NeverSmiling

Wallet: 0xfcdc071df7080c214196bb0b3b751e5417f9d8e3 Window: 2026-05-18 to 2026-06-02 (16 calendar days, 16 active) Universe: 173,640 trades · 6,360 unique markets · $1,022,056 gross BUY notional Net computed P/L: +$4,337 on $1,022,056 deployed = +0.42% ROI (16 days)

The Punchline

This is not a directional betting wallet. NeverSmiling is a both-sides spread-capture bot running the full menu of short-duration BTC, ETH, SOL, and XRP Up/Down markets on Polymarket. It buys both Up and Down on 79.2% of the markets it touches, locking in sub-$1.00 paired costs as a structural profit mechanism, then tilts asymmetrically toward the side it believes will resolve correctly. The directional tilt is real - at 3x+ dominance ratio the dominant side wins 78.8% of the time - but the spread is the engine that makes small directional errors affordable.

The economics are tight. $1.02M deployed returns $4,337 computed - a 0.42% ROI over 16 days. At Polymarket's own reported figure of $20,977 the ROI is ~2.05%. Either way, this is a low-margin, high-volume, industrial operation, not a discretionary trading book. The bot generates edge by being everywhere at once - 173,640 fills across 6,360 markets - and collecting micro-profits on each that aggregate into a meaningful absolute dollar figure.

---

What He Trades

The universe is the full short-duration crypto Up/Down product suite. From the CSV:

btc-updown-15m-*   BTC 15-minute windows   (dominant by volume, early in window)
btc-updown-5m-*    BTC 5-minute windows
eth-updown-5m-*    ETH 5-minute windows
sol-updown-5m-*    SOL 5-minute windows
sol-updown-15m-*   SOL 15-minute windows
xrp-updown-5m-*    XRP 5-minute windows
xrp-updown-15m-*   XRP 15-minute windows

This is broader than a typical single-asset microstructure wallet. The bot is running across the full crypto Up/Down vertical - every available asset and every available duration appears in the CSV sample. All 173,640 trades classify as Crypto.

No sports. No politics. No longer-duration crypto markets. No hourly or 4-hour windows visible in the data. The focus is exclusively on the 5-minute and 15-minute Up/Down series where the orderbook resets every window and paired-cost opportunities refresh continuously.

Trade size by asset cannot be precisely broken out from the aggregated data, but the top_markets list is heavily BTC-weighted, with individual BTC windows seeing 30-208 trades and $930-$1,766 of volume per window. The SOL and XRP windows visible in the CSV tail show smaller clips ($1-$4 per fill), consistent with thinner orderbooks and lower-liquidity assets.

---

The Order of Operations - One Market, Trade by Trade

Illustrating the strategy with Bitcoin Up or Down - May 18, 8:15AM-8:30AM ET (btc-updown-15m-1779106500), which resolved Down. This is the earliest full market visible in the tail of the CSV and shows the complete paired-cost entry pattern.

Walk-through:

1. Entry phase (12:15-12:17 UTC). The bot enters Up at prices from $0.39 to $0.49 - mid-range coin-flip zone. These are multiple small clips of $2-$17, walking the orderbook. Simultaneously it is building the Down position at $0.20-$0.36.

1. Both-sides construction. Over the first 12 minutes of the 15-minute window, the bot accumulates Up and Down simultaneously. At 12:21:49 it hits the Down side heavily - a burst of 8 fills at $0.33 for a combined ~$34. This is the dominant leg building phase.

1. Near-certainty sweep (12:27:13). The bot buys 107.47 shares of Down at $0.91 for $98.42. This is its largest single clip in the window - buying the near-certain side as the market approaches resolution. This is a high-conviction dominant-leg top-up.

1. Longshot hedge (12:27:20-12:28:56). It simultaneously buys Up at $0.06-$0.08 - 170 shares for ~$11.62 total. This is the cheap hedge that occasionally pays 14x if the market flips. The longshot ROI on sub-$0.10 buys is +7.2% because these occasionally hit.

1. Resolution: Down wins. All Down shares pay $1.00. All Up shares pay $0.00. Net on this market: Down wins, the $98.42 near-certainty fill pays ~$107 gross, the $34 mid-range Down buys pay ~$43, total receipts vs total outlay determine the per-market P/L.

This pattern repeats across every market: simultaneous construction of both sides, dominant leg typically at mid-price ($0.30-$0.60), explicit near-certainty top-ups when conviction is high ($0.90+), cheap longshot hedge at $0.02-$0.15. The whole sequence unfolds within the 5-15 minute window.

