the annualized return on working capital is multiples of the nominal figure.

The strategy is a hybrid spread-capture plus directional market maker. On 33% of markets it buys both outcomes; on the other 67% it takes one-sided directional bets. The both-sides markets have a median paired cost of $1.059, which is above break-even but with 40% of those pairs locking in sub-$1.00 guaranteed spread. The directional markets show a strongly calibrated accuracy curve, with the 3x+ dominance bucket hitting 77.7% dominant-side wins across 8,653 markets. That win rate is the load-bearing structural fact of this wallet.

The bot trades 24/7. There is no sleep window. Every UTC hour shows positive trade volume ranging from 26,222 trades (hour 21) to 34,373 trades (hour 10). Win rates by hour range from 53.9% (hour 13) to 71.1% (hour 23). The operation is fully automated and always on.

What He Trades

The universe collapses to one category:

Within Crypto, all four major slug patterns appear: btc-updown-5m-*, btc-updown-15m-*, eth-updown-5m-*, eth-updown-15m-*, sol-updown-5m-*, sol-updown-15m-*, xrp-updown-5m-*, xrp-updown-15m-*. The CSV also shows xrp-up-or-down-may-5-2026-7pm-et (a longer-duration XRP market), confirming the bot also touches hourly and non-standard-window markets opportunistically.

The asset breadth distinguishes this wallet from single-asset operators. The bot operates simultaneously on multiple assets within the same timestamp. At 23:57:40 UTC on May 5, it fires BUY orders into btc-updown-15m, btc-updown-5m, and eth-updown-5m within the same two-second burst. The concurrency is systematic, not coincidental.

Both-sides participation: 33.3% of markets had both outcomes purchased. This is the structural feature that separates this wallet from SirMartingale (0% both-sides) and pure market makers (60%+ both-sides). The bot is a hybrid: it hedges when uncertain and goes one-sided when confident.

The Order of Operations: One Market, Trade by Trade

The most illustrative example of the both-sides paired strategy in the CSV is the btc-updown-5m-1775693100 market (Bitcoin Up or Down, April 8, 8:05PM-8:10PM ET), which resolved "Up." This market shows the bot's full behavioral signature.

Walk-through: The bot opens with a Down position at $0.42-$0.48, spending about $19.55. The market then signals Up to the bot (perhaps via spot tape). It adds a large Up position at $0.80, pairing the market. Almost simultaneously it sells its Down shares at $0.20-$0.25, taking a ~$7 loss on that leg. It then continues loading Up at $0.72-$0.89 as the BTC tape confirms the move. The market resolves Up. Net result: the Down hedge cost ~$7 in losses, and the Up position won with all shares paying $1.00. The hedge was the insurance policy, not the core bet.

This walk-through shows the bot's logic: enter both sides when uncertain, sell the losing leg when the signal clarifies, hold the winning leg through resolution.

Why It Works: The Math

The strategy has three stacked positive-EV sources:

Source 1: Sub-$1.00 paired spread (43.6% of paired markets)

For a paired market with cost = $0.92 (22nd percentile of paired cost distribution):
  Both sides pay $1.00 on the winner.
  Locked profit per share pair = $1.00 - $0.92 = $0.08
  On 50 paired shares: guaranteed +$4.00 regardless of outcome.
  Scale across 12,635 markets, 43.6% sub-$1.00:
  = 5,509 markets with guaranteed profit.

Source 2: Directional accuracy at high dominance

3x+ dominance bucket: 8,653 markets, 77.7% dom-side win rate
  Mean paired cost: $1.030 (slightly above break-even)
  Dom-side win rate 77.7% vs 22.3% lose rate

  For every 100 markets at avg dominance 5x, avg cost $1.03:
    77.7 wins: dom-side collects $1.00/share; non-dom loses $0
    22.3 losses: dom-side loses; non-dom collects but is smaller
    Net directional edge per market: positive at 77.7% win rate
    vs 50% break-even on a purely random coin flip.

Source 3: Active SELL exits (15.2% of trades are SELLs)

105,866 SELL trades collecting $1,132,065 in SELL proceeds.
SELLs fire before resolution on winning positions to lock gains.
The CSV shows SELLs at $0.20-$0.35 on the losing leg (cutting losses),
and SELLs at $0.50+ on the winning leg (locking gains early).

