LIL222

Wallet: 0xeaca59cb5e10e0be128b005a0f84465d2ed80729 Window: 2026-04-05 → 2026-04-29 (25 calendar days, 25 active) Universe: 34,724 trades · 3,500 markets · 3,500 events · $22,855 gross notional Net cash-flow P/L: +$6,267 on $11,146 deployed = +56.22% ROI in 25 days

The Punchline

LIL222 is the cleanest single-strategy bot in this entire library. There is no blend, no hedging, no directional view, no category rotation, no day-of-week tilt to debate. The whole book runs one rule, executed 21,919 times in 25 days:

That is literally the entire strategy. 100.00% of his 21,919 BUY tickets fired at exactly $0.01. 98.55% of his 12,805 SELL tickets fired at exactly $0.02. No exceptions worth talking about. Not a single buy at $0.02. Not a single buy at $0.005. He sits at the absolute floor of Polymarket's tick grid (1-cent minimum) and waits.

He is functionally a passive maker on the bottom rung of the orderbook of dying penny contracts. He's not predicting Bitcoin. He's not picking sides. He doesn't care if BTC goes up or down, and the data confirms it: his win rate is 1.36%, which is worse than random (random on a binary would be ~50%) - because he is systematically buying the side that's losing. That is on purpose. Losing is the source of his edge.

The trade is structural, not directional. Polymarket's binary contract pays $1.00 to the winner and $0.00 to the loser. So a share you bought at $0.01 has two paths: sell it for $0.02 (a 100% return, the ordinary outcome) or hold it through the rare 1.36% reversal where it pays $1.00 (a 99× return). Both branches are profitable in expectation when entry is rigidly anchored at the floor. Lil222 spends 25 days mechanically harvesting both branches and walks out with 56% on his BUY notional in three weeks of 24/7 operation.

This is not a "smart" strategy in the sense of having a market view. It is infrastructure. He is the limit-order at the bottom of every penny book on btc-updown-5m. Retail panic-sells into him; he flips to retail FOMO at $0.02 or eats the lottery ticket on the rare flip.

What He Trades - The Universe Is One Thing

Of 3,500 markets touched in 25 days, every single one is a Polymarket "Bitcoin Up or Down" 5-minute crypto microstructure market. The Crypto category alone holds 100% of trades, 100% of volume, and 100% of P/L. There is no soccer, no NFL, no politics, no other crypto pair. He doesn't even touch the 15-minute or hourly crypto markets that share the same orderbook architecture - only the 5-minute slice.

btc-updown-5m-1777262400 → "Bitcoin Up or Down - April 27, 12:00AM-12:05AM ET"
          btc-updown-5m-1776907500 → "Bitcoin Up or Down - April 22, 9:25PM-9:30PM ET"
          btc-updown-5m-1777450500 → "Bitcoin Up or Down - April 29, 4:15AM-4:20AM ET"

A new market opens every five minutes, 288 markets per day. Lil222 touches roughly 140 of them per day on average, which means he picks up about half of the available 5-min BTC market inventory. The other half are dead because the orderbook never offers a $0.01 contra-side to lift.

The selection rule is implicit but obvious: only enter when the market has gapped enough that one side is trading at $0.01. Most markets open near $0.50/$0.50 and stay there. They never hit $0.01 on either side. Lil222 ignores those - his algorithm only trips when the BTC tick has carried one side to the floor in the final minute or two of the window.

That makes his average market-touch a bet on mean reversion in the last 90 seconds of a 5-minute crypto market. Not on direction.