---

Why It Works - The Math

The strategy has three stacked positive-EV components:

Component 1: Paired cost below $1.00
  Median paired cost:        $0.9836
  % markets sub-$1.00:       54.4%
  % markets sub-$0.97:       46.5%
  Per-market spread on sub-0.97:   ~$0.03 per paired dollar
  Computed spread P/L (16 days):   $160 (conservative - only fully-paired shares)

Component 2: Directional accuracy on dominant leg
  Dom 1.0-1.5x → 53.7% win rate (near coin flip, minimal edge)
  Dom 1.5-2.0x → 66.2% win rate (+16% above fair value)
  Dom 2.0-3.0x → 74.2% win rate (+24% above fair value)
  Dom 3.0x+    → 78.8% win rate (+29% above fair value)
  
  4,034 total both-sides markets × implied EV gain on dominant leg
  vs hedge tax on non-dominant leg

Component 3: Longshot lottery on cheap hedge leg
  Sub-$0.10 entries: 4,932 trades, 7.5% win rate, +7.2% ROI
  Sub-$0.20 entries: 15,240 trades, 15.3% win rate, +6.2% ROI
  These win rates are slightly above the fair-value implied probability,
  or the occasional 13x payout on a $0.07 buy lifts the average enough
  to net positive even on imperfect calibration.

The hedge tax is the dominant drag. Total hedge tax (USDC spent on non-dominant legs that ultimately lost) computed at $369,270. The directional wins must cover this. With $4,337 net P/L, the directional accuracy is barely covering the hedge cost. This explains why the ROI is thin - the strategy is balanced on a knife edge between spread capture and hedge tax.

<pre><code>For a representative market with dominance ratio 3x: Deploy $6 on dominant side @ $0.60 → 10 shares Deploy $2 on non-dominant @ $0.30 → 6.7 shares Total deployed: $8.00 Paired cost: $0.60 + $0.30 = $0.90 (sub-$1.00, spread locked)

If dominant wins (78.8% probability): Collect $10 on dominant, $0 on hedge Net: +$2.00 on $8.00 deployed = +25%

If dominant loses (21.2% probability): Collect $6.70 on non-dominant, $0 on dominant Net: -$1.30 on $8.00 deployed = -16.25%

EV = 0.788 * 2.00 + 0.212 * (-1.30) = +$1.576 - $0.276 = +$1.30 per $8.00 = +16.25% EV per cycle at 3x dominance

At 1.5x dominance (66.2% win rate): Deploy $6 dominant @ $0.60, $4 non-dominant @ $0.40 If dom wins: collect $10, net +$0 If dom loses: collect $10, net +$0 Paired cost = $1.00 → zero spread, zero directional edge at 1.5x/50-50 price </code></pre>

The math confirms: the strategy generates meaningful EV only at dominance ratios above 2.0x and paired costs below $1.00. The 1.0-1.5x bucket at 53.7% win rate is noise, and those markets likely bleed slightly after fees.

---

Phase 1 - Trader Profile

Scale and Activity

Trade Size Distribution

The size profile is near-uniform - the top 5% carries only 28% of capital, vs 37% for SirMartingale. This is a well-dispersed clip structure. The max fill of $267.09 is only 76x the median - very tight ceiling. No power-law concentration. The bot is spraying equal-weight clips across every available market, not sizing up on high-conviction calls.

Execution Speed

72% of consecutive fills come within 10 seconds of each other. Multiple same-second and same-minute bursts visible in the CSV. This is fully automated. The 22.3-second mean gap vs 2-second median indicates burst-and-pause execution - flurries of 5-15 fills per market window followed by pauses between windows.

Trading Hours

The bot is active in all 24 UTC hours. The overnight hours (00:00-06:00 UTC) still show 4,500-5,200 trades per hour - reduced but not zero. Peak hours are 14:00-20:00 UTC (~9,000-10,000+ trades/hour). This is genuinely a 24/7 operation with modest overnight volume reduction, unlike SirMartingale's hard sleep window.

Archetype: SPREAD CAPTURE + DIRECTIONAL TILT

---

Phase 2 - Core Strategy Identification

Both-sides participation: 79.2%

5,034 of 6,360 unique markets show both-side purchases. This is the defining number. Any wallet with 79.2% both-sides rate is primarily a spread-capture / market-making operation. The remaining 20.8% of markets (1,326) are one-sided - likely cases where the bot's signal was too strong on one outcome to bother with the hedge.

The strategy is a hybrid A+B:

• A (Both-Sides Spread Capture): 79.2% both-sides rate, median paired cost $0.9836, 54.4% sub-$1.00
• B (Directional Betting): Strong dominance-ratio accuracy signal (78.8% at 3x+), clear conviction scaling

It is not:

• A latency arbitrageur (trades across too many assets too evenly, no SELL leg)
• A copy-trader (no detectable lag pattern)
• A DCA accumulator (each market is fresh, no returning to prior positions)
• A pure longshot bot (39% of capital is in $0.60+ entries)

The no-SELL signature is the most operationally significant fact. Every single one of 173,640 trades is a BUY. The wallet holds all positions to resolution. Combined with the both-sides structure, this means the strategy earns its P/L from settlement payouts, not from active exit management. This is fundamentally different from SirMartingale's SELL-engine approach.