The P/L decomposition confirms: the realized total of +$258,687 is achieved despite a structural hedge tax of -$980,261 (the capital lost on the non-dominant legs across all paired markets where the dominant side won). The gross bet on all winning outcomes must exceed $1.24M to produce the net positive result after absorbing that hedge tax. It does.

Phase 1: Trader Profile

Scale and Activity

Trade Size Distribution

The max of $281.93 is only 52× the median. This is a bounded, high-repetition sizing model. The top-5% share of capital at 32.2% is moderate. The size distribution is not power-law; it is a fairly flat curve with a hard ceiling around $280. No single trade can blow up the book.

Execution Signature

The 0.0-second median and 81.8% sub-10s share confirm a high-speed bot. The CSV shows batches of 5-10 fills at identical timestamps (e.g., nine trades all at 00:05:09 UTC on April 9). Same-second multi-leg fan-outs are the entry signature. The 15.1-second mean is elevated because the SELL exits are time-delayed relative to the BUY entries.

Trading Hours (UTC)

The bot is active in every single UTC hour. The histogram ranges from 26,222 trades (hour 21) to 34,373 trades (hour 10). There is no zero-trade hour, no sleep window, no weekend break. This is a 24/7 automated system.

The hour-7 loss (-$2,242) is the only negative P/L hour. Hours 4, 7, 13, and 21 are identified as the worst by the filter analysis.

Archetype: HYBRID SPREAD + DIRECTIONAL BOT

Phase 2: Core Strategy Identification

Both-sides participation rate: 33.3%

12,635 of 37,982 markets had both outcomes purchased. This places the wallet in the hybrid zone: too few pairs to be a pure market maker, too many to be a pure directional bettor.

Classification:

The wallet is A + B: spread capture on paired markets (A) combined with directional betting on one-sided markets (B). The two modes are distinguished by the dominance ratio of the paired markets.

Evidence for Spread Capture (A): - 12,635 both-sides markets - Median paired cost $1.059 (6% above break-even) - 43.6% of paired markets achieve sub-$1.00 paired cost - 40.2% achieve sub-$0.97 paired cost (locking at least 3 cents guaranteed) - Second-side lag median of 140 seconds (intentional pairing, not accidental)

Evidence for Directional Betting (B): - 25,347 one-sided markets (66.7% of universe) - 64.4% overall win rate on resolved BUYs - Price-band ROI shows positive returns across all 10 bands - 77.7% dominant-side win rate at 3x+ dominance

Evidence against pure market making: - Median paired cost $1.059 is not consistently below $1.00 (a pure MM would require this) - 56.4% of paired markets are above $1.00 paired cost (these markets lose money on the spread and must win directionally)

The bot is not a static MM. It is a dynamic paired-entry directional system where the pairing is both a hedge and a spread-capture tool.

Phase 3: Dominance Ratio Analysis

This is the most important quantitative section for understanding the directional signal.

The jump from 64.3% at 2-3x to 77.7% at 3x+ is the critical finding. The bot's high-conviction tier (where it allocates 3+ times more to one side than the other) wins 77.7% of the time. This is not noise. Across 8,653 markets, the standard error on a 77.7% estimate is about 0.45 percentage points. The true win rate is between 76.8% and 78.6% at 95% confidence.

The 1.0-1.5x bucket at 53.9% is near random. These are the markets where the bot is essentially pairing for spread, not expressing directional conviction. The paired cost there ($1.089) means the guaranteed spread is negative, so these markets only profit if the dominant side wins, which at 53.9% is barely better than chance.

The optimal filter implied by this data: focus replication on the 3x+ dominance markets. The 8,653-market sample with 77.7% win rate is the alpha concentrate. See Phase 7 for the filter result: the high-conviction filter produces +$187,602 P/L at +6.43% ROI vs the 3.02% baseline.

Phase 4: Entry Price Analysis

The calibration is near-perfect. The win rate column tracks the implied probability almost exactly across all 10 bands. This is the signature of a well-calibrated market: the bot is not finding systematic mispricings in any single price band.

The ROI inversion is the key structural insight. The highest ROI (53.4%) is at the lowest price band ($0.00-$0.10), and ROI declines monotonically to 1.1% at the $0.90-$1.00 band. But capital allocation is the inverse: 45% of all capital is deployed in the $0.70-$1.00 zone with the lowest ROI.

Why does the bot deploy so much capital at low ROI? Two reasons: 1. The high-price bands are the dominant leg of paired positions. When the bot allocates 5x to "Up" vs "Down," the large Up position is at a high implied probability. 2. The low-price bands ($0.00-$0.20) are the small hedge legs. They carry high ROI because when they unexpectedly win, the payout is large relative to cost, but they rarely win (7.5% and 16.7% win rates respectively).

The bot intentionally over-weights the high-probability dominant leg and uses the low-probability hedge leg as cheap insurance. The aggregate ROI is 3.02%.

Sub-bucket concentration check: Unlike LIL222 or SirMartingale's 101-cent spread, this wallet has no single dominant cent. The entry price distribution is genuinely spread across all price points, consistent with a bot that enters at whatever the CLOB offers for its desired position size, not one that anchors to a specific tick.

Phase 5: Category and Vertical Breakdown

The category breakdown is already shown above. The interesting sub-vertical analysis is by asset and duration, which the CSV and slug patterns reveal:

Asset coverage: BTC (primary), ETH (secondary), SOL (tertiary), XRP (quaternary). The volume hierarchy follows liquidity: BTC markets have the most depth, so the bot can deploy more per market.

Duration coverage: Both 5m and 15m windows are active. The 15m markets allow more time for the directional signal to play out, which may explain why some of the best single-market P/L trades in best_markets_by_pnl are 15m windows.

Top single-market performers (by P/L):

The ETH market (eth-updown-15m-1775688000, the April 9 3AM ET market) generating +$889 on $988 deployed (90% ROI) in one market is the extreme tail of the distribution. This is where a paired trade resolved with a sub-$1.00 cost and the dominant side won decisively.

Worst single-market performers:

The worst markets are concentrated early in the window. Bitcoin Up or Down April 9 2:05AM-2:10AM ET: -$3,332 on $3,332 deployed, 0 wins. All 167 trades in that market went to zero. The bot placed the wrong directional bet and the hedge leg (if any) also lost. This is the maximum per-market loss exposure: ~$3,300.

Phase 6: Timing and Execution Analysis

Hourly P/L Heat Map

The bot generates positive P/L in 23 of 24 hours (hour 7 is the sole loser at -$2,242). The top hours by absolute P/L:

Hour 5 UTC (+$27,173) is the single highest-P/L hour despite a moderate 64.8% win rate. This is 1 AM Eastern, when crypto vol is moderate and competition on the Polymarket CLOB may be thinner, allowing the bot's signal to find better-priced entries.

Hour 23 UTC has the highest win rate (71.1%), which corresponds to 7 PM Eastern. This is end-of-US-business-day, when BTC/ETH vol typically spikes after equity market close.

Day of week analysis:

Wednesday is the only negative P/L day (-$3,227) despite a 65.8% win rate, suggesting adverse market conditions or an unusually high proportion of paired markets where both the spread and directional calls underperformed. Tuesday has the highest absolute P/L (+$56,886). Sunday has the highest ROI (+4.58%) despite the lowest win rate (59.7%), suggesting the bot's entry prices were particularly favorable on Sundays (likely due to thinner competition).

Second-side lag on paired markets: 140 seconds median. This is fast enough to be intentional pairing (not coincidental two-legged betting) but slow enough to suggest the bot is watching the first leg's price action before committing the second leg. A true simultaneous MM would have a lag close to zero.

Burst pattern: The CSV shows consistent batches of 5-15 fills within 2-10 seconds, followed by gaps of 10-60 seconds. This is a "burst and wait" pattern: the bot fans out a position across multiple fills to walk the orderbook without moving the price, then waits for the market to update.

Phase 7: Filter Experiments

Finding 1: The high-conviction filter is the most powerful single filter. Restricting to markets with dominance ratio 2x or higher, dominant side only, produces 163,380 trades at 75.8% WR and +6.43% ROI on $2.92M of capital. That's +$187,602 from 28% of the capital base. The lift is real and large.

Finding 2: The price $0.30-$0.70 filter produces modest absolute lift on the subset (+4.99% ROI vs 3.02% baseline), but it removes 58% of the capital and 58% of the P/L. Net effect on the overall book is neutral to negative.

Finding 3: The category filter is a near-no-op. Crypto is 99% of the book. Filtering to Crypto alone barely changes anything.

Finding 4: The worst-hour exclusion marginally improves ROI (3.03% vs 3.02%) by removing the hour-7 loss (-$2,242). The effect is tiny. The bot is already avoiding most bad hours via its own signal quality.

Finding 5: The combined filter (price + high conviction) is the most concentrated alpha: +5.34% ROI on $1.99M of capital, +$106,356 P/L. This is the purist's replication target.

Phase 8: Rolling Window Consistency

Every rolling window closes positive. The 7-day windows range from +$39,723 (the Apr 30 window) to +$103,786 (the Apr 16 window). The variance is significant (2.6x between best and worst 7-day), reflecting the day-of-week effects (Wednesday drag) and intraweek vol in crypto markets. But even the worst 7-day window generates ~$40K.

Weekly stair-step:

The weekly breakdown shows the cumulative P/L step from $64K (Week 15) to $127K (Week 16) to $178K (Week 17) to $244K (Week 18) to $259K (partial Week 19). The growth is not linear: Week 17 was the weakest full week at +$50,332, while Weeks 16 and 18 each added ~$63-66K. The variation tracks crypto vol: lower-vol weeks produce thinner edges and fewer profitable markets.

Phase 9: P/L Decomposition

The spread component is a net NEGATIVE. Despite 40% of paired markets having sub-$1.00 paired costs, the overall spread P/L across all 12,635 paired markets is -$142,045. The 60% of markets with above-break-even paired costs drag the average below zero. This is the critical implication: this wallet is NOT primarily a spread-capture strategy. The spread is a side effect of the hedging behavior. The primary alpha is directional.

The directional P/L (total realized minus all spread and hedge components) must be approximately +$1.38M to explain the +$258K net after absorbing the -$980K hedge tax and -$142K spread drag. That directional alpha comes from the 77.7% win rate on high-conviction bets.

Why the hedge tax is so large: At mean paired cost of $1.047 across all 12,635 paired markets, the non-dominant leg averages 15-20 cents per share. With $980K deployed on losing hedge legs, the bot is spending real money on insurance. The insurance pays off by reducing the variance of outcomes and, occasionally, saving the book when the dominant side is wrong.

Phase 10: Strategy Specification

One-sentence summary: A 24/7 multi-asset crypto Up/Down bot that pairs both outcomes when market pricing is close, tilts heavily to one side when confident (3x+), and actively sells the losing leg and winning leg for early exits.

Market selection: All crypto Up/Down markets on Polymarket: BTC, ETH, SOL, XRP in 5m and 15m windows. Opportunistic participation in longer-window markets (hourly, XRP 7PM ET).

Entry logic: Two modes based on signal confidence. Low confidence: buy both sides with roughly equal sizing, capturing spread when paired cost is below $1.00. High confidence: buy both sides with 3x+ tilt to the dominant side. In extreme conviction (implied by the 3x+ dominant-side fill pattern), the hedge leg is minimized to ~15-20% of total exposure.

Exit logic: Active SELLs on the losing leg (cutting to ~$0.20-$0.35 as the market moves against) and winning leg pre-resolution (selling at $0.50-$0.95 to lock gains before the window closes). 15.2% of trades are SELLs.

Sizing model: $5-$282 per fill, $5.42 median, bounded hard at $282. No Kelly scaling visible. Trade count scales to fill the target position size across multiple fills.

Edge source: Primary = directional accuracy at high dominance (77.7% at 3x+). Secondary = spread capture on the 40% of paired markets with sub-$1.00 paired cost. The edge source most likely traces to spot-tape reading for BTC/ETH and a model-based signal for SOL/XRP.

Risk management: Both-sides hedging on 33% of markets bounds per-market catastrophic loss. Hard clip size cap at ~$280. Maximum single-market loss observed: -$3,332. Average daily drawdown risk: small relative to average daily P/L of ~$9,580.

Weaknesses: Wednesday structural underperformance. Hour-7 UTC consistently negative. The hedge tax (-$980K over 27 days) is the largest single drag on performance. If the directional signal degrades (win rate at 3x+ drops below 65%), the strategy becomes unprofitable because the hedge tax is fixed but the directional wins that cover it are not.

Wallet: 0x89b5cdaaa4866c1e738406712012a630b4078beb Window: 2026-04-09 → 2026-05-05 (27 active / 27 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: 33.27% (12,635 of 37,982 markets)
• Median paired cost: $1.0593
• Mean paired cost: $1.0474
• Paired cost % under $1.00: 43.6%
• Paired cost % under $0.97: 40.2%
• Median 2nd-side hedge lag: 140s

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: 27 of 27 (100.0%)
• Rolling 7-day P/L range: +$8,882 → +$103,786
• Rolling 15-day windows green: 27 of 27 (100.0%)
• Rolling 15-day P/L range: +$8,882 → +$168,145

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 05:13 UTC.

Wallet: 0x89b5cdaaa4866c1e738406712012a630b4078beb Window: 2026-04-09 to 2026-05-05 Baseline: 588,539 resolved BUYs | 64.4% WR | $8,565,499 deployed | +$258,554 P/L | +3.02% ROI

The headline result

One filter produces large, real, positive lift. The others are no-ops or dilutive on absolute P/L.

The high-conviction filter (dominance 2x+, dominant leg only) concentrates 73% of the P/L into 28% of the capital at a 6.43% ROI vs 3.02% baseline. This is the actionable finding. Everything else in the filter battery is a structural rearrangement that either has no effect (category filter) or removes good fills along with bad ones (price band filter).

Filter results table

Filter-by-filter commentary

1. Price band filter ($0.30-$0.70) DILUTIVE ON ABSOLUTE P/L

Applying the standard sweet-spot filter retains 245,858 trades at 55.6% WR, producing +$118,484 P/L at +4.99% ROI. The ROI improves (4.99% vs 3.02%), but absolute P/L falls by -$140,070 because the filter removes 58% of the capital and 54% of the winning trades.

The mechanism is structural, not pathological. This wallet's capital-weighting is heavily skewed toward the $0.70-$0.90 zone (the dominant legs of high-dominance paired positions). Those positions have low individual ROI (2.0-2.1%) but enormous capital allocation and aggregate P/L (+$85,585 between $0.70 and $0.90). Removing them to "clean up" the price band cuts the book's best absolute-dollar contributor.

For capital-constrained replicators: the $0.30-$0.70 filter does concentrate ROI. If you have $500K of working capital and want to extract maximum return per dollar deployed, the sweet-spot filter is appropriate because you cannot deploy $8.5M anyway. But do not apply it to a full replication of the strategy architecture.

The other side of this finding: The $0.00-$0.10 band at +53.4% ROI ($28K deployed, +$14,961 P/L) is the hedge leg of many paired positions. Removing it by applying a $0.30+ floor costs +$14,961 in P/L. The hedge legs are cheap insurance that occasionally pays out large; filtering them out is not free.

2. High-conviction filter (dominance 2x+) MEANINGFUL LIFT

This is the only filter in the battery that produces genuine, material improvement in risk-adjusted returns while retaining a large fraction of absolute P/L.

Mechanism: The dominance ratio is a direct proxy for the bot's own confidence in its directional signal. Markets where the bot allocates 2x or more to one side are markets where its internal model favored that side strongly. As Phase 3 showed, those markets hit 64.3% (2-3x) and 77.7% (3x+) dominant-side win rates. The filter keeps exactly these markets and their dominant-side allocations.

Numbers: - 163,380 trades retained out of 588,539 (27.8%) - $2,917,346 capital retained out of $8,565,499 (34.1%) - +$187,602 P/L retained out of +$258,554 (72.6%) - ROI lifts from 3.02% to 6.43%

Put differently: 34% of the capital produces 73% of the P/L. The remaining 66% of capital (the low-dominance trades, the pure spread positions, the hedge legs) produces only 27% of the P/L.

For replicators: If you are building a bot to replicate this strategy, focus exclusively on markets where you have 2x+ confidence in the dominant side. Skip the markets where you would pair near-equally. Those near-equal pairs are how this wallet hedges its uncertainty; if you replicate only the high-conviction subset, you eliminate the hedge tax and concentrate on the directional alpha.

3. Category filter (Crypto only) NO-OP

Crypto is 99.2% of the book by trade count and 99.6% of BUY notional. Filtering to Crypto alone removes 5,142 trades (0.87% of the book) and $36,631 of capital (0.43%), producing P/L of +$258,055 vs +$258,554 baseline. The difference is -$499.

This filter is structurally inapplicable to a portfolio that is already almost entirely one category. The "Other" markets (the 0.8% non-crypto slice) generate +$499 at +1.36% ROI, which is below the 3.02% baseline but positive. Removing them is slightly dilutive on absolute P/L, nearly invisible on ROI.

For replicators: There is no reason to apply a category filter here. The bot is already crypto-only in practice. If building a replication bot, simply target the crypto Up/Down slug patterns and ignore everything else by construction.

4. Hour exclusion filter (worst 4 hours: 4, 7, 13, 21) NO-OP

Excluding the four worst-performing UTC hours (4, 7, 13, 21) removes 92,538 trades and $1,200,369 of capital, producing +$223,434 P/L at +3.03% ROI. The ROI improvement is less than 0.01 percentage points. The absolute P/L loss is -$35,120.

The reason the filter barely helps: only hour 7 is negative (-$2,242 P/L). Hours 4 (+$1,126), 13 (+$14,332), and 21 (+$21,904) are all positive contributors. Removing them costs real P/L. The filter algorithm identifies these as "worst hours" based on relative WR, but WR alone is not sufficient evidence to filter hours on a strategy where capital allocation per hour is the dominant driver of absolute P/L.

Hour 7 (UTC): -$2,242 P/L on 29,396 trades. This is the only genuinely negative hour. However, at the scale of this book ($8.57M deployed), losing $2,242 in one hour represents a 0.026% hourly loss, not a structural problem. Excluding hour 7 alone would save $2,242 but cost the execution infrastructure of firing nearly 30K trades. Not worth optimizing for.

For replicators: Run the bot 24/7. The hour-exclusion optimization is below the noise threshold for this strategy.

5. Combined filter (price $0.30-$0.70 + high-conviction 2x+) USEFUL FOR CAPITAL-CONSTRAINED REPLICATION

The combined filter retains 203,267 trades at 56.3% WR, +$106,356 P/L, +5.34% ROI. This is the most concentrated alpha for a capital-constrained replicator.

The price filter reduces the combined set slightly from the pure high-conviction result (+$187,602, 6.43%) because it strips out the $0.70-$0.90 dominant legs (which were the high-dominance, high-capital positions). The combined filter is therefore slightly worse than the pure dominance filter alone.

Verdict: Use the high-conviction filter alone (not combined with price). The price filter removes exactly the positions that the dominance filter was designed to keep.

What filters would add genuine value (but require data we don't have)

The standard PR&R filter battery does not capture the optimal levers for this strategy. The genuinely useful refinements would require:

Bottom line for replication

The filter analysis produces one actionable recommendation and several non-recommendations:

DO apply the dominance filter (2x+ for replication, 3x+ for alpha concentration). This is the most powerful single lever. It concentrates ROI from 3.02% to 6.43% while retaining 73% of absolute P/L from 34% of the capital.

DO NOT apply the $0.30-$0.70 price filter unless you are capital-constrained. This filter removes the high-capital dominant legs that produce the strategy's largest absolute P/L contributions. It lifts per-dollar ROI but costs real dollars.

DO NOT apply the hour exclusion filter. Only one hour is negative (-$2,242) and the cost of excluding the other "worst" hours (which are positive) outweighs the gain from skipping hour 7.

DO NOT apply the category filter. It is structurally a no-op.

The single most useful thing a replicator can do is build the dominance-ratio classifier into the entry logic and use it as the primary position-sizing signal.

Source wallet: 0x89b5cdaaa4866c1e738406712012a630b4078beb Strategy: Hybrid both-sides paired entry with high-dominance directional tilt across multi-asset crypto Up/Down markets Reference book: $8.57M BUY notional over 27 days | +$258,687 net P/L | +3.02% ROI on deployed | 24/7 operation

One-paragraph operator brief

Build a Polymarket bot that continuously monitors all active crypto Up/Down markets (BTC, ETH, SOL, XRP across 5m and 15m windows), applies a real-time directional signal (spot tape or equivalent), and enters each market in one of two modes: (a) high-conviction mode when the signal favors one side by 3x or more in sizing, or (b) paired-hedge mode when the signal is weak and a sub-$1.00 paired cost is achievable. In high-conviction mode, buy the dominant side large and the hedge side small. In paired-hedge mode, buy both sides at equal-ish sizing if and only if the combined VWAP is below $1.00. Exit losing legs actively by selling at $0.20-$0.35 when the market moves against the dominant position. Exit winning legs pre-resolution at $0.50-$0.95 to lock gains. Run 24/7. Cap individual fills at $280. Expect 3-6% monthly ROI on deployed capital, with the high-conviction filter pushing that to 6-7% at the cost of lower gross volume.

1. Market Selection

Asset priority for capital allocation: BTC (deepest orderbook, most volume), ETH (second), SOL and XRP (thinner, smaller fills). The reference wallet fires on all four simultaneously. A scaled-down replicator with limited working capital should prioritize BTC and ETH.

Both-sides gate: Only enter paired mode if the combined VWAP across both outcomes is below $1.02 (target sub-$1.00 but allow 2 cents of slippage). Above $1.05, enter directional mode only.

2. Entry Logic

The entry has two distinct modes. The mode selection is driven by the directional signal strength.

def select_mode(market, signal_confidence):
    """
    signal_confidence: float in [0, 1] representing bot's
    probability estimate that the UP side wins.
    0.5 = no edge. 0.8 = strong UP signal. 0.2 = strong DOWN.

    Returns: ('directional', dominant_side, dominance_ratio)
             OR ('paired', None, None)
    """
    # Compute how far signal is from 50/50
    edge = abs(signal_confidence - 0.5)

    if edge >= 0.20:  # Signal says 70%+ probability on one side
        dominant_side = "Up" if signal_confidence >= 0.5 else "Down"
        dominance_ratio = signal_confidence / (1 - signal_confidence)
        return ('directional', dominant_side, dominance_ratio)

    # Weak signal: check if paired cost is attractive
    clob_up_ask = market.up_side.best_ask
    clob_down_ask = market.down_side.best_ask
    paired_cost = clob_up_ask + clob_down_ask

    if paired_cost < 1.02:  # Spread is available
        return ('paired', None, 1.0)  # Equal sizing

    return None  # Skip market

Directional mode (high-conviction): - Dominant side: floor_usdc * dominance_ratio allocated - Hedge side: floor_usdc * 1.0 allocated - Dominance ratio target: 3.0x minimum for the high-conviction filter - Walk the orderbook with 5-15 sequential fills to build position without price impact

Paired mode (spread capture): - Both sides: roughly equal allocation - Only enter if paired VWAP achieves below $1.02 - Smaller absolute sizing than directional mode (sub-$10 per leg typical)

3. Exit Logic

The reference wallet fires SELLs actively on both legs, not just holding to resolution. 15.2% of trades are SELLs, generating $1.13M in pre-resolution proceeds.

Losing-leg exit (cut the hedge):

def manage_losing_leg(position, market):
    """
    When market price moves against the hedge/non-dominant leg,
    sell it back into the orderbook to recover capital.
    """
    current_bid = market.get_bid(position.outcome)

    # Sell if bid has fallen to 30-35 cents range
    # (the reference shows sells at $0.20-$0.35)
    if current_bid <= 0.35 and position.usdc_cost > 5.0:
        qty_to_sell = position.shares * 0.80  # Keep 20% as residual
        post_sell_order(market, position.outcome, qty_to_sell, current_bid)

Winning-leg exit (lock gains pre-resolution):

def manage_winning_leg(position, market):
    """
    When the dominant-leg position is showing unrealized gains
    (price has moved toward $1.00), sell tranches to lock.
    """
    current_bid = market.get_bid(position.outcome)
    seconds_remaining = seconds_until_close(market)

    # Ladder sells at $0.50, $0.70, $0.85, $0.95
    for target_price in [0.50, 0.70, 0.85, 0.95]:
        if current_bid >= target_price:
            tranche = position.shares * 0.25
            post_sell_order(market, position.outcome, tranche, current_bid)

    # If market about to close with a winning position held,
    # let the remainder settle at $1.00
    if seconds_remaining < 30:
        cancel_unfilled_asks()  # Let residual go to settlement

4. Sizing Model

The reference wallet uses bounded, repetition-heavy sizing. No Kelly scaling visible. The ceiling is hard at $282 per fill.

Why not Kelly: Kelly sizing would concentrate capital into the highest-edge bets. This wallet instead sizes uniformly across thousands of markets, trading throughput for concentration. The result is lower per-market ROI but extremely stable aggregate returns. For a first-build replication, uniform bounded sizing is safer and easier to implement.

The capacity ceiling: At $280 per fill and 21,800 fills per day, the reference wallet deploys $317K/day of BUY notional. At smaller scale (1,000 fills/day, $25/fill), you are deploying ~$25K/day, which cycles a $500K balance roughly 20 times in 27 days. The economics scale nearly linearly until orderbook depth limits apply.

5. Both-Sides Allocation

The reference wallet's pairing logic is not random. The 140-second median second-side lag indicates the bot watches the first leg's price action before committing the second leg. This is an informed pairing, not mechanical.

Pairing rules:

def compute_pair_sizing(market, signal_confidence, base_clip):
    """
    Returns (dominant_usdc, hedge_usdc) sizing for a paired entry.
    """
    edge = abs(signal_confidence - 0.5)

    if edge >= 0.30:  # Very strong signal: 80%+ probability
        dominance_ratio = 5.0  # 5x allocation on dominant
    elif edge >= 0.20:  # Strong signal: 70%+ probability
        dominance_ratio = 3.0
    elif edge >= 0.10:  # Moderate signal: 60%+ probability
        dominance_ratio = 1.5
    else:  # Weak signal: near 50/50
        dominance_ratio = 1.0  # Pure spread play

    dominant_usdc = base_clip * dominance_ratio
    hedge_usdc = base_clip * 1.0  # Hedge always at base clip

    return (dominant_usdc, hedge_usdc)

The second-side rule: After entering the first leg, wait up to 120 seconds and observe the price action. If the first leg is moving against you (price falling on your entry), reduce the hedge leg size by 50%. If the first leg is moving in your favor (price rising toward $1.00), execute the hedge leg at full size to lock the paired cost structure.

6. Hour Scheduling

The reference wallet runs 24/7 with no sleep window. However, not all hours are equal.

Wednesday caution: Wednesday is the only negative P/L weekday (-$3,227). Reduce sizing to 75% on Wednesdays until you have enough data to characterize the Wednesday pattern in your own market conditions.

Sunday opportunity: Sunday has the highest ROI (+4.58%) despite the lowest absolute WR (59.7%). This is consistent with thinner competition from professional CLOB participants on weekends. Run at full size on Sundays.

7. Operational Requirements

8. Risk Profile

The strategy is structurally bounded-loss per trade. The combination of hard fill caps and both-sides hedging means no single market can cause a catastrophic drawdown. The risk is slow bleed if the directional signal degrades across many markets simultaneously over multiple days.

9. Diagnostic Checklist

Run weekly. Compare to reference wallet benchmarks.

10. What This Playbook Deliberately Does Not Include

No pure spread-only mode without directional filtering. The spread-capture mechanism alone (buying both sides at sub-$1.00 paired cost) is not the strategy's primary alpha source. At median paired cost $1.059, the spread is actually a net negative across all paired markets combined (-$142K over 27 days). The paired trades are only profitable because the dominant-side directional accuracy covers the spread cost and then some.

No aggressive Kelly sizing on high-dominance bets. The reference wallet uses bounded, repetitive sizing even at 3x+ dominance. Kelly would concentrate capital into the highest-confidence positions, which sounds appealing but would increase per-market variance and require more working capital to manage drawdowns. The reference strategy's uniformity is a deliberate risk-reduction choice.

No 1-minute or sub-5-minute markets. The reference wallet's universe is 5m and 15m windows. Sub-5m markets (if they exist) have insufficient time for the both-sides pairing workflow to execute cleanly. The bot needs at least 60-120 seconds from signal detection to full position entry.

No non-crypto assets. The 0.8% "Other" allocation in the reference wallet generates positive but below-baseline ROI. The directional signal is crypto-spot-based. Extending to sports or politics markets would require entirely different signal infrastructure.

No waiting for paired cost to naturally fall below $1.00 before entering. The bot enters when the signal fires, whether paired cost is $0.95 or $1.09. The paired cost filtering is a gate only for the pure-spread mode (when there is no directional signal). In directional mode, paired cost is irrelevant; the signal justifies the entry regardless.

No single-market stop-loss. The reference wallet does not stop out of positions that are moving against the dominant side (beyond selling the hedge leg). It holds the dominant leg through resolution. The bounded sizing makes stop-losses unnecessary: the worst per-market outcome is a ~$280-$380 loss, which the strategy's positive-EV structure absorbs comfortably across thousands of markets.

The reference wallet's design is disciplined and self-consistent. Every "improvement" a builder might be tempted to add is something the original operator implicitly tested and discarded. The bounded sizing, the 24/7 schedule, the paired-hedge behavior, and the dominance-ratio-driven entry mode selection are all load-bearing components. Trust the architecture.