The Order of Operations - One Market, Trade by Trade

This is the cleanest single-trade trace in the dataset. April 22, 9:25PM-9:30PM ET market (UTC: 01:25-01:30 on April 23):

Walk through what's happening:

1. 01:28:04 - minute 3 of the 5-minute window. BTC has ticked enough that "Down" is trading at the floor. Lil222 hits the $0.01 bid for 590 shares of "Down" across six fills in the same second (sub-second bot fan-out - note the identical timestamp on six separate transaction hashes).
2. 01:28:12 - eight seconds later. Someone bumps the orderbook. The $0.02 ask is now lift-able. Lil222 immediately sells 391 of his 590 "Down" shares at $0.02. He's already net positive on this market. He still holds 199 "Down" shares as a residual lottery ticket.
3. 01:30:14 - fourteen seconds after the market window closes. The "Up" outcome is about to be declared the winner (the resolution oracle hasn't fired yet but the price action signaled it). Lil222 sees "Up" still being offered at $0.01 by panicking pre-resolution sellers and buys 500 shares of the winning side at $0.01.
4. Settlement. "Up" wins. The 500 "Up" shares pay $1.00 each = $500. The 199 unsold "Down" shares pay $0.00. Net: -$5.81 -$5.00 +$7.81 +$500 = +$497.

That is a $5 ticket returning $500. A 99× gross return on the winning leg, banked in roughly two minutes.

The market resolution is even cleaner in single-fill cases. The April 27, 12:00AM-12:05AM ET market is one trade in the entire CSV: one BUY at 04:04:52 UTC (12:04:52 AM ET - eight seconds before the window closes) for 500 shares of "Up" at $0.01. "Up" wins. $5.00 in, $500 out, +$495 P/L on a single fill.

This is what 296 of his 21,919 BUYs look like (the 1.36% that win). The other 21,623 lose their $0.01 (or get partially recouped via the $0.02 sell). The wins individually are an order of magnitude larger than the losses. The expected value is positive because the 99× payout on wins crushes the 1× loss on misses.

Why It Works - The Penny Math

Treat a single $0.01 share as a probabilistic instrument:

p(sell at $0.02 before close)          ≈ 56%   (12,620 / 21,919 BUY tickets had a SELL counterpart)
          p(hold to settlement & win at $1.00)   ≈ 1.4%
          p(hold to settlement & lose at $0.00)  ≈ 42.6%

The expected value per $0.01 ticket:

EV  =  0.56 × $0.02   +   0.014 × $1.00   +   0.426 × $0.00
              =  $0.0112        +   $0.0140         +   $0
              =  $0.0252

A $0.01 ticket has a $0.0252 expected exit value. That's a +152% expected return per ticket before any other adjustments. Even after cancellation rates, slippage, gas, and the unsold residual that mostly resolves to zero, the realized number lands at +56% net ROI on cash deployed across 25 days. The strategy is not subtle. It is the kind of edge that gets harvested by whoever shows up first with a low-latency bot, and then the edge dies when other bots compete.

LIL222 is one of those bots. The fact that he's still printing 56% in late April 2026 means either (a) the market is too small for serious competition to bother yet, or (b) the orderbook tick at $0.01 is small enough that even with 3-4 bots competing, the absolute scale doesn't attract serious capital. His total 25-day capital deployed is $11,146 - eleven thousand dollars. This is the kind of edge that prints percentage returns but not dollar returns big enough to fund anyone's lifestyle. He's grinding $250/day on a margin that nobody bigger will fight him for.

Phase 1 - Trader Profile

Scale & Activity

• 21,919 BUYs + 12,805 SELLs = 34,724 trades in 25 days
• BUY notional $11,146 · SELL notional $11,709 · gross turnover $22,855
• ~1,389 trades per day · active every single calendar day in the range
• 3,500 unique markets across 3,500 unique events → exactly one market per event (the 5-min BTC schedule produces a fresh event every 5 minutes; he never double-touches the same market)

Trade Size Distribution (capped, not power-law)

Notice the hard ceiling at $5.00 - that's the P99 mark, and the max single fill is $151.29. He's running fixed-size clips of either ~$0.05 (5 shares × $0.01) or ~$5.00 (500 shares × $0.01), with rare outliers when he piles into a single market. There's no scaling-by-conviction here. There's no Kelly sizing. There's no edge-weighted bet. Every market gets the same $5 lottery ticket regardless. The book is uniform-size, which is exactly what you'd expect from a bot that doesn't have a directional model - there's nothing to scale on.

