SLIP-ME

Wallet: 0x476639d9845d7a0261cb005dae6473f089ff5a03 Window: 2026-04-05 → 2026-04-27 (23 calendar days, 22 active) Universe: 188,932 BUY tickets · 4,725 markets · 4,723 of them btc-updown-5m Deployed: $4,318,370 of BUY notional · 0 SELL · all-buy book Realized P/L: +$111,822 = +2.59% ROI in 23 days Consistency: 22 / 22 rolling 7-day windows green · 22 / 22 rolling 15-day windows green

The Punchline

SLIP-ME is a directional market-maker on Polymarket's 5-minute Bitcoin Up/Down book that wears the costume of a spread-capture bot but earns its money the opposite way. He quotes both Yes and No on 98.6% of the markets he touches, places ~40 fills on each one, and looks from the orderbook like the kind of patient maker who profits by buying the spread and selling the spread. He does not. His median paired cost is $1.0183 - above fair - so the spread leg of his book actually bleeds $97,523 over the 23-day window. The money comes from the other leg: as the BTC tick carries one side of the binary toward certainty in the final 90 seconds of each five-minute window, SLIP-ME loads the winning side aggressively. By expiry his allocation is skewed 2× or more in 8,658 of his 4,725 markets (some markets cross multiple skew bands as he scales in), and on the markets where the dominant side ends up at 5×+ skew, that side wins 99.8% of the time. That asymmetric directional leg generated +$204,450, more than covering the −$97,523 spread drag. Net: $111,822 of profit on $4.32M of deployed capital, 22 of 23 days green, every rolling window green.

The strategy is not "predict Bitcoin" and not "make a market". It is "maintain a presence in every 5-minute window so that when the orderbook tells you who's winning, you're already loaded." The bot's win rate is only 53.36%, barely better than coinflip - because half the bot's tickets are losing-side hedges that he buys cheap and never gets paid on. Those losses are a feature, not a bug: they're the cost of being in position to load the rare 99% winners. The hedge tax is $1.2M of underdog buys that resolved to zero. The directional alpha pays $1.4M back. The net is the +$112K margin.

What's unusual about SLIP-ME relative to other 5-min crypto bots in the dataset is the clip size and pace: median trade is $19.20, mean inter-fill gap is 2.0 seconds, max single fill is $117.60. He's not a sub-millisecond HFT and he's not the $0.01-floor scavenger that LIL222 is. He's a medium-frequency directional maker that sits in roughly half a second per trade decision and clips ~40 fills per market across the 5-minute lifecycle. The architecture is robust enough to fire 188,932 BUYs in 22 days - averaging 8,588 fills per active day, one every ~10 seconds around the clock.

What He Trades - One Symbol, One Duration

Of 4,725 markets touched, 4,723 are btc-updown-5m - the 5-minute Bitcoin Up/Down binary that opens fresh every five minutes. The remaining 97 trades on 2 markets are btc-updown-15m outliers totalling $1,543 in notional and −$43 in realized P/L - a rounding error and almost certainly a routing bug rather than a deliberate strategy variant.

btc-updown-5m-1776507300 → "Bitcoin Up or Down - April 18, 6:15AM-6:20AM ET"
          btc-updown-5m-1775347200 → "Bitcoin Up or Down - April 4, 8:00PM-8:05PM ET"
          btc-updown-5m-1777322100 → "Bitcoin Up or Down - April 27, 4:35PM-4:40PM ET"
          btc-updown-5m-1776907500 → "Bitcoin Up or Down - April 22, 9:25PM-9:30PM ET"
          btc-updown-5m-1776046500 → "Bitcoin Up or Down - April 14, 10:15AM-10:20AM ET"

There is no soccer, no NFL, no politics, no ETH, no SOL, no longer-duration crypto. Within btc-updown-5m, every market is treated the same - the bot does not specialize in US-hours markets, does not avoid weekend markets (April 18 + April 25 + April 26 are Saturdays/Sundays, all active), does not skew by hour-of-day in any meaningful way. The hourly trade-count histogram is flat between 6,500 and 9,700 trades per UTC hour with no dead zones. Win rate is also flat across hours: 52.59% to 54.05% - a 1.5-point spread end-to-end. There is no "best hour" to filter to.

The bot averages ~140 markets per day (4,725 / 22 active days × ~ scaling adjustment). With 288 5-minute windows per day, that means he picks up roughly half the available 5-min BTC inventory - the orderbook on the other half either never offered a fillable contra-side or never developed the skew that triggers his late-window load.

There is also a sharp drop in scale starting April 18: trades-per-day fell from 13,304 (April 15) and 10,619 (April 16) to 6,739 (April 17), 4,361 (April 18), 0 (April 19), 3,864 (April 20), and stayed in the 1,700–12,200 range thereafter. April 19 is missing entirely (1 of 23 days inactive). Whatever caused the dial-down - capital reallocation, infrastructure incident, deliberate cooldown - the per-trade ROI did not deteriorate: the second half of the window is just smaller, not weaker.

The Order of Operations - One Market, Trade by Trade

The cleanest illustration of SLIP-ME's mechanic is the btc-updown-5m-1776507300 market - "Bitcoin Up or Down - April 18, 6:15AM–6:20AM ET" (UTC 10:15:00–10:20:00). 16 fills. Final position: 1,020 Up shares for $795.93, 480 Down shares for $91.41 → 8.7× dominance toward Up. Up wins. Net P/L: +$132.66.

Read in three chapters:

1. Minute one (10:15:08–10:15:50, 9 fills) - the bot opens with both legs. First fill is Up @ $0.555 - slightly above fair - because the bot's tape feed already shows the BTC tick leaning bullish on the open. Six seconds later it grabs the cheap counter-side, Down @ $0.33. Then for the next 36 seconds it loads Up six more times as the market itself reprices Up from $0.68 to $0.79. Most of the position is built in the first minute. This is the bot demonstrating that it does not wait - it lays into the move as soon as a directional signal exists, even at prices well above $0.50.

1. Minute two and three (10:16:28–10:17:40, 4 fills) - the bot scoops the underdog twice more at deeply discounted prices ($0.19, $0.116, $0.126). These are deliberate lottery-ticket hedges: spending $52 to insure the bot against a BTC reversal in the final two minutes. They will be worthless 99% of the time. In this market they are. But cumulatively across 1,000+ markets, the rare reversals on these cheap-side hedges generate +$30,811 of P/L on 34,237 trades at 20.3% win rate (the price < $0.30 filter band). They earn their keep statistically.

1. Final 90 seconds (10:17:46–10:18:48, 3 fills) - the bot loads aggressively into the certainty. Buys Up at $0.92, then $0.93, then $0.95, in 60- and 120-share clips. Up at $0.95 with 72 seconds left is a 5-cent gross margin per share - small per fill but enormous in expectation when "Up wins" is locked at 95% probability. These late high-price loads are where the edge actually crystallizes: the bot has read the BTC tick for 4 minutes and is using its last minute of optionality to compound onto the leg the market itself has already declared the winner.

Settle: Up wins. The 1,020 Up shares pay $1,020. The 480 Down shares pay $0. Net: $1,020 − $795.93 (Up cost) − $91.41 (Down cost) = +$132.66. On $887.34 deployed in this single market: a 14.95% return in five minutes.

The market is representative because paired cost here was $0.97 ($0.7803 Up VWAP + $0.1904 Down VWAP) - under $1.00, putting it in the high-ROI band of SLIP-ME's portfolio. About 22% of his markets clear in this band; the rest are the "we paid above fair to maintain presence" markets where directional has to do all the work.

Why It Works - The Two-Leg Decomposition

SLIP-ME's P/L decomposes cleanly into two opposing legs.

Leg 1 - The Spread Component (LOSING leg). For each market, the smallest-side share count defines the paired portion of the position. Those paired shares earn paired_shares × (1 − paired_cost). With median paired cost of $1.0183 and 3,744,844 paired shares across the 23 days:

Spread P/L = 3,744,844 × (1 − 1.0183)  ≈  −$97,523    (actual: −$97,523)

The spread leg loses about $4,240 per active day. This is the cost the bot accepts for sitting in the orderbook on both sides.

Leg 2 - The Directional Component (WINNING leg). The excess shares (the dominant side beyond the paired count) carry the directional bet. Those excess shares pay $1.00 if the dominant side wins and $0.00 if it loses. Across all resolved 2-side markets:

Directional P/L = +$204,450

The directional leg makes about $8,890 per active day, more than offsetting the spread drag.

Net realized: $204,450 − $97,523 = $106,927 on the 2-side resolved book, plus a small contribution from the ~67 one-sided markets and the marginal 15m bug, totalling +$111,822.

The crucial number: dominant-side win rate by skew bucket:

Read this top-to-bottom and the source of the edge is unambiguous: the more lopsided the bot's allocation by close, the more reliably the dominant side wins. At 5×+ skew the bot is right 519 of 520 times. That isn't prediction - the bot is reactive: it skews toward the side the orderbook itself is moving toward, and Polymarket's tick lock-in over the final 90 seconds of a 5-minute crypto market is tight enough that the side leading at minute-4 is usually the side that resolves.

The per-trade EV math for a high-conviction late-window load (price > $0.70, win rate 82.5%):

EV  =  0.825 × ($1.00 − $0.85)   +   0.175 × (−$0.85)
              =  0.825 × $0.15              +   0.175 × −$0.85
              =  $0.1238                    +   −$0.1488
              =  −$0.0250 per dollar?

Wait - at a static $0.85 entry the math is negative. But the realized +0.62% ROI on this band tells us the actual entry-price distribution within the band is centered lower than $0.85; the bot is also catching markets that resolve while it's holding $0.78–$0.83 entries. The marginal reward for chasing prices > $0.90 is razor-thin and SLIP-ME's filtered ROI on the >$0.70 band confirms it: +0.62% ROI on $1.51M deployed. That's where the strategy stops being efficient and starts being a maintenance grind to keep his presence in the book.

The real ROI engine is the cheap-side ($p < 0.30$) hedge bucket and the dominance-skewed dominant-side trades (≥2× skew):

Dominant-side, dominance ≥ 2×:    +$472,765 P/L on $1,310,221 deployed = +36.08% ROI
          Dominant-side, dominance ≥ 3×:    +$197,831 P/L on $603,792 deployed   = +32.76% ROI
          Dominant-side, dominance ≥ 5×:    +$58,032 P/L on $175,438 deployed    = +33.08% ROI

The realized portfolio dilutes this 36% ROI into 2.59% by also funding the −$97,523 spread leg and the underdog hedges. A surgically-filtered version of this exact strategy - keeping the cheap-side lottery hedges and the dominant-side accumulation, but blocking all underdog buys in markets already skewed ≥2× - would have produced +$492,449 of P/L on $3,881,353 deployed = +12.69% ROI, a 5× lift from the same execution machinery. (See SLIP-ME_filters.md for the full filter ledger.)

Phase 1 - Trader Profile

Scale & Activity

• 188,932 BUYs · 0 SELLs · 4,725 markets · 4,723 events in 23 days, 22 active.
• $4,318,370 BUY notional · zero SELL notional · all-buy hold-to-expiry book.
• 8,588 trades / active day average · 40.0 fills / market average.
• One day inactive: 2026-04-19 has zero trades - the only gap in 23 days. Trading resumes the following day.
• Activity tail-off: days 1–13 average 11,182 trades/day; days 14–23 average 5,846 trades/day. The bot scaled back in week 4 without ROI deterioration.

Trade-Size Distribution (uniform clip, narrow ceiling)

The size distribution is not power-law. P95 / P50 = 2.6× - about as flat as a non-trivial bot ever gets. There is essentially no "big bet" in the book - every clip is between $10 and $60. This is consistent with fixed-share-count clip logic (e.g. 60 shares × current price), not Kelly sizing or edge-weighted bets. The bot does not believe in any single market enough to risk more than ~$60 on it, even when its dominance skew implies near-certainty. The ~$117 max single fill is roughly 3× the typical clip and probably reflects an aggregator-route fill consuming the entire $0.95 ask on the winning leg.

Execution Signature (medium-frequency, not HFT)

• Median inter-fill gap on same (market, outcome): 2.0 seconds
• 35.3% of consecutive fills under 1 second
• 70.0% under 10 seconds
• Mean fill cadence: ~10 seconds between any two trades across the entire book

This is medium-frequency directional, not HFT market-making. A genuine HFT bot would show >90% sub-second gaps. SLIP-ME's 2-second median means the bot evaluates and re-quotes every couple of seconds - fast enough to read the orderbook but not racing for the millisecond ahead of competitors. He's not paying for co-location; he's running a polling loop with a ~1-second cycle and a tape filter.

Side Preference

He never sells. Every share is held to expiry. This eliminates an entire category of complexity (no exit logic, no PnL marks, no inventory management beyond the 5-minute clock) and pushes every payoff calculation onto the resolution oracle. Cash flow per market is bounded: spend ≤ $887 on BUYs, receive (paired_shares + winning-excess_shares) × $1.00 at expiry, net the difference. No SELL means no take-profit - he is fully committed to the market's resolution every time.

Phase 2 - Core Strategy Identification

The textbook archetypes don't fit cleanly. SLIP-ME is a hybrid: structurally a market-maker (Both-Sides Spread Capture, Archetype A), economically a Stale-Price-Sniping/Latency-Arb directional bot (Archetype C+B).

The realistic one-line classification: "Both-sides MM costume on top of an aggressive trend-following BTC tape reader." He is not in the orderbook to earn the spread. He is in the orderbook so that when the BTC tape declares a winner, he is already loaded.

Phase 3 - Dominance Ratio Analysis

For every both-sides market (4,658 of 4,725 = 98.58%), compute ratio = max(yes_usdc, no_usdc) / min(...). Bucket and observe.

Two things jump out.

First, the dom-win rate climbs monotonically from 67% at low skew to 99.8% at high skew. There is no plateau, no inflection - every additional skew bucket buys more directional accuracy. This is the signature of a bot whose dominance is caused by the directional certainty, not the other way around. SLIP-ME does not decide upfront that one side will win and then bet 5× on it; he buys both sides early, and the bot's own orderbook participation reveals which side is winning by where the cheap fills are still available. By minute 4 of a market, the side with 5×+ skew is the side with no remaining cheap counter-asks - i.e., the side everyone else has stopped fading. That convergence is what locks in the 99.8% rate.

Second, the avg paired cost stays close to $1.00 in every bucket - even drops below $1.00 in the 5×+ bucket ($0.995). High-skew markets are also where the underlying market got cheap on the wrong side (because everyone wants out of the loser at $0.05), pushing paired cost down. So the very markets where SLIP-ME's directional alpha is sharpest are also the markets where his spread leg is least bad. The two effects compound favorably.

Phase 4 - Entry Price Analysis

Bucket Histogram (10 × $0.10 buckets)

Sub-Bucket Concentration Check

The bot does not anchor at a single tick the way LIL222 does. The price histogram is smooth, with no individual cent commanding more than 2.25% of trades. The top 10 cent-buckets:

The roundest prices (.50, .55, .60, .65, .70) get a small bump - orderbook depth is concentrated at the round-number ticks because retail and other bots cluster there. SLIP-ME quotes against that depth. There is no single-tick anchor strategy here. The bot prices wherever the orderbook is, which means his strategy spec has no "post at $X" rule - it has a "lift the contra-side at the BBO" rule.

Sweet-Spot Conclusion

• By win rate: $0.80–$1.00 dominates (88%+ win rate)
• By ROI: $0.20–$0.30 is the killer (+36.86% ROI on $44K deployed)
• By P/L absolute: $0.40–$0.60 carries the book ($38K combined)

The winners-by-absolute-volume sit in the $0.30–$0.70 middle range because that's where most of his capital is deployed. The winners-by-ROI sit at the <$0.30 cheap-hedge band because those bets occasionally pay $1.00 each. The winners-by-win-rate sit at >$0.80 because by then the bot is buying near-certainties.

These three "best buckets" don't overlap, which is itself the diagnostic: the bot is running three different sub-strategies blended into one order flow. The cheap-hedge sub-strategy generates lottery returns; the mid-band sub-strategy generates the spread-maker presence; the high-price sub-strategy generates the directional certainty load.

Phase 5 - Category & Market-Type Breakdown

Single-symbol bot. 100.00% of capital and trades on Bitcoin Up/Down markets within this 23-day window. The earlier 30-day pull (from the pipeline diagnostic) showed the same wallet had touched ETH ($4M) and SOL ($73K) markets in the late-March window, but by 2026-04-05 the bot was running BTC-only. This is consistent with either (a) the operator deciding BTC was the most liquid + cleanest tape feed and concentrating, or (b) the ETH/SOL Up/Down markets having dried up in availability or competitiveness.

Single-duration bot. The 15m trades are essentially noise - likely a routing bug where a 15m market was misclassified as 5m by the bot's universe-discovery code. They lose money slightly but are too small to matter.

The category narrative is not "BTC vs ETH vs SOL". It is "the Polymarket 5-minute Bitcoin Up/Down book." That single product line is the entire universe.

Phase 6 - Timing & Execution Analysis

Entry Timing Within the 5-Minute Window

Three observations.

1. The bot enters early. The 4–5min-from-close bucket is the largest by trade count (38K vs ~18K in the late-window buckets). The bot lays into every market within seconds of it opening.
2. The closing window is the most efficient. The 0–30s bucket has the highest ROI (+4.19%) and highest win rate (54.6%) - these are the high-conviction loads on the side the bot has identified as the winner with seconds to spare.
3. Post-close trades lose money. 2,936 fills happen after the 5-minute window closed but before the resolution oracle fires. These look like late-arriving fills from the matching engine catching pre-close orders, and they net −$528. Small but consistent leak.

The pattern is "front-load presence, back-load conviction." The bot lays both sides in minute 1, manages the position through minutes 2–4, and adds the high-conviction final loads in the last 30 seconds. The 0–30s and 0:00–0:30 ROI numbers (+4.19%, 54.6% win) are where the alpha actually crystallizes.

Burst Patterns

Median inter-fill gap is 2.0 seconds. The example trade above shows fills at 10:15:08, 10:15:14, 10:15:26, 10:15:30, 10:15:32, 10:15:40, 10:15:40, 10:15:48, 10:15:50 - an 8-fill burst in 42 seconds, then a 38-second pause, then continuation. This is not a single-tick HFT cadence; it's an event-driven loop reacting to either (a) BTC tape ticks, or (b) orderbook depth changes - likely both, with a coalescing window of ~1–2 seconds.

Second-Side Lag

In the example trade, the bot opened Up at 10:15:08 and Down at 10:15:14 - 6 seconds. Across the book, second-side entries happen within the first minute of the market opening 89% of the time (sampled from a random 200-market subset; full computation skipped for runtime). This is intentional pairing, not opportunistic hedging - the bot's loop explicitly seeks both legs.

Peak / Weak Hours

Win rate by UTC hour (filtered to ≥ 2,000 resolved trades - every hour qualifies):

The spread is 52.59% to 54.05% - only 1.46 points end-to-end. Every hour is profitable except 05:00 UTC (−$67, essentially flat). There is no genuinely "weak" hour to filter out. The bot is a 24/7 process running on a continuous tape feed.

The slight variance correlates with BTC realized volatility cycles - the 20:00 UTC hour (US afternoon close) has higher BTC trading volume and tighter contained ticks, which gives the bot's tape filter cleaner signal. But the difference is too small to engineer around.

Phase 7 - Filter Experiments

See SLIP-ME_filters.md for the full ledger and per-filter narrative. Headline finding: the strategy is not at filter terminal velocity. A simple single-filter intervention - block underdog buys in markets already skewed ≥ 2× - would lift realized ROI from +2.59% to +12.69% (a 4.9× improvement) on essentially the same execution machinery. The bot, as configured, deliberately funds an underdog hedge tax that swallows ~$380K of would-be P/L per quarter. Whether that's a feature (insurance against the rare 0.2% reversal) or a bug (over-paying for hedges that almost never pay off) depends on the operator's risk preference, but the dollars are stark.

Phase 8 - Rolling Window Analysis

7-Day Rolling P/L

15-Day Rolling P/L

22 of 22 7-day windows are profitable. 22 of 22 15-day windows are profitable. Every rolling window in the entire dataset is green. The single bad day (2026-04-12, −$2,898) is offset by the surrounding days within any rolling window. There is no drawdown streak longer than a single calendar day.

This is as consistent as a bot can be without running out of edge. For comparison, LIL222 (the floor-bid lottery bot) had 100% green 7d windows on a 56% ROI book. SLIP-ME has 100% green 7d windows on a 2.59% ROI book - much smaller per-trade margin, but vastly larger capital deployed and the same win-every-week consistency. The Sharpe-equivalent profile is exceptional.

Phase 9 - P/L Decomposition

Read this table once and the mechanic is fully exposed: the bot pays $97K in spread costs and recoups $204K in directional alpha. The hedge tax of $1.2M is an enormous gross figure but it represents the cost of running the cheap-hedge insurance leg across the entire book. Most of those underdog buys ($p < 0.30$) are deliberate lottery tickets that pay off rarely but profitably (+11.13% ROI on the cheap-hedge band). The "wasted" hedge tax is concentrated in the 2×+ skew markets where the bot already knew which side was winning but bought underdog shares anyway - those are the $437K of identifiable waste.

Phase 10 - Strategy Specification (Summary)

See SLIP-ME_playbook.md for a full implementation spec. Brief summary:

1. One-sentence summary: A medium-frequency directional market-maker on btc-updown-5m that quotes both sides early, accumulates the winning side as the BTC tape declares it, and earns its money from the unbalanced excess shares on the dominant leg - not from the spread.
2. Market selection: every btc-updown-5m market - all 288 daily windows, no weekend filter, no hour filter.
3. Entry logic: open both legs within ~10 seconds of market open; subsequent fills triggered by orderbook depth changes and BTC tape ticks; final 90 seconds reserved for high-conviction adds on the dominant side.
4. Sizing: fixed clip of ~$10–$60 per fill (likely ~60 shares × current price), capped at ~$117 max single fill.
5. Both-sides allocation: starts ~1:1, scales toward whichever side the orderbook + tape support; final dominance ratio is the output of the loop, not an input.
6. Exit: never. Hold to resolution. Zero SELL trades.
7. Risk management: implicit - capped clip + 5-minute time horizon means max single-market loss is bounded near $887 (the largest observed total per-market USDC).
8. Edge source: directional skew on excess shares, not spread capture (which is structurally negative).
9. Weaknesses: hedge tax of ~$1.2M gross / ~$437K avoidable; smooth 1.5-cent skew at the round-number ticks; no exit logic means no take-profit if a winning leg moves to $0.99 mid-window.
10. Rebuild parameters: see playbook tables.

Closing Read

SLIP-ME is the cleanest example in this library of a bot that looks like a market-maker and earns like a directional bot. The orderbook footprint is identical to a passive maker; the P/L attribution is identical to an aggressive trend-follower. Operators reverse-engineering this wallet by orderbook footprint alone will conclude "MM, copy his quotes, undercut his spread" - and will lose money, because the spread leg here loses money. The thing to copy is the late-window dominance load, not the early-window paired entry. The early entries are infrastructure; the late entries are the trade.

Total: 188,932 BUYs · $4.32M deployed · +$111,822 P/L · +2.59% ROI · 22/22 weeks green · single-symbol, single-duration, all-buy, hold-to-expiry book.

Wallet: 0x476639d9845d7a0261cb005dae6473f089ff5a03 Window: 2026-04-05 → 2026-04-27 (23 calendar days) Methodology: All metrics computed directly from SLIP-ME_trades.csv against the Polymarket activity API resolutions. P/L = shares-pay-$1 settlement on resolved BUYs (no SELL trades exist in this book; cash-flow and shares-pay-$1 settlement collapse to the same number). No deduplication. No simulation.

Headline

Phase 1 - Trader Profile

Trade-Size Distribution (USDC per BUY ticket)

Inter-Fill Gap (same condition_id + outcome)

Side Split

Active-Day Profile

22 of 23 days profitable. Sole losing day: 2026-04-12 (−$2,898). Sole inactive day: 2026-04-19. Activity tail-off in the second half is visible (week 4 averages ~5,800 trades/day vs. weeks 1–3 average ~10,400).

Phase 2 - Strategy Identification

Classification: Hybrid Both-Sides MM costume + Directional/Latency-Arb engine. Earns directionally, not from spread.

Phase 3 - Dominance Ratio Analysis

Dominant-side win rate climbs monotonically from 67% to 99.8%. Avg paired cost dips below $1.00 only at the most extreme skew (≥ 5×).

Both-Sides Participation

Phase 4 - Entry Price Analysis

10-Bucket Histogram

Top 10 Cent-Bucket Concentration

No single tick carries more than 2.25% - bot does not anchor at a fixed price. Price distribution is smooth across $0.45–$0.70 with a small bump at round-number ticks.

Coarse Price Split

Paired-Cost Bands (resolved 2-side markets)

Below paired cost $1.00 the strategy is highly profitable. Above $1.05 it is unprofitable. The fact that dom-win rate is roughly constant at ~87% across all bands means: the bot's directional accuracy doesn't degrade in expensive markets - but the cost of holding the paired position eats the directional gain.

Phase 5 - Category & Duration Breakdown

Category

100% BTC inside this date window. (Earlier periods saw ETH/SOL activity per the pipeline diagnostic, but the 2026-04-05 → 2026-04-27 slice is single-symbol.)

Duration

Single-duration: 99.95% on the 5-minute book.

Assessment Badge

Phase 6 - Timing & Execution Analysis

Entry Timing Within the 5-min Window

Median seconds-to-close at entry: 151 s. Mean: