BTC5MScour

Wallet: 0x68f6c51dbb0ecb12ee66a937434e6190c9f19533 Window: 2026-06-10 to 2026-07-06 (27 calendar days, 26 active) Universe: 23,638 trades across 2,744 markets, all btc-updown-5m-* Account P/L (Polymarket verified): +$26,520.22 on $1,497,961.61 deployed = +1.77% ROI

The Punchline

This is a near-certainty scouring bot that operates exclusively on Bitcoin 5-minute Up/Down markets. The strategy is not directional prediction. It is not market-making. It is not latency arbitrage in the spot-to-CLOB sense. It is something simpler and more mechanical: lift every stale offer priced below rational fair value in the final seconds of a 5-minute BTC market window.

With 98.8% of capital deployed at prices of $0.90 or above, the wallet is buying outcomes that are already nearly certain and collecting the $0.01-$0.04 per-share spread between entry price and $1.00 settlement. Do this 22,000 times and the aggregate P/L is reliable, if thin. The 1.77% ROI on $1.5 million deployed sounds unimpressive until you remember the capital cycles multiple times per day inside 5-minute windows. The actual working capital is a fraction of the gross deployed notional.

There is one departure from pure spread-capture: a concentrated cluster of trades on June 21 between 2:25AM and 3:40AM ET that generated over $13,000 in P/L across nine consecutive 5-minute markets. During this window, the bot entered at prices as low as $0.10-$0.40, capturing enormous per-share gains on correct resolutions. Whether this represents a second signal mode or fortunate positioning in a volatile BTC move is the central unanswered question in the dataset.

The bottom line: +$26,520 earned primarily by being faster than other bots at lifting residual stale offers in the final moments of BTC 5-minute windows, with a secondary payoff from correctly-positioned deep-value entries during at least one episode of significant BTC volatility.

---

What This Wallet Trades

The universe is precisely one market type: btc-updown-5m-*. All 23,638 trades, all 2,744 markets, all in Crypto, all Bitcoin 5-minute Up/Down. This is the most concentrated universe in the PR&R dataset. No ETH, no 15-minute or hourly BTC, no other assets, no other categories.

btc-updown-5m-*    23,638 trades   $1,497,961 BUY notional   2,744 unique markets
All other slugs         0 trades   $0

Within that universe, the bot touches almost every 5-minute window that opens during its operating hours. The 2,744 unique markets over 26 active days equals roughly 105 unique markets per day, which maps approximately to 105 separate 5-minute windows per active day: the bot is present in essentially every available market each day.

Entry price is the most revealing dimension:

The $0.90-$1.00 band is 98.5% of trades and 98.6% of capital. This is the spread-capture engine. The other bands together represent 1.2% of capital and, paradoxically, contribute more P/L per dollar than the core engine.

---

The Order of Operations - One Market, Trade by Trade

The clearest single-market trace of the standard pattern is *Bitcoin Up or Down - July 1, 6:40AM-6:45AM ET*. This is a routine market with 20 fills, all on the Down side, all resolving as wins. It illustrates the typical near-certainty sweep.

Resolution: Down wins. All 20 fills resolve as wins. Total deployed: $1,388.21. Total returned at $1.00/share: roughly $1,388.21 plus spread. Approximate P/L: +$27.50 on $1,388.21 = +1.98% for a single 5-minute window.

Walk-through:

1. 10:43:19 UTC - The market opened at 10:40 ET (14:40 UTC). At 10:43:19, there are roughly 40 seconds remaining in the window. The bot sees Down-side offers available at $0.97. It sweeps 250 shares for $242.50 in two fills. The Down side at $0.97 implies 97% probability. The bot's assessment is that the "Down" resolution is very near-certain, making $0.97 a good buy.

1. 10:43:26 through 10:43:44 - Over the next 25 seconds, the bot walks up the orderbook, lifting offers at $0.97, $0.98, and eventually $0.99. Multiple fills in the same second indicate the bot is fanning out across available resting asks simultaneously.

1. 10:44:04 through 10:44:08 - The bot continues at $0.99. These are the final fills before resolution at 10:45 UTC. Total exposure has reached $1,388.

1. Resolution - All 20 fills resolve as wins. The $0.01-$0.03 spread between entry and $1.00 payout is realized on every share.

This pattern repeats roughly 105 times per day. The bot identifies which side of the current 5-minute market is going to resolve correctly (based on real-time BTC price vs the reference threshold), then sweeps the orderbook on that side at whatever price is available between $0.96 and $0.99.

---

The June 21 Anomaly - A Second Mode

The June 21 early morning cluster is qualitatively different from the spread-capture baseline. The best 10 markets by P/L are overwhelmingly from this two-hour stretch:

The June 21, 2:55AM-3:00AM ET market generated +$9,523 on $1,464 deployed, a return of 650%. This is only possible if a very large portion of those 285 fills were at very low prices (sub-$0.20) and resolved as winners. Cross-referencing with the worst markets list: June 21, 3:05AM-3:10AM ET was a complete loss at -$2,460, and June 21, 1:50AM-1:55AM ET lost -$1,932. The bot was buying both Up and Down across adjacent windows at low prices, and the aggregate was massively positive because it was on the right side during a large BTC directional move.

The interpretation: During the June 21 3AM ET window, BTC made a large move. The bot (or its operator) entered heavily on the correct side at prices reflecting genuine uncertainty ($0.10-$0.40) across multiple consecutive windows. The markets where the correct call was made paid $1.00 per share, generating 250-900% per-share returns. The markets where the wrong side was bought lost their entire stake. The net across the cluster was +$13,000+ because the correct positions were sized larger or the win/loss ratio within the cluster was favorable.

This is not pure spread-capture behavior. It suggests a secondary signal mode that activates during high-volatility BTC periods, where the bot shifts from near-certainty sweeping to directional positioning at lower prices.

---

Why It Works - The Math

The spread-capture core is mechanically positive EV whenever the entry price is below the true resolution probability:

<pre><code>Standard sweep trade (Near-certainty mode): Entry price: $0.97 (market implies 97% win probability) True win probability: ~99.3% (actual win rate at $0.90-$1.00 band) EV per share: 0.993 × ($1.00 - $0.97) - 0.007 × $0.97 = 0.993 × $0.03 - 0.007 × $0.97 = $0.0298 - $0.0068 = +$0.0230 per share

At median entry $0.97: Gross per-trade EV: +$0.023 per share Capital per fill: ~$63 (mean USDC) Shares per fill: ~$63 / $0.97 = ~65 shares EV per fill: ~$1.49

23,636 fills × $1.49 expected EV = ~$35,200 expected

Realized: +$26,291 (trading P/L) - below simple EV expectation because losing markets (0% win rate, complete losses) consume some of the tail: 994 complete losses × ~$63 avg = ~$62,600 lost</code></pre>

The math works because the bot's 99.3% win rate in the $0.90+ band exceeds the entry price probability by 1.5-2.3 percentage points on average. That gap - the bot's information advantage about whether a market is truly near-certain - is the entire source of the spread-capture edge.

The secondary June 21 mode has very different math: entries at $0.10-$0.20 with realized win rates far above the implied probabilities (58% actual win rate vs 10-20% implied) suggest either advance knowledge of the resolution direction or a momentum signal that predicted the BTC move correctly.

---

Phase 1 - Trader Profile

The size distribution is tightly bounded with an unusual bimodal shape. The median is $34 but the P95 and P99 are nearly identical ($242.50 and $247.50), suggesting the bot has a hard limit around $250 per fill with a cluster of fills also at a smaller clip size around $30-100. The max of $878 is a single outlier (roughly 3.5x the P99), likely a fill in the June 21 anomaly cluster.

The top-5% share of capital at 23.1% is extremely low - far below a power-law distribution. This confirms that the bot spreads capital uniformly across many fills rather than concentrating into conviction bets.

Inter-trade gap analysis:

This is fully automated. No manual trader operates at 1-second median inter-fill gaps across 23,000+ trades.

The bot trades all 24 hours with no dead zone:

Hour 06 UTC is the single highest absolute P/L hour (+$10,440) and hour 07 UTC is the single worst (-$3,316). Both are during the June 21 anomaly window. Removing June 21 from hours 06 and 07 UTC would likely normalize both close to the other hours' ~+$1,000-$1,200 range.

---

Phase 2 - Core Strategy Identification

Both-sides participation rate: 0.26% (7 of 2,744 markets). For practical purposes, this is zero. The wallet is directional per market in 99.74% of cases.

This is not a market maker. It is not a copy trader. It is not a DCA accumulator. It is a C: near-certainty scouring bot operating exclusively in the final window of short-duration markets, with episodic elements of B: directional betting during high-volatility BTC periods.

The 7 both-sides markets have a median paired cost of $1.05 (above $1.00), meaning those paired positions were not spread captures - they were likely accidents or hedges. The 3.0x+ dominance bucket shows 5/5 dominant-side wins with a mean paired cost of $0.877, but the sample of 7 markets is too small to draw conclusions.

---

Phase 3 - Dominance Ratio Analysis

The 3.0x+ bucket shows 5/5 dominant-side wins at a mean paired cost below $1.00 - technically these are genuine spread captures with directional tilt. But 7 total both-sides markets out of 2,744 is statistically negligible. The dominance framework does not describe this trader; it describes an edge case.

---

Phase 4 - Entry Price Analysis

The sub-bucket view at the per-cent level confirms the hypothesis: the $0.97, $0.98, and $0.99 price points carry the overwhelming majority of the $0.90+ capital. The CSV sample shows nearly every fill in the $0.90+ range landing at $0.97, $0.98, or $0.99, with the distribution shifting toward $0.99 in markets with 2-3 minutes remaining and toward $0.97 in markets with 4+ minutes remaining. The bot is price-walking the orderbook from its current best ask down to $0.96, lifting all available offers.

The anomalous lower-band fills ($0.10-$0.50) appear during high-volatility windows where the market's probability is genuinely uncertain. The $0.10-$0.20 band's 58.4% actual win rate against 15% implied probability is the single most anomalous data point in the dataset - it suggests either insider information, a reliable technical signal, or very fortunate timing during the June 21 BTC move.

---

Phase 5 - Category and Vertical Breakdown

Single-category book. The ROI of +1.75% on crypto trading appears low but is consistent with a high-velocity near-certainty strategy where each fill earns 1-3 cents per dollar deployed.

There is no sub-category breakdown possible (single asset, single duration). The within-category analysis reduces to time-of-day and per-market patterns already covered.

---

Phase 6 - Timing and Execution

Hourly P/L (UTC)

The 06 UTC and 07 UTC outliers are both artifacts of the June 21 anomaly cluster. Hours 04, 10, 16, 18, 19, 20 all show 100% win rates - these are hours where the near-certainty engine fired consistently with no bad fills.

The bot has no sleep window. All 24 hours have trades. This is fundamentally different from SirMartingale's US-session scheduling. The BTC 5-minute market runs continuously and the bot participates continuously.

Day-of-week patterns

Sunday's 78.9% win rate is dramatically lower than every other day (all above 95%). Yet Sunday generates the highest ROI at 4.12%. This is the June 21 effect: June 21 was a Sunday, and the early-morning anomaly cluster with its mixed wins and losses at low entry prices drives down the raw win rate while inflating ROI.

Friday's +$8,024 P/L and Sunday's +$7,944 together account for 60% of total P/L while representing only 31% of total trades.

---

Phase 7 - Filter Experiments

Full filter commentary is in the Filters tab.

---

Phase 8 - Rolling Window Consistency

Week 25's lower win rate (89.9% vs 97-99% in other weeks) is the June 21 anomaly pulling the average down. Week 27 at 99.3% win rate is the baseline near-certainty engine at its cleanest.

100% of all rolling 7-day and 15-day windows close green. The cumulative P/L line in the daily data is monotonically positive except for a small dip on June 24 (cumulative drops from $18,239 to $16,659) and June 28 ($19,790 to $19,583). The strategy has no extended losing streaks.

---

Phase 9 - P/L Decomposition

This decomposition is unusual: the SELL leg is essentially absent. The wallet holds 23,634 positions to settlement. There is no exit management, no SELL engine, no active position management. The bot buys near-certainties and waits. This is the structural inverse of SirMartingale, who sells aggressively before settlement.