---

Phase 3 - Dominance Ratio Analysis

This is the most informative phase for this wallet.

The mean paired cost at 2.0x+ sits right at $1.00, meaning the spread capture contribution from these high-conviction markets is essentially nil - the bot is expressing pure directional conviction, not locking in a riskless spread. The spread P/L comes from the 1.0-1.5x bucket where mean paired cost is $0.939 - 6 cents below $1.00.

Critical insight: The strategy's two components are negatively correlated in their contribution. The high-spread markets (1.0-1.5x, paired cost $0.939) deliver low directional accuracy (53.7%). The high-accuracy markets (3.0x+, win rate 78.8%) have no spread at all (paired cost $1.000). The total P/L is the sum of these two partially-independent engines.

Second-side lag: Median 50 seconds between entering first and second side of a paired market. This is fast pairing - intentional simultaneous construction within the same market window, not opportunistic hedging. A lag under 60 seconds means the second side is bought as part of the same execution sequence, confirming these are paired by design.

---

Phase 4 - Entry Price Analysis

The win-rate calibration is perfect - 7.5% wins at $0.00-$0.10 entries correctly reflects ~5-10% implied probability, and 95.4% wins at $0.90-$1.00 entries correctly reflects ~95%+ implied probability. The market is pricing these outcomes accurately on average.

The ROI inversion reveals the structural tension: cheap entries earn positive ROI because the lottery payoffs on occasional wins exceed the cost basis. Expensive entries lose ROI because the margin on each win is razor-thin and any losses are devastating to ROI. The bot is losing $4,517 on its near-certainty buys and making $1,611 on its longshot buys - these two wings partially cancel each other out.

Sub-bucket inspection: The price distribution across the full range shows no single-tick concentration. Capital spreads fairly evenly from $0.30 to $0.70 with the heaviest concentration in the $0.40-$0.60 band ($338,505 combined = 33% of capital). This is the coin-flip zone where paired-cost spread capture works best.

---

Phase 5 - Category and Market-Type Breakdown

Single-category book. The interesting breakdown is by asset (derived from market slugs):

The top_markets list is exclusively BTC - the highest-volume single windows see 208 trades and $1,766 of volume. BTC dominates by absolute dollar volume. The best_markets list includes one-trade BTC wins of $639 and $614, suggesting high-dominance single-clip plays on near-certainty BTC resolution moments.

Assessment: Crypto - +0.42% ROI - thin but positive across the full universe.

---

Phase 6 - Timing and Execution

Hourly P/L (UTC)

The hourly P/L variation is noisy rather than structural. No single hour shows more than $2,300 of P/L on ~7,000 trades - the per-trade edge is so small that random variance in outcome resolution dominates the hourly pattern. The worst hours identified by the filter system (hours 1, 14, 15, 23) have slightly below-average win rates but the signal is weak.

The bot is genuinely 24/7. Unlike SirMartingale's hard sleep window, the hour histogram shows 4,545-10,148 trades in every UTC hour. The overnight reduction (hours 00-06 at ~4,500-5,200 trades) vs daytime (hours 14-20 at ~9,000-10,000 trades) reflects BTC/ETH market activity levels, not an operator sleep schedule.

Day-of-week P/L

Saturday is the best day (+2.05% ROI, +$2,885). Thursday and Friday are positive. Tuesday and Sunday are negative. The pattern is consistent with the weekend competition-reduction thesis - fewer competing market makers active on Saturdays.

---

Phase 7 - Filter Experiments

The filter picture for this wallet is the opposite of SirMartingale: the standard filters do work, not because the unfiltered book is broken, but because the worst-performing zones (near-certainty buys, overnight hours) are genuinely dragging the ROI down.

The price-band filter (0.30-0.70) is the single most useful lever: it strips the $0.90-$1.00 zone that loses $4,517 and the $0.80-$0.90 zone that loses $1,507, while keeping the coin-flip zone that earns positive ROI. The improvement is +$2,531 absolute and +0.72 percentage points of ROI - real, not cosmetic. See the Filters tab for full commentary.

---

Phase 8 - Rolling Window Consistency

The rolling window profile is weak by elite-trader standards. 56% of 7-day windows positive means the strategy has meaningful variance - it can lose money for a full week. The first rolling 7-day windows are mostly negative (May 18-22 all show negative or near-zero daily P/L in the rolling7 series).

Week 2 (May 25-31) was the standout at +$3,017 - nearly 70% of the total computed P/L came from one week. Week 1 barely broke even. The cumulative line is not monotonic; it shows meaningful negative periods in the first half.