Execution Signature

• Median inter-fill gap on same (market, outcome): 0.0 seconds
• 84.3% of consecutive fills under 1 second · 99.0% under 10s · 100% under 60s
• This is definitive bot territory - sub-second fan-out across multiple transaction hashes in the same second is not human reaction time

The April 22 9:25PM trace above shows six separate BUY transactions all stamped 01:28:04 UTC. That's the bot splitting one logical order across multiple on-chain fills in parallel - atomic multi-leg fan-out, common for penny markets where the orderbook depth at any single price point is shallow.

Archetype

Pure passive maker / penny-longshot harvester. Not directional, not market-making in the both-sides sense, not signal-following. He's the resting bid at the floor of every 5-min BTC market that gaps to $0.01.

Phase 2 - Core Strategy Identification

Both-sides participation: only 1.60%

Of 3,500 markets touched, only 56 had both YES and NO sides bought. This rules out classical market-making (which would show 60%+) and rules out hedge-the-tail strategies. He's almost always one-sided per market: he buys whichever side has gapped to $0.01, not both sides.

Paired cost on the rare both-sides markets: $0.0200

When he does buy both sides, the paired cost is essentially the floor: $0.01 + $0.01 = $0.02. That means a guaranteed payout of $1.00 on whichever side wins - a $0.98 spread per paired share, or a 49× return on paired capital with zero directional risk. These 56 markets generated $4,470 in P/L on tiny capital - about 7% of his book. The both-sides cases aren't his main strategy but when they happen they're free money.

What he is not doing

• Not a market maker (1.6% both-sides rate vs MM threshold of 60%)
• Not directional (1.36% win rate is worse than coin-flip random - by design, because he's buying losing-side tickets when they're cheap)
• Not a stale-price sniper in the classical sense (he doesn't wait for the resolution to be obvious - he enters in the final 90s of an active market and lets noise produce the win)
• Not a copy-trader (every market is a fresh 5-min BTC event; nothing to copy)
• Not DCA (each market gets one cluster of trades in a 2-3 minute window, then he never returns)

Classification

Pure passive penny-longshot harvester. The single closest peer in this library is bonereaper, which runs the same 5-min BTC strategy but with slightly different price targets and a more active SELL profile. LIL222 is bonereaper distilled to its purest form - every parameter except the entry price has been removed.

Phase 3 - Dominance Ratio Analysis

The classical dominance analysis essentially doesn't apply here because only 56 of 3,500 markets have both sides bought. The breakdown:

The dom-side win rate at 1.0–1.5× is 15.1% - well below the 50% you'd expect from a random model. That confirms even on the 56 both-sides markets, he is systematically tilting toward the losing side, because that's the one that's at $0.01. The dominance ratio isn't expressing conviction; it's expressing whichever side gapped further down.

There is no conviction curve here. The strategy doesn't have one. The signal is the orderbook floor, not directional information.

Phase 4 - Entry Price Analysis

100% of his BUY capital lives in the bottom $0.10 price band. Drilling deeper:

• $0.01 exactly: 21,919 of 21,919 BUYs (100.00%)
• $0.02 exactly: 0 BUYs
• $0.03+: 0 BUYs

There is no second band. There is no taper. He is locked to the floor.

Sell-side distribution:

• $0.02 exactly: 12,620 of 12,805 SELLs (98.55%)
• $0.03: 57 SELLs (0.45%)
• $0.04: 51 SELLs (0.40%)
• $0.05+: 77 SELLs (0.60%)

The handful of $0.03+ exits are almost certainly cases where his $0.02 ask got partially filled and the rest had to reprice up as the orderbook moved. The default exit is unconditionally $0.02. He doesn't try to wait for $0.05. He doesn't trail the price up. The strategy is "double-or-nothing": $0.01→$0.02 = 100% gain, take it instantly.