The $229.54 in rewards/other is minimal relative to trading P/L (+$26,291) - this is a trading-P/L-driven wallet, not a liquidity farming operation.

The hedge tax ($2,144.93) refers to the losing sides of the 7 both-sides markets. This is trivial relative to total P/L.

---

Phase 10 - Strategy Specification

One-sentence summary: A near-certainty sweep bot that identifies the resolving side of Bitcoin 5-minute Up/Down markets in their final 60-120 seconds, then lifts all available orderbook offers between $0.96 and $0.99 on that side, collecting the residual spread at settlement.

Edge source (primary): Information advantage about resolution direction in the final seconds of 5-minute markets, enabling purchase of offers priced below true resolution probability.

Edge source (secondary, episodic): Directional positioning during high-volatility BTC periods when the correct side is available at $0.10-$0.40, generating 250-900% per-share returns on correct calls.

What works: Entering the $0.90-$0.99 band when resolution direction is known. 99.3% win rate. The secondary deep-value mode when active.

What drags: Complete losses on wrong-side entries (994 losing markets, total -$X in losses). The $0.60-$0.90 band is negative ROI (-$1,096 combined) suggesting the bot occasionally enters at mid-range prices where the signal is weaker.

Full implementable spec is in the Playbook tab.

Wallet: 0x68f6c51dbb0ecb12ee66a937434e6190c9f19533 Window: 2026-06-10 → 2026-07-06 (26 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: 0.26% (7 of 2,744 markets)
• Median paired cost: $1.0515
• Mean paired cost: $1.1843
• Paired cost % under $1.00: 42.9%
• Paired cost % under $0.97: 42.9%
• Median 2nd-side hedge lag: 102s

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: +$751 → +$11,662
• Rolling 15-day windows green: 27 of 27 (100.0%)
• Rolling 15-day P/L range: +$751 → +$18,031

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 2026-07-09 09:45 UTC.

Wallet: 0x68f6c51dbb0ecb12ee66a937434e6190c9f19533 Window: 2026-06-10 to 2026-07-06 Baseline: 23,636 resolved BUYs · 95.79% WR · $1,497,962 deployed · +$26,291 trading P/L · +1.75% ROI Account P/L (Polymarket verified): +$26,520.22

---

The headline result

One filter produces spectacular-looking results. It is not replicable. The combined stack (high-conviction dominance + exclude worst hours) hits 94% ROI on 38 trades with $1,777 deployed, but that $1,671 P/L comes almost entirely from the June 21 anomaly cluster. Excluding those four worst hours from the book removes the volatile outlier markets where the bot entered at deep-value prices ($0.10-$0.40) and collected enormous per-share gains. The filter looks like a brilliant refinement; it is actually just a lens for viewing the June 21 event.

The price-band filter is the most important verdict: applying the standard "$0.30-$0.70 sweet spot" cuts 98.5% of the book's capital and returns only the 502 mid-range fills that happen to have a higher ROI per dollar - but absolute P/L collapses from +$26,291 to +$1,143. The filter does not identify a portable alpha; it isolates a tiny sample that includes several June 21 anomaly fills.

The most useful filtering insight for this wallet is negative: the strategy cannot be meaningfully improved by any of the standard filters because the capital is already concentrated where the edge is.

---

Filter results table

---

Filter-by-filter commentary

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

Applying this filter reduces the qualifying population from 23,636 trades to 502. The 502 trades represent 1.0% of all trades and 1.1% of all capital. P/L collapses from +$26,291 to +$1,143, a reduction of -$25,148.

The filter looks attractive superficially - ROI jumps from 1.75% to 7.27%. But this is selection bias, not a genuine improvement. The 502 mid-range fills include some of the June 21 anomaly deep-value entries (the $0.40-$0.50 band with 205 trades returning +24.7% ROI and the $0.50-$0.60 band with 94 trades returning +72.8%). Strip those out and the remaining mid-range fills are often negative (the $0.60-$0.70 band loses -$2,623 at -56.2% ROI, and the $0.80-$0.90 band loses -$160).

More fundamentally: the strategy's operating mode is $0.90-$0.99 entry. Applying a $0.30-$0.70 filter removes 98.6% of the strategy and keeps a tiny residual that does not represent any coherent sub-strategy. This filter destroys, not refines.