The rolling 15-day series shows the same pattern: first 6 entries are negative, then a regime shift around May 24-25 where the book starts consistently printing positive.

Interpretation: With only +0.42% ROI per period, any stretch of slightly-below-average resolution outcomes produces a losing week. The edge is real but thin enough that weekly variance is large relative to the edge.

---

Phase 9 - P/L Decomposition

The decomposition reveals the hedge tax magnitude. $369,270 was spent on losing legs of both-sides markets. The dominant-leg wins must cover this plus generate the $4,337 net. This requires the dominant leg to return approximately $373,607 above cost - which with 78.8% accuracy on 1,420 high-conviction markets and 66-74% on the others is achievable but tight.

The $160 spread P/L is negligible relative to the overall book size - the spread capture mechanism is real but minor. The primary driver is directional accuracy, not spread.

P/L validation caveat: Polymarket reports $20,977 vs computed $4,337, a gap of $16,640. This exceeds the 10% tolerance threshold. The most likely explanation is open positions in the snapshot window - markets that have not yet resolved - that Polymarket marks at last-traded price (above cost in aggregate) but our system counts as unresolved (excluded or at cost). The true cash-flow P/L when all positions resolve is likely between $4,337 and $20,977.

---

Phase 10 - Strategy Specification

One-sentence summary: A 24/7 automated both-sides spread-capture and directional-tilt bot that buys Up and Down simultaneously on 5-minute and 15-minute BTC, ETH, SOL, and XRP Up/Down markets, locking in sub-$1.00 paired costs and tilting 2-5x toward the side with a higher directional signal, holding all positions to settlement.

Edge source: Two partially-independent mechanisms: (1) sub-$1.00 paired costs on 54% of markets generating $0.01-$0.05 per dollar of paired capital, and (2) directional accuracy at high dominance ratios (78.8% at 3x+) generating alpha above the fair-value hedge tax cost.

What works: The $0.30-$0.70 price band (+1.14% ROI vs +0.42% baseline). High-dominance markets (2x+, 74-79% win rate). Saturdays (+2.05% ROI). Hours 11:00 and 22:00 UTC.

What drags: Near-certainty buys ($0.90-$1.00) lose $4,517 at -3.8% ROI. The first week of the observation window was approximately breakeven to slightly negative before the book hit its stride.

What replicators must do: Implement the combined filter (price 0.30-0.70 + exclude worst 4 hours) to lift ROI from +0.42% to +1.59% without losing meaningful absolute P/L. See the playbook for the full implementable spec.

Wallet: 0xfcdc071df7080c214196bb0b3b751e5417f9d8e3 Window: 2026-05-18 → 2026-06-02 (16 active / 16 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: 79.15% (5,034 of 6,360 markets)
• Median paired cost: $0.9836
• Mean paired cost: $0.9758
• Paired cost % under $1.00: 54.4%
• Paired cost % under $0.97: 46.5%
• Median 2nd-side hedge lag: 50s

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 16 (75.0%)
• Rolling 7-day P/L range: -$694 → +$4,250
• Rolling 15-day windows green: 12 of 16 (75.0%)
• Rolling 15-day P/L range: -$694 → +$4,591

Weekly P/L

Phase 9 - P/L Decomposition

Phase 10 - Top Markets by Volume

Top 10 winners by P/L

Top 10 losers by P/L

Report generated 2026-07-04 08:09 UTC.

Wallet: 0xfcdc071df7080c214196bb0b3b751e5417f9d8e3 Window: 2026-05-18 to 2026-06-02 Baseline: 173,640 BUYs · 48.5% WR · $1,022,056 deployed · +$4,337 P/L · +0.42% ROI

---

The headline result

Two filters add real lift. One is a no-op. The rest are identity-equivalent or inapplicable. The combined filter - price band 0.30-0.70 plus excluding the four worst hourly buckets - reduces trade count by 52% but lifts ROI from +0.42% to +1.59% while improving absolute P/L by +$3,586. The improvement is structural: it removes the near-certainty zone that bleeds money at -3.8% ROI and the worst overnight hours.

The high-conviction dominance filter (dom ≥ 2x, dominant leg only) performs counterintuitively: it shows a high win rate of 82.9% but lower absolute P/L than the price-band filter alone. This is because stripping the non-dominant legs removes cheap shares that occasionally pay out, while the high-cost dominant-leg fills at $0.70+ still earn thin margins.

---

Filter results table

---

Filter-by-filter commentary

1. Price band filter (0.30-0.70) → MEANINGFUL LIFT

Result: +$6,868 P/L on $600,239 deployed (+1.14% ROI) vs +$4,337 on $1,022,056 (+0.42%). Absolute P/L improves by +$2,531 while cutting deployed capital in half.