This is the most extreme price-band concentration of any wallet I've ever seen. It is essentially a one-pixel strategy.

Phase 5 - Category & Vertical Breakdown

There is one row. There are no other categories. 100% of trades, 100% of volume, 100% of P/L is in the Crypto vertical, and inside that vertical 100% is 5-minute Bitcoin Up/Down. He doesn't touch any other Polymarket asset class. He has voluntarily chosen the most narrow universe possible.

Phase 6 - Timing & Execution

Hourly P/L (UTC)

The bot runs 24/7, but performance varies meaningfully by hour. Best and worst windows by ROI:

The single best ROI hour is 04:00 UTC (+268%), which is midnight ET - the deep overnight US window when retail liquidity is thin and the orderbook has the widest gaps. The worst is 19:00 UTC (−18%), which is afternoon ET when retail attention is highest and the spreads are tightest. The pattern is consistent: he prints the best returns when nobody else is watching the orderbook.

Only one of 24 hours posts a net loss (19:00 UTC, -$95). The rest are all positive, ranging from +$11 to +$1,001. That's how broad and consistent the edge is.

Day-of-week P/L

Tuesday and Thursday are the workhorses (~+100% ROI each). Wednesday and Sunday are weak (~+9% and +6% ROI). There's no strong story for why those two midweek days are the standouts - most likely it's just sample variance on small absolute numbers. The bot doesn't attempt to time the week.

Burst patterns and accumulation windows

Same-second multi-fill is the norm - 84% of consecutive fills on the same (market, outcome) happen within 1 second. The April-22-9:25PM example (six BUYs in the same second, then five SELLs eight seconds later) is the signature pattern.

Median accumulation window per market: roughly 90 seconds (entry late in the 5-min window, exit moments later). He never returns to a market after it resolves - every event is one-touch.

Phase 7 - Filter Experiments

Most filters are useless or destructive on this trader. The price-band filters return zero trades because 100% of his book is in the $0.00–$0.10 band. The dominance filter is meaningless because he's barely both-sides at all. Only the hour-exclusion filters add anything, and they add very little - about $190 of incremental P/L (3.4% lift) by skipping the four worst-performing hours of the day.

The strategy is already running at filter terminal velocity. There's no further refinement to extract by tightening selection. The base strategy is the optimization. Filters that work on directional or MM books do nothing here because the underlying engine doesn't have the dimensions those filters operate on.

The one practical refinement would be to skip hour 19:00 UTC entirely - it's the only outright loser hour, and excluding it cleanly adds about +$95 / +0.6% ROI without throwing out any meaningful sample.

Phase 8 - Rolling Window Consistency

This is one of the most consistent edge profiles in the library. 96% of all rolling windows close green, and the only "red" window is a -$2 dip in the early days when the bot was still warming up. The cumulative line is monotonically rising for essentially the entire observation period.

The cumulative trajectory:

Day  1 (Apr 5):  +$13       Day 14 (Apr 18):  +$1,732
          Day  5 (Apr 9):  +$592      Day 19 (Apr 23):  +$4,009
          Day 10 (Apr 14): +$835      Day 25 (Apr 29):  +$5,563

Notice the acceleration in week 3-4. Daily P/L averages ~$60 in the first 10 days, then ~$330 in the last 10 days. Two possible explanations:

1. Capital scaling. He may have increased the per-market clip size as the strategy proved out - the daily BUY notional climbs from ~$200 in week 1 to ~$800 in week 4.
2. Market expansion. The bot may have widened its market-selection rules to touch more 5-min markets per day (1,418 trades on April 20 vs 408 on April 5 - 3.5× the velocity).

Either way, there's no mean-reversion to fear - the strategy keeps getting more profitable as he scales it. No drawdown deeper than $76 on any single day. The book never goes underwater for more than a few hours. It is the closest thing to a risk-free yield instrument I have seen in this dataset.

Phase 9 - P/L Decomposition

Breaking the edge into its two engines:

1. The SELL engine generates $563 alone (SELL proceeds exceed BUY cost by themselves). That's a +5.1% return on capital from market-making activity only, before any resolution outcomes. This is the orderbook-bounce harvester - every $0.01 share flipped at $0.02 banks $0.01.

1. The settlement engine generates the remaining $5,661 from the 296 winning tickets that paid $1.00 each on a $0.01 cost basis. This is the lottery component - 1.36% hit rate at 99× gross payout = ~+35% expected return per ticket on the held residual.

The two engines run in tandem on the same shares: he tries to flip them at $0.02 first, then if that doesn't happen, the share auto-converts into a held lottery ticket at expiry. Both branches are positive-EV, which is why the strategy is so resilient. Even if the SELL fill rate dropped to zero, the settlement engine alone would still be profitable. Even if the win rate dropped from 1.4% to 0.7%, the SELL engine alone would still be profitable. The two paths backstop each other.

The hedge-tax line ($168) is essentially nothing - because he barely touches both sides, the hedge cost is negligible.

Phase 10 - Strategy Specification (short form; full detail in playbook.md)

One-sentence summary: A single-purpose Polymarket bot that places $0.01 bids on the cheapest leg of every 5-minute Bitcoin Up/Down market and posts $0.02 asks on whatever fills, holding the unsold residual to settlement.

Edge source: Two stacked positive-EV engines:

1. Orderbook-bounce arbitrage ($0.01→$0.02 = 100% gain, 56% fill rate).
2. Tail-event harvest (1.4% win rate at 99× payout = +35% per held share).

What works: Off-hours trading windows (00–07 UTC). Tuesday and Thursday calendar days. The $0.01-bid/$0.02-ask price wedge. Same-second multi-leg fan-out execution.

What drags: Hour 19:00 UTC (the only net-loss hour). Weekend daytime windows.

Risk profile: Maximum daily drawdown observed: $76. 96% of all rolling windows green. The strategy is bounded-loss by construction - every fill is $0.01-$5.00, and the worst case per fill is -$0.01 to -$5.00. It is structurally incapable of a large blowup.

Why this is uniquely interesting: Most edge strategies in prediction markets blend signal and structure (directional view + sizing + hedge). LIL222 has stripped every dimension except the entry price and the exit price. It's a one-rule bot operating on a one-tick price wedge. The fact that 56% net ROI in 25 days is achievable on this little capital with this little code is what makes it worth replicating.

The playbook spec translates this into runnable thresholds and bankroll math.

Window: 2026-04-05 → 2026-04-29 (25 calendar days, 25 active) Source CSV: LIL222_trades.csv Methodology: Cash-flow P/L = -buy_usdc + sell_usdc + remaining_share_payout. Remaining shares settle at $1 if outcome won, $0 if lost, or mark at last-traded price if open. No tx-hash dedup (atomic multi-leg fills are real trades).

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: 1.60% (56 of 3,500 markets)
• Median paired cost: $0.0200
• Mean paired cost: $0.0200
• Paired cost % under $1.00: 100.0%
• Paired cost % under $0.97: 100.0%
• Paired cost % under $0.95: 100.0%
• Median 2nd-side hedge lag: 38s (0.6 min)
• % of pairs hedged within 60s: 69.6%
• % of pairs hedged within 1hr: 100.0%

Dominance buckets (5-tier)

Phase 4 - Entry-Price Analysis (cash-flow P/L)

Phase 5 - Category & Vertical Breakdown

Phase 6 - Timing & Execution

Net P/L by hour (UTC)

Net P/L by day-of-week

Phase 7 - Filter Experiments

Phase 8 - Rolling Window Consistency

• Rolling 7-day windows green: 24 of 25 (96.0%)
• Rolling 7-day P/L range: -$2 → +$3,031
• Rolling 15-day windows green: 24 of 25 (96.0%)
• Rolling 15-day P/L range: -$2 → +$4,794

Daily P/L (BUY trades, cash-flow allocated)

Phase 9 - P/L Decomposition

Theoretical structural attribution

Phase 10 - Top Markets

Top 25 by BUY notional

Top 15 winners by P/L

Top 15 losers by P/L

Report generated 2026-04-29 01:41 from LIL222_trades.csv.

Wallet: 0xeaca59cb5e10e0be128b005a0f84465d2ed80729 Window: 2026-04-05 → 2026-04-29 Baseline: 21,919 BUYs · 1.36% WR · $11,146 deployed · +$5,563 P/L · +49.91% ROI

The headline result

On this trader, almost no filter helps. That is itself the story.

Most of the filters in the standard PR&R framework are designed for traders with multiple dimensions of variance - directional vs spread, multiple categories, multiple price bands, multiple market types. LIL222 has compressed all of those dimensions to a single rule (bid $0.01, ask $0.02 on 5-min BTC). There's nothing left to filter on. The strategy is already at filter terminal velocity - every fill the bot already takes was selected by the underlying entry logic, and additional gating mostly subtracts good fills along with a tiny number of bad ones.

The one thing that does help: dropping the worst hours of the day. That alone adds about +$190 of P/L (3.4% lift). Everything else is noise or destruction.

Filter results table

Filter-by-filter commentary

1. Price band filters → useless

100% of his book lives in the $0.00-$0.10 band (in fact, all 21,919 BUYs sit at exactly $0.01). Any filter that targets a higher band returns zero qualifying trades and zero P/L. The classical "$0.30-$0.70 sweet spot" filter, which is the workhorse for most directional bettors, is a non-event for LIL222 because he doesn't trade outside the floor.

Verdict: N/A. The filter framework's price dimension doesn't apply.

2. High-conviction filter (dominance ≥ 2×, dominant leg only) → destructive

This filter would normally extract the alpha leg from a market-maker by ignoring the hedge side and keeping only the directional bet. LIL222 doesn't market-make - only 56 of 3,500 markets have both sides bought, and only 2 of those exceed 2× dominance. The filter qualifies just 14 trades, all losers, totaling -$7 P/L.

Verdict: Destructive. Strips out 99.94% of the book to capture a tiny noise sample.

3. Category filter → meaningless

100% of his trades are Crypto. There is one category. There is nothing to filter on.

Verdict: Identity transformation. Returns the baseline.

4. Exclude single worst hour (19:00 UTC) → small lift

Hour 19:00 UTC is the only outright net-loss hour in the day (-$95 over 1,139 trades, -18.4% ROI). Excluding it drops 1,139 trades and adds +$95 to P/L (+0.6% ROI lift).

This is the only filter on the menu that delivers a clean, intuitive, positive result. It's small in absolute terms because the baseline is small in absolute terms - but the percentage lift is real and the rule is unambiguous.

Verdict: ✅ Worth implementing. Skip 19:00–20:00 UTC. Costs you 1,139 fills, saves you $95.

5. Exclude worst 4 hours → larger lift, larger sample loss

Drops the four worst-performing hours (cumulative across the bottom of the hourly P/L distribution): some combination of 19:00, 03:00, 11:00, and 15:00. Removes 3,765 trades and adds +$190 P/L (+12 ROI points).

The catch: this is the maximum you can squeeze out of hour-filtering on the observed data without overfitting. Pushing further (excluding 6 hours, 8 hours, etc.) starts cutting into samples whose negative ROI is within noise of the positive ones.

Verdict: ✅ The single most productive refinement available. Sacrifices 17% of trade volume for a 3.4% lift in absolute P/L and a 24% lift in realized ROI on the trades that remain.

6. STACK (high-conv + skip worst hour + skip losing cats) → broken

The stacked filter is dominated by the high-conviction component, which already collapsed the sample to 14 trades. Adding hour and category exclusions on top doesn't recover any signal. Output: same 14 losers as the high-conv filter alone, -$7.

Verdict: Don't stack high-conviction here. It's incompatible with the strategy's structure.

What filters would add value if you could measure them

The most useful refinements aren't in the standard framework because they require orderbook depth and microstructure data we don't have from the trade log alone. Hypothetically:

None of these can be computed from the trade CSV alone. If LIL222 himself is using any of them, we can't tell from the wallet history - they would be invisible filters expressed in his entry code.

Bottom line for replication

The base strategy already extracts the available edge. The only practical post-strategy filter is:

That's it. Don't overcomplicate it. The dollars saved by hour-filtering are real but small ($95-$190 over 25 days). The dollars saved by not trying to layer on more sophisticated filters than the underlying strategy can support are larger - because every additional gating rule on a sub-second bot adds latency, and latency is the one thing this strategy genuinely cannot afford.

If you replicate this bot, the implementation lift comes from adding a max-fill-rate ceiling (so you don't over-allocate to a single hot market) and scaling the per-market clip with available capital, not from filter selection. See playbook.md for that spec.

Source wallet: 0xeaca59cb5e10e0be128b005a0f84465d2ed80729 Strategy: Passive penny-longshot harvester on 5-min Bitcoin Up/Down markets Reference book: $11,146 BUY notional → +$6,267 net P/L → +56.22% ROI in 25 days

One-paragraph operator brief

Build a Polymarket bot that fires a $0.01 BUY on whichever side of every 5-min btc-updown-5m-* market is currently at the orderbook floor, then immediately posts a $0.02 ASK on the same shares. Hold any unsold residual to settlement. Don't look at BTC. Don't look at funding. Don't look at anyone else's trades. Run it 24/7. Skip the 19:00 UTC hour. Cap per-market spend at $5. Expect ~+50% ROI per month on $10K of working capital, with worst-day drawdown bounded by the daily fill rate × $5 = roughly $750/day in absolute downside.

1. Market selection

Notes:

• The 5-min window is what makes this work - the resolution clock creates the price gaps that drop one side to $0.01 in the final 60-90 seconds. Longer windows don't gap as often or as deeply. Shorter windows don't exist on Polymarket as a stable inventory.
• New markets open every 5 minutes (288/day). Expect to touch ~140/day = roughly 50% of available inventory based on which side gaps to the floor.

2. Entry logic

def should_enter(market):
              if market.duration != "5m":          return False
              if market.asset != "BTC":            return False
              if not market.is_live:               return False
              if time_until_close(market) > 180:   return False   # only enter in last 3 min
              if time_until_close(market) < 5:     return False   # don't fire after window closes
              
              floor_side = side_with_ask_at(market, 0.01)
              if floor_side is None:               return False   # no $0.01 ask available
              
              return floor_side

3. Exit logic (the SELL leg)

def manage_position(position):
              # Try to immediately post the $0.02 ASK
              post_ask(position.shares, price=0.02, expires=position.market.close_time)
              
              # If the ASK doesn't fill before market close, hold to settlement
              # (no stop-loss, no trailing exit, no repricing)

The reason for not waiting at $0.03+: the bot is competing with other bots for the same $0.02 fill. If you set your ask at $0.03 and another bot sets $0.02, theirs fills first and yours becomes the held lottery ticket - but you've also lost the $0.01 immediate gain. The marginal utility of $0.03 vs $0.02 is small ($0.01 extra on a $0.05 ticket = +20%), but the marginal probability of your ask getting hit drops from ~50% to <10% because you're behind every $0.02 ask in the queue. Lock the $0.02 immediately.

4. Sizing model

The reference wallet uses fixed clips (no scaling-by-edge, no Kelly). Recommended sizing for replication:

The strategy's edge degrades when your per-market clip approaches the depth at the $0.01 ASK - you'll partially fill, push the price up off the floor, and lose the entry condition. The reference wallet at $5/market is operating well below the depth wall. Above ~$25/market on most markets you'll start moving prices.

Recommended max bankroll for this strategy: ~$50K-$75K. Beyond that the alpha doesn't scale and you'll need to fragment across multiple wallets to avoid self-impacting your own fills.

5. Bankroll math

Using observed reference numbers as a baseline:

Monthly performance assumptions (extrapolated from 25-day window):
            Capital deployed per month:    ~$13,400 (BUY notional)
            Net P/L per month:             ~+$7,500 (cash-flow)
            Expected ROI per month:        ~+56% on capital deployed
            
          Required liquid working capital: ~$2,000-3,000 (peak instantaneous exposure;
                                                          most capital cycles within minutes)

Important: the +56% ROI is on deployed capital, not on bankroll. The bot only deploys ~$11K total over 25 days but cycles it constantly - on any given day, instantaneous exposure rarely exceeds $2-3K. So the capital efficiency is much higher than the surface ROI number suggests:

On a $5,000 standby bankroll → $7,500 monthly net P/L → +150% monthly ROI on bankroll
          On a $10,000 standby bankroll → $7,500 monthly net P/L → +75% monthly ROI on bankroll

The constraining input is not capital but fill rate - how many of the available 288 daily 5-min markets actually gap to $0.01 on a side you can lift. The reference wallet hits ~140/day, which suggests bot competition isn't yet saturating the floor.

6. Hour exclusions (the only filter that matters)

Conservative recommendation: skip only 19:00 UTC. The other three are noise on the small absolute numbers and may not persist.

Aggressive recommendation: skip all four (~17% of trades, +$190 P/L lift, +12 ROI points on remaining trades). Use this if your fill ceiling is the constraint rather than your hour budget.

7. Operational requirements

8. Risk profile

The strategy is structurally incapable of a large blowup. Every dollar at risk is a $0.01 ticket. The worst 25 days the model could plausibly produce would still be +$0 not -$5K, because the SELL engine alone (orderbook bounce harvesting) backstops the position.

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

Run weekly:

10. What this playbook deliberately does NOT include

• No directional BTC view. Don't add one. The trader doesn't have one and the data confirms a directional view actively destroys the edge.
• No Kelly/Optimal sizing. Fixed clips are optimal here because every ticket has the same EV.
• No category expansion. Don't try to apply this to ETH/SOL/sports/etc. The 5-min BTC orderbook is the specific microstructure that produces the $0.01-floor gap. Other markets don't behave the same way.
• No copy-trading. This is a self-contained edge - there's no signal to follow from another wallet.
• No hedging. The both-sides participation rate of 1.6% confirms hedging is not part of the strategy. Don't add a hedge.
• No machine learning. There is nothing for a model to learn here. The rule is the rule.

The whole point of LIL222's edge is that the strategy is small, mechanical, and unhedged. Every "improvement" you might be tempted to add is something he tried implicitly and rejected. Trust the simplicity.

TL;DR - implementable in 100 lines of Python

# Pseudocode - outline only
          while True:
              for market in poll_markets("btc-updown-5m-*"):
                  if not 5 < seconds_until_close(market) < 180:
                      continue
                  if utc_hour(now()) == 19:
                      continue   # skip the one losing hour
                  floor_side = side_with_ask_at(market.orderbook, 0.01)
                  if floor_side is None:
                      continue
                  # Lift $0.01 ASK with up to $5 of buying power
                  fills = lift_ask(market, side=floor_side, price=0.01, max_usdc=5.00)
                  # Immediately post $0.02 ASK on filled shares
                  if fills.shares > 0:
                      post_ask(market, side=floor_side, price=0.02,
                               shares=fills.shares, expires=market.close_time)
                  # Unsold shares auto-settle at $1 (win) or $0 (loss)

Run this 24/7. Reconcile daily. Expect ~+50%/month ROI on $10K of deployed working capital, bounded by the available daily 5-min market inventory and your latency to the matching engine.

That's the entire strategy. There is nothing else.