2. High-conviction dominance filter (dom ≥ 2x, dominant leg only) → MISLEADING

This filter surfaces 111 trades across the 5 both-sides markets with dominance ratio ≥ 2x (all in the 3.0x+ bucket). Win rate is 97.3%, ROI is +46.8%, and absolute P/L is +$2,594. This looks excellent.

It is not replicable. The 7 both-sides markets in this book were not the result of a deliberate market-making strategy - they appear to be incidental cases where the bot entered both sides of the same market during the June 21 volatility cluster. The 3.0x+ bucket's 5/5 dominant-side wins and $0.877 mean paired cost are the footprint of the June 21 anomaly, where the bot happened to buy more shares on the winning side than the losing side.

Applying this filter as a forward-looking selection rule would require deliberately engineering both-sides entries with 3.0x+ dominance, which is not the base strategy. The filter is not applicable as a replication guide.

3. Category filter → NO-OP

100% of trades are Crypto. The filter is identity-equivalent to baseline. No P/L change, no win rate change.

This is expected for a single-category book. The filter adds no information.

4. Hour exclusion filter (exclude hours 0, 5, 6, 7 UTC) → DESTRUCTIVE

The four "worst" hours are 00 UTC (-$420 P/L), 05 UTC (listed as worst despite +$1,147 P/L), 06 UTC (+$10,440 P/L), and 07 UTC (-$3,316 P/L).

Excluding these four hours removes 4,799 trades and $263,243 of deployed capital from the qualifying set. P/L drops from +$26,291 to +$18,439. The filter costs -$7,852 in P/L.

The mechanism is straightforward: hour 06 UTC is the single highest-P/L hour in the book at +$10,440, driven almost entirely by the June 21 anomaly. Any filter that identifies hour 06 as a "worst" hour is misclassifying it. The filter algorithm apparently ranks hours 0, 5, 6, and 7 as worst based on some combination of win rate and raw P/L - hour 07 at -$3,316 is genuinely bad, but hour 06 at +$10,440 is the second-best performing hour in the book.

The hour filter should not be applied here. There are no genuinely bad hours to exclude: the anomalous behaviors (hours 06 and 07 on June 21) are tied together and cannot be separated by a simple hour-exclusion rule.

5. Combined stack filter → NOT APPLICABLE

The combined filter (high-conviction dom ≥ 2x + exclude worst hours) produces 38 trades with 84.2% win rate, $1,777 deployed, and +$1,671 P/L (+94.0% ROI).

The 94% ROI headline is eye-catching. It is not a replicable edge. The 38 qualifying trades are the intersection of the 7 both-sides markets (after hour exclusion) that happened to include some large-position June 21 anomaly fills on the dominant side. This is a sample of 38 trades selected retrospectively from a 27-day window - it describes what happened, not a forward-looking entry rule.

Do not interpret 94% ROI on 38 trades as a viable sub-strategy.

---

What filters would actually help

The standard battery misses the genuine exploitable dimensions of this wallet. Useful refinements would operate on different axes:

The most actionable insight from filter analysis: the $0.60-$0.90 entry band should be avoided. It is neither near-certainty sweep territory ($0.90+) nor genuine deep-value territory ($0.10-$0.40). The 244 fills in this band deploy $8,635 and generate -$473 net - a -5.5% ROI that drags the overall book. A replicator should configure the bot to skip offers in the $0.60-$0.90 range and focus exclusively on $0.90-$0.99 sweeps plus deliberate deep-value entries with a separate signal.

---

Bottom line for replication

Three concrete filter recommendations:

1. Do NOT apply the $0.30-$0.70 price filter. It removes 98.5% of the strategy's operating capital for a superficially attractive ROI improvement that is entirely selection-bias from June 21 anomaly fills.

1. Do NOT apply the hour exclusion filter. The four "worst hours" include hour 06 UTC, which is the highest-P/L hour in the book. Excluding it costs -$10,440 of P/L.

1. DO consider excluding the $0.60-$0.90 entry band. These 244 fills are between the near-certainty zone and the deep-value zone, and they generate negative ROI (-5.5%). A bot that skips offers in this price range and focuses on $0.90-$0.99 (and separately on $0.10-$0.40 with a volatility signal) would likely outperform the current configuration.