The mechanism is clear from the price-band P/L table. The zones this filter removes:

• $0.00-$0.10: +$373 P/L (kept by filter? No - removed). Wait - this zone earns +7.2% ROI. The filter removes it but also removes the $0.10-$0.30 zone which earns +3.7-6.2%.
• $0.70-$0.80: -$437 P/L at -0.35% ROI. Removed by filter.
• $0.80-$0.90: -$1,507 P/L at -1.44% ROI. Removed by filter.
• $0.90-$1.00: -$4,517 P/L at -3.79% ROI. Removed by filter.

The three expensive zones ($0.70-$1.00) collectively lose -$6,461. The filter removes these bleeders at the cost of also removing the cheap zones ($0.00-$0.30) that collectively earn +$3,943. Net filter benefit: -$6,461 removed minus +$3,943 removed = +$2,518 improvement (matches the observed +$2,531 within rounding).

Why the expensive zone bleeds: Buying a $0.95 outcome that wins 95.4% of the time earns only $0.05 per share on a win but loses $0.95 on the rare loss. The bot's actual win rate of 95.4% at $0.90-$1.00 entries is close to but slightly below the break-even required (~97%+ at those prices). Result: systematic small losses every time the "nearly certain" outcome fails.

Recommendation: Apply the $0.30-$0.70 filter as a default entry gate for any replicator. It is the single most impactful standalone filter.

2. High-conviction dominance filter (dom ≥ 2x, dominant leg only) → WEAK LIFT - USE CAREFULLY

Result: +$2,644 P/L on $315,447 deployed (+0.84% ROI). Win rate jumps to 82.9% as expected from the dominance analysis (74-79% at 2x+). But absolute P/L is lower than the price-band filter alone.

The issue is structural: the high-conviction dominant leg is often the expensive leg. When the bot tilts 3x+ toward Down at $0.70, the dominant-leg trades are concentrated in the $0.60-$0.90 price range - exactly the zone that earns weak or negative ROI. The 74-79% win rate beats the implied probability at those prices, but not by enough to overcome the thin margin.

Compared to the price-band filter: The price filter earns +$6,868 on $600,239 (1.14%). The dominance filter earns +$2,644 on $315,447 (0.84%). The price filter wins on both absolute P/L and ROI. The dominance filter is useful as a complement, not a replacement.

The genuine value of this filter: It identifies the 2,537 highest-conviction markets (dom 2x+) where the directional signal is strongest. A replicator who can only take on limited capital should weight these markets heavily. But for a full-book replication, the price-band filter does more work.

3. Top category filter (Crypto) → NO-OP

100% of trades are Crypto. The filter is identity-equivalent to baseline. No insight, no lift.

4. Exclude worst 4 hours (UTC 01, 14, 15, 23) → MODEST LIFT

Result: +$5,858 P/L on $845,898 deployed (+0.69% ROI). Absolute improvement +$1,521, ROI improvement +0.27pp.

The four worst hours by P/L:

• Hour 10: -$1,437
• Hour 12: -$1,483
• Hour 23: -$1,153
• Hour 06: -$835

But the filter system uses hours 1, 14, 15, 23 (the lowest win-rate hours). The win-rate-based selection and the P/L-based selection disagree, which weakens the filter's effectiveness. Excluding hours 14 and 15 (14:00-16:00 UTC, US market open) removes a high-volume window where the bot struggles with win rate (44.2% and 45.9%) - these are genuine bad hours driven by US-open volatility confusion.

The hour filter works, but it is not a primary lever. Excluding 32,153 trades (18.5% of the book) for +$1,521 uplift is a reasonable trade if you want to reduce operational exposure, but the price-band filter does more per dollar of capital removed.

5. Combined filter (price 0.30-0.70 + exclude worst 4 hours) → MEANINGFUL LIFT - BEST STACK

Result: +$7,923 P/L on $499,735 deployed (+1.59% ROI). The best single configuration.

The combined filter cuts the book roughly in half by trade count and capital, but lifts ROI from +0.42% to +1.59% - nearly a 4x improvement in return per dollar deployed. Absolute P/L improves by +$3,586 (+83% better than unfiltered).

This is the correct operating configuration for a replicator who wants to allocate the same working capital more efficiently. Instead of scattering $1M across all markets and all hours, concentrate $500K into the coin-flip zone during productive hours.

---

What filters would add value if measurable

The filters above work on observable dimensions. Several additional refinements are hypothetically valuable but require data beyond the trade CSV:

---

Bottom line for replication

Three actionable conclusions from this filter analysis:

1. Apply the $0.30-$0.70 price gate. This is the single highest-value filter, improving ROI from +0.42% to +1.14% and adding +$2,531 absolute P/L by stripping the near-certainty zone that systematically loses money.

1. Skip hours 14 and 15 UTC (10-11am ET). The US market open creates a win-rate drag at 44-46%. These are high-volume hours (9,866 and 9,806 trades respectively) that produce negative absolute P/L contribution. Avoiding them costs volume but improves quality.