The single most useful "filter" you can apply is a band gap: operate at $0.90-$0.99 for spread capture, operate at $0.10-$0.40 when a volatility trigger fires, and skip the $0.40-$0.90 middle zone entirely.

Source wallet: 0x68f6c51dbb0ecb12ee66a937434e6190c9f19533 Strategy: Near-certainty scouring on BTC 5-minute Up/Down markets with episodic deep-value mode during high-volatility BTC periods Reference book: $1,497,962 deployed → +$26,291 trading P/L → +$26,520 account total → +1.77% ROI in 27 days

---

One-paragraph operator brief

Build a Polymarket bot that monitors the live BTC price against the resolution threshold for every active btc-updown-5m-* market. In the final 60-120 seconds of each window, when the resolution direction is already determined by the current BTC spot price, sweep the orderbook on the resolving side from $0.96 up to $0.99, lifting all available offers. Hold all positions to settlement at $1.00. Run 24 hours per day, 7 days per week - the BTC 5-minute series never sleeps. Skip offers priced between $0.60 and $0.90 (bad ROI band). Optionally, activate a secondary deep-value mode during high-volatility BTC periods to enter at $0.10-$0.40 on the direction-signaled side when implied probability and spot price diverge significantly. Expect roughly 105 market entries per day with $450-$600 average capital per market, generating +$8-$15 net P/L per market on the spread-capture mode alone.

---

1. Market selection

The wallet demonstrates that 2,744 markets over 26 active days = roughly 105 markets per active day in the btc-updown-5m series. Polymarket posts approximately 288 new 5-minute BTC windows per day (one every 5 minutes). The bot is participating in roughly 36% of all available windows - it is selectively entering markets where its conditions are met, not running in every market unconditionally.

The entire P/L comes from this one series. No diversification into ETH, 15-minute windows, or other categories will help or is needed.

---

2. Entry logic

The core entry signal is resolution certainty: BTC spot price has moved past the market's threshold price with enough margin that resolution is not in doubt before the window closes.

def should_enter(market, btc_spot_now):
    # Slug whitelist
    if not market.slug.startswith("btc-updown-5m-"):
        return None

    # Timing - only enter in final 60-120 seconds
    sec_left = seconds_until_close(market)
    if sec_left > 120 or sec_left < 5:
        return None  # too early or too late

    # Price band - avoid the no-man's land
    # Only enter if we'll pay $0.90-$0.99 OR signal justifies $0.10-$0.40
    clob_best_ask_up = market.up_side.best_ask
    clob_best_ask_down = market.down_side.best_ask

    # Spot certainty check
    threshold = market.btc_threshold_price
    margin_pct = abs(btc_spot_now - threshold) / threshold

    if margin_pct < 0.0005:   # within 0.05% of threshold = ambiguous
        return None

    direction = "Up" if btc_spot_now > threshold else "Down"
    best_ask = clob_best_ask_up if direction == "Up" else clob_best_ask_down

    # Standard near-certainty mode: only enter at $0.90-$0.99
    if 0.90 <= best_ask <= 0.99:
        return direction

    # Deep-value mode: enter at $0.10-$0.40 if volatility flag is set
    if 0.10 <= best_ask <= 0.40 and volatility_flag_active():
        return direction

    # Skip $0.40-$0.90 range entirely (negative ROI band)
    return None

---

3. Exit logic

There is no exit logic. All positions are held to settlement.

The reference wallet has only 2 SELL transactions across 23,638 BUY transactions. The strategy is to buy near-certainties and collect $1.00 at settlement. No active position management is needed or productive.

def manage_position(position):
    # Do nothing. Every position settles at $1.00 (win) or $0.00 (loss).
    # The entry-price discipline handles the risk; exits would only add
    # transaction costs and reduce effective ROI on thin $0.01-$0.03 spreads.
    pass

Why not sell at $0.99 to "lock in" the gain? The market is about to settle at $1.00. The incremental gain from selling at $0.99 vs waiting for settlement is $0.01 per share. The transaction cost and execution risk of an active sell in the final 10 seconds is not worth $0.01/share.

---

4. Sizing model

The reference wallet deploys $63.38 mean per fill and $34.08 median per fill across markets averaging 8-9 fills each at $450-$600 total per market. The maximum per-market deployment visible in the top-markets table is roughly $4,400 ($4,410 for the June 21, 11:00PM market with 69 fills).

Note that each 5-minute market settles and capital is returned within 5 minutes. Instantaneous exposure is only a fraction of daily deployment: at any given moment, the bot has capital at risk in at most 2-3 concurrent markets. The daily throughput figure is the sum of cycling the same capital many times.

The sizing within each market should walk the orderbook from the current best ask up to $0.99, deploying the target per-market clip across multiple fills at different price levels. The reference shows 8-20 fills per market across prices ranging from $0.96 to $0.99.

def compute_clip_size(bankroll, market_depth):
    per_fill_base = bankroll * 0.003    # 0.3% of bankroll per fill
    per_fill_max = bankroll * 0.015     # 1.5% max per fill
    per_market_target = bankroll * 0.05 # 5% of bankroll per market total
    return min(per_fill_base, per_fill_max), per_market_target

---

5. Bankroll math

The reference book mechanics extrapolated:

Daily activity (from 26-day observation):
  Markets entered per day:      ~105 unique 5-min windows
  Fills per market:             ~8-9 fills (mean)
  Capital per market:           ~$546 (mean, = $1,497,962 / 2,744)
  Capital cycled per day:       ~$57,229 ($1,497,962 / 26 days)
  P/L per day:                  ~$1,011 ($26,291 / 26 days)
  Average P/L per market:       ~$9.58 ($26,291 / 2,744)

Effective ROI per capital cycle:
  $9.58 P/L / $546 per market = +1.75% per cycle
  Capital recycles ~12× per day (105 markets × $546 / $57,229 effective)

Monthly extrapolation (30 days):
  Capital cycled monthly:       ~$1,717,000
  Expected trading P/L monthly: ~$30,000
  Required peak working capital: ~$5,000-$10,000
  (because capital returns within 5 min; cycling capital, not locking it up)

The ROI figure of 1.77% per cycle is the per-capital-cycle return, not the annualized return on standing capital. With a $10,000 bankroll and aggressive cycling, the effective monthly ROI on standing capital can be much higher than the per-cycle headline suggests.

On $5,000 working capital:
  Can support ~9 concurrent markets × $546 each = $4,914 instantaneous
  Monthly net P/L at same efficiency: ~$15,000
  Monthly ROI on $5,000 standing capital: ~300%

On $25,000 working capital:
  Monthly net P/L at same efficiency: ~$75,000
  Monthly ROI on $25,000 standing capital: ~300%

Above $25,000 per-market deployment, orderbook depth on most BTC 5-minute markets will thin out. The reference book's mean $546 per market is well within the observed liquidity. Scaling to 5x the reference deployment per market may encounter adverse price impact at the $0.96-$0.97 level.

---

6. Hour scheduling and operational rhythm

Unlike SirMartingale's US-session bot, this strategy runs 24 hours per day, 7 days per week. BTC 5-minute markets open every 5 minutes continuously. The reference wallet has trades in every hour of the day with no dead zones.

The only scheduling consideration: if you are running the bot manually or with a cron-style scheduler, ensure it recovers from outages within 5 minutes (before the current market window closes). An outage of one market window costs only the missed P/L for that window (~$9.58 expected).

---

7. The deep-value secondary mode

The June 21, 2:55AM-3:00AM ET market (+$9,522 P/L on $1,464 deployed = 650% return) is the strongest signal that the bot has a second operating mode. The conditions appear to be:

1. BTC makes a large directional move (1%+ in a 5-minute window or less)
2. The market's resolution probability shifts rapidly from ~50/50 to near-certain
3. The bot enters during the transition - when prices are still at $0.10-$0.40 - rather than waiting until the market fully prices in the move at $0.96-$0.99

Implementing this mode requires:

def volatility_flag_active(btc_data):
    # Compute rolling 5-minute realized vol on BTC
    realized_vol_5m = compute_realized_vol(btc_data, window_minutes=5)
    
    # Flag is active when 5-minute vol exceeds 2x its trailing 60-minute average
    baseline_vol = compute_realized_vol(btc_data, window_minutes=60)
    return realized_vol_5m > (2.0 * baseline_vol)

def deep_value_entry_size(bankroll, entry_price):
    # Smaller position than standard mode: 0.5% of bankroll per fill
    # The variance is higher so size down
    return bankroll * 0.005

This mode is optional. The base strategy (near-certainty sweeping alone) generates consistent positive P/L. The deep-value mode adds high-variance upside during BTC volatility events but introduces complete-loss risk on individual positions.

---

8. Risk profile

The strategy is structurally bounded-loss per market. The maximum possible loss on any single 5-minute market is the capital deployed in that market ($500-$1,500 typical). There is no uncapped downside because positions are binary: they pay $1.00 (win) or $0.00 (loss).

---

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

Run weekly:

---

10. What this playbook deliberately does not include

• No prediction model. This strategy requires no forecast of where BTC is going. It only needs to know where BTC is right now relative to a 5-minute resolution threshold. Any attempt to add a directional prediction layer will introduce complexity without benefit - the edge is that the market is already resolved, not that you predicted it.

• No SELL engine. The reference wallet has 2 SELLs out of 23,638 trades. Active exit management would consume the $0.01-$0.03 per-share spread in transaction costs and execution slippage. Hold to settlement.

• No $0.40-$0.90 band entries in standard mode. This band has negative ROI in the reference book. The bot has no reliable signal for mid-range probabilities. The deep-value mode at $0.10-$0.40 during volatility is different - there the implied probability is far below the actual resolution probability, creating genuine edge.

• No position sizing by conviction. All standard-mode entries are sized uniformly at 0.3% of bankroll per fill, regardless of the certainty margin. A spot price 2% past the threshold is not meaningfully more certain than one 0.1% past - both resolve the same way.

• No multi-asset diversification. The reference wallet is 100% BTC 5-minute, period. ETH markets have different threshold dynamics and liquidity profiles. Attempting to replicate the same strategy on ETH 5-minute markets without separate calibration may not produce the same win rates.

• No copy-trading. Each 5-minute market is self-contained. There is nothing to copy from another wallet; by the time you could observe another wallet's entry, the market is likely within 30 seconds of resolution.

• No Kelly sizing or Martingale. The strategy's per-fill EV is small and consistent. Optimal sizing is a small fixed clip, not a volatility-adjusted Kelly fraction or a loss-chasing doubling scheme.

---

TL;DR - implementable in ~100 lines of Python

async def run_sweep_bot():
    btc_feed = await connect_btc_spot_websocket()       # real-time BTC mid-price
    clob_feed = await connect_polymarket_clob_ws()       # L2 for btc-updown-5m markets
    
    while True:
        for market in active_markets("btc-updown-5m"):
            sec_left = seconds_until_close(market)
            if not (5 < sec_left < 120):
                continue
            
            # Identify resolution direction
            btc_price = btc_feed.latest()
            threshold = market.resolution_threshold
            margin = abs(btc_price - threshold) / threshold
            
            if margin < 0.0005:    # too ambiguous
                continue
            
            direction = "Up" if btc_price > threshold else "Down"
            best_ask = market.side(direction).best_ask
            
            # Standard near-certainty mode
            if 0.96 <= best_ask <= 0.99:
                clip = min(bankroll * 0.003, 65.0)
                await walk_book_buy(market, direction, max_usdc=clip * 9)
                # Walk from best_ask to $0.99 in multiple fills
                
            # Deep-value mode (optional, requires vol signal)
            elif 0.10 <= best_ask <= 0.40 and btc_vol_elevated():
                clip = bankroll * 0.005
                await market_buy(market, direction, usdc=clip)
            
            # else: skip - no entry in $0.40-$0.90 band
        
        await asyncio.sleep(1.0)
    
    # All positions auto-settle at $1.00 (win) or $0.00 (loss)
    # No exit management needed

Run this 24/7. Monitor daily win rate and average entry price. The strategy earns +1.77% per capital cycle; at 105 cycles per day with fast capital recycling, the effective monthly ROI on working capital is substantial. The edge is mechanical, not predictive: you are buying certainty that the market is still pricing as probability.