1. Do not attempt to apply the dominance filter alone. It improves win rate to 82.9% but produces lower absolute P/L than the simpler price-band filter. If you want to use dominance as a supplement, combine it with the price filter - only take dominant-leg positions in the $0.30-$0.70 zone at 2x+ conviction.

The combined filter is not just the best filter configuration - it reveals what this strategy actually is at its core: a coin-flip-zone paired-cost bot that should stay in its lane and skip the near-certainty traps. See the playbook for the full implementable spec.

Source wallet: 0xfcdc071df7080c214196bb0b3b751e5417f9d8e3 Strategy: Both-sides spread capture with directional tilt on crypto Up/Down markets Reference book: $1,022,056 BUY notional · +$4,337 computed P/L · +0.42% ROI (16 days) Filtered book (recommended configuration): $499,735 deployed · +$7,923 P/L · +1.59% ROI

---

One-paragraph operator brief

Build a Polymarket bot that simultaneously buys both Up and Down on 5-minute and 15-minute BTC, ETH, SOL, and XRP Up/Down markets, targeting combined VWAP below $0.97, with entries restricted to the $0.30-$0.70 price band only. On each market, tilt the allocation 2-5x toward the side your directional model favors. Hold all positions to settlement - no SELL engine required. Skip hours 14:00-16:00 UTC (US market open). Run 24/7 otherwise. Expect approximately 400 unique markets per day, $30K-$50K of daily BUY notional, and +1.5-2.0% ROI per 16-day cycle on deployed capital at reference scale. The strategy earns through two stacked mechanisms: locked-in spread on sub-$0.97 paired costs, and directional accuracy (78.8% at 3x+ dominance) that exceeds the cost of the hedge leg.

---

1. Market selection

Asset priority order (based on volume concentration in reference book): BTC 15m and 5m first (highest absolute dollar volume), then ETH 5m, SOL 5m and 15m, XRP 5m and 15m. The bot should operate on all of them, but if capital-constrained, prioritize BTC.

Market eligibility check before entering each window:

def is_eligible_market(market):
    if market.asset not in ("BTC","ETH","SOL","XRP"):   return False
    if market.duration not in ("5m","15m"):              return False
    if utc_hour(now()) in (14, 15):                     return False  # US open drag
    pct_elapsed = elapsed_seconds / market.duration_seconds
    if pct_elapsed < 0.10 or pct_elapsed > 0.90:        return False
    if market.up_ask - market.up_bid > 0.15:             return False  # thin book
    return True

---

2. Entry logic

The bot enters both sides of every eligible market. The question is not whether to enter, but how much to put on each side.

Step 1: Compute the directional signal

def get_directional_signal(market, spot_feeds):
    # Use spot price relative to market threshold to compute fair probability
    # The exact model is unknown - reverse-engineered from the dominance
    # pattern: when the bot has 3x+ conviction, it is right 78.8% of the time
    
    fair_prob_up = compute_fair_prob(market, spot_feeds)
    clob_prob_up = market.up_side.mid_price
    
    gap = fair_prob_up - clob_prob_up  # positive = Up is underpriced
    return gap, fair_prob_up

Step 2: Compute the target allocation

def get_target_allocation(fair_prob_up, total_clip):
    # Split clip between Up and Down based on directional confidence
    # Target: dominant/non-dominant ratio of 2x-5x at high conviction
    
    if abs(fair_prob_up - 0.50) < 0.05:          # near-50/50
        up_frac = 0.50                           # equal split
    elif fair_prob_up > 0.50:                    # Up favored
        skew = min(5.0, 1.0 + (fair_prob_up - 0.50) * 10)  # 1x to 5x
        up_frac = skew / (skew + 1.0)
    else:                                        # Down favored
        skew = min(5.0, 1.0 + (0.50 - fair_prob_up) * 10)
        up_frac = 1.0 / (skew + 1.0)
    
    return up_frac * total_clip, (1.0 - up_frac) * total_clip

Step 3: Apply the price-band gate

This is the most important filter. Only place clips at prices between $0.30 and $0.70:

def should_fill_at_price(price):
    return 0.30 <= price <= 0.70

# When walking the orderbook, skip any depth outside this band
# On the dominant side: if the dominant side's best ask is >$0.70, skip
# On the non-dominant side: if the non-dominant side's best ask is >$0.70, skip

Step 4: Walk the orderbook

Fan out 3-8 clips per side to walk the available depth:

def enter_market(market, up_budget, down_budget):
    # Up side
    up_fills = walk_book_buy(market, "Up", max_usdc=up_budget,
                              price_min=0.30, price_max=0.70)
    # Down side  
    down_fills = walk_book_buy(market, "Down", max_usdc=down_budget,
                                price_min=0.30, price_max=0.70)
    
    return up_fills, down_fills

The paired-cost check: Before committing to a market, compute the expected paired cost. If the best available ask on Up is $0.55 and on Down is $0.55, the paired cost is $1.10 - above $1.00, no spread locked in. The bot should still enter if its directional accuracy is high enough (3x+ dominance can still be profitable at paired cost $1.00 via directional alpha alone), but should reduce total clip size.

def compute_paired_cost(market):
    up_vwap_estimate = market.up_side.ask_at_depth(target_usdc / 2)
    down_vwap_estimate = market.down_side.ask_at_depth(target_usdc / 2)
    return up_vwap_estimate + down_vwap_estimate

# If paired_cost > 1.00 and dominance < 2x: skip market
# If paired_cost > 1.00 and dominance >= 2x: reduce clip to 50% of normal

---

3. Exit logic

There is no exit logic. The wallet holds every position to settlement - 173,640 BUYs and 0 SELLs. This is by design, not omission.

The rationale: on 5-15 minute markets, the positions resolve within the window. The transaction cost of posting and managing SELL orders in a thin CLOB exceeds the marginal benefit of early exit for positions sized at $1-$10. The spread is locked in at entry; the settlement payout is the correct exit mechanism.

The one exception worth implementing: cancel any open limit orders before market resolution to avoid stale fills in the final 30 seconds of a window. The CSV shows no partial fills suggesting the bot places market orders (hitting the ask), so this may not apply.

# Settlement is automatic on Polymarket
# No action required - positions pay out at $1.00 (win) or $0.00 (loss)
# All P/L realized at resolution time

---

4. Sizing model

Flat clip per market, scaled by paired cost and conviction:

Size scaling table by bankroll:

Do not scale linearly above $50K working capital. At reference scale, the bot deploys ~$64K/day on $1M working capital across 398 markets. Beyond $25K working capital per wallet, you start moving prices on thinner SOL/XRP books. Fragment across multiple wallets.

The reference wallet's clip structure:

• Median: $3.51
• P95: $16.41
• Max: $267.09 (one outlier in 173,640 fills)
• Top 5% share: 28.3%

This near-uniform size distribution is correct for a spread-capture operation. Do not Kelly-size; Kelly only makes sense when you can accurately estimate win probability per trade. The spread-capture mechanism does not depend on directional accuracy - it locks in profit mechanically.

---

5. Both-sides allocation

The most distinctive feature of this strategy is the paired-entry design. Every market entry consists of two simultaneous clips - one on Up, one on Down - with asymmetric sizing driven by the directional signal.

Target ratios by dominance bucket:

The reference book shows 1,513 markets at 1.0-1.5x, 984 at 1.5-2.0x, 1,117 at 2.0-3.0x, and 1,420 at 3.0x+. That distribution - roughly equal across conviction buckets - implies the directional signal fires at all confidence levels continuously. A replicator should expect similar distribution.

Second-side lag: The reference wallet pairs with a median 50-second lag between first and second side. For a bot, both sides can be entered simultaneously (0-second lag). This is better - it eliminates the risk that the second side's price moves before you can enter it.

---

6. Bankroll math

Reference book extrapolated to monthly cadence:

  16-day computed P/L:           +$4,337  (conservative)
  16-day Polymarket-reported P/L: +$20,977 (may include open MTM)
  Monthly midpoint estimate:      ~$8,000-$40,000 depending on resolution
  
  Capital deployed per 16 days:   $1,022,056
  Average daily deployed:         $63,878
  Working capital required:       ~$10,000-$20,000
    (capital cycles within 5-15 min windows; peak instantaneous exposure
     is a fraction of the cumulative monthly notional)

At filtered configuration (0.30-0.70 band):
  16-day P/L: +$7,923 on $499,735 deployed
  Monthly run rate: ~$15,000 P/L on ~$940K monthly deployed
  ROI per dollar deployed: +1.59% per 16 days = +3.0% per 30 days
  
Required working capital: ~$10,000
  (30-day notional of ~$940K cycles through ~$10K at any moment
   since each position turns over in 5-15 minutes)

The capital efficiency of this strategy is extremely high on paper. $10K of working capital cycles through $940K of annual notional because each dollar is redeployed every 5-15 minutes. The cash-on-cash ROI on actual working capital is therefore much higher than the 1.59% ROI per dollar deployed suggests:

Working capital: $10,000
Daily cycles: ~400 markets × avg 5min resolution = 400 × 12/hr × ~18 active hrs
  ≈ but constrained by actual fill rate and clip size

More practically: $64K deployed per day / $10K working capital = 6.4x daily turnover
Monthly: $1.9M notional / $10K capital = 190x annual turnover multiple

---

7. Hour scheduling

The bot has no hard sleep requirement. The reference wallet runs all 24 hours. The only genuine dead zone is 14:00-16:00 UTC where the US equity open creates correlated volatility that degrades the directional signal. The overnight reduction is optional - the overnight hours are profitable, just lower-volume.

Saturday operations: Saturday shows +2.05% ROI vs +0.42% baseline - nearly 5x better. Do not skip Saturdays. This is the highest-alpha single day of the week, likely due to reduced competition from professional market makers. Weekend hours 12:00-20:00 UTC should be prioritized.

---

8. Risk profile

---

9. Diagnostic checklist: is the bot still working?

Run weekly:

---

10. What this playbook deliberately does NOT include

No SELL engine. Unlike SirMartingale, NeverSmiling derives no P/L from active exit management. Adding a SELL engine to this strategy changes its fundamental nature and introduces execution complexity that the thin-margin structure cannot absorb. The spread is locked in at entry; settlement is the correct and only exit.

No Kelly sizing. The spread-capture mechanism does not have a calculable per-trade win probability that Kelly can optimize against. Flat clips scaled by paired cost and conviction is the correct approach. Kelly would require knowing the exact fair probability of each outcome - if you knew that precisely enough to Kelly-size, you would not need the hedge leg at all.

No high-conviction single-side bets. The reference wallet places 0 trades that are pure directional with no hedge. Even at 3x+ dominance, it buys the non-dominant side. This is correct: the non-dominant side occasionally wins (at 21.2% frequency at 3x+ dominance), and those wins are cheap to buy ($0.20-$0.30 per share at a 3x-dominant market). Removing the hedge transforms this into a directional strategy with higher per-win payouts but also higher per-loss costs. That is a different strategy.

No price targets above $0.70 or below $0.30. The filter analysis is definitive: the $0.70-$1.00 zone loses money systematically. The $0.00-$0.30 zone earns positive ROI but the absolute contribution ($3,943 earned) is smaller than what the $0.70-$1.00 zone loses ($6,461 lost) even though both zones have roughly similar capital allocations. By removing both, the $0.30-$0.70 band captures the sweet spot where the spread-capture mechanism and the directional signal both operate efficiently.

No market-type expansion beyond crypto Up/Down. The paired-cost mechanism requires outcomes that sum to exactly $1.00, fast resolution windows, and continuous refreshing of new markets. Sports markets have these properties but lack the spot-tape fair-value model that drives the directional signal. Do not attempt to port this strategy to sports without a sport-specific fair-value model.

---

TL;DR - implementable structure

# NeverSmiling replication - pseudocode outline
# Full implementation requires Polymarket CLOB WebSocket + spot data feeds

async def run_neversmiling():
    spot_feeds = await connect_spot_feeds(["BTC","ETH","SOL","XRP"])
    clob = await connect_polymarket_clob_ws()
    
    while True:
        # Skip US open
        if utc_hour(now()) in (14, 15):
            await sleep(60)
            continue
        
        for market in active_markets_matching(
            ["btc-updown-5m", "btc-updown-15m", "eth-updown-5m",
             "sol-updown-5m", "sol-updown-15m", "xrp-updown-5m", "xrp-updown-15m"]
        ):
            if not is_eligible_market(market):
                continue
            
            # Compute fair value and directional signal
            spot = spot_feeds.latest(market.asset)
            fair_prob_up = compute_fair_prob(market, spot)
            gap = fair_prob_up - market.up_side.mid_price
            
            # Compute paired cost
            paired_cost = estimate_paired_cost(market, target_usdc=10)
            
            # Skip markets with poor paired cost and weak signal
            if paired_cost > 1.00 and abs(gap) < 0.15:
                continue
            
            # Compute allocation split
            dom_frac = 0.50 + min(0.30, abs(gap) * 1.5)  # 50% to 80% dominant
            total_clip = compute_clip_size(paired_cost, abs(gap), available_capital())
            
            if gap > 0:  # Up favored
                up_budget = total_clip * dom_frac
                down_budget = total_clip * (1 - dom_frac)
            else:        # Down favored
                up_budget = total_clip * (1 - dom_frac)
                down_budget = total_clip * dom_frac
            
            # Enter both sides - only in $0.30-$0.70 band
            await walk_book_buy(market, "Up",   up_budget,   min_price=0.30, max_price=0.70)
            await walk_book_buy(market, "Down", down_budget, min_price=0.30, max_price=0.70)
            
            # No exit logic needed - settle automatically at resolution
        
        await sleep(1)

# Expected at $10K working capital:
#   ~400 markets/day, ~$35-50K daily deployed, +1.5-2.0% ROI per cycle
#   Saturday best day, avoid hours 14-16 UTC
#   Monitor paired cost weekly - strategy degrades if mean > 0.99

Run 24/7 except hours 14-16 UTC. Reconcile daily against expected paired-cost and win-rate diagnostics. The strategy earns through discipline and volume, not through any single spectacular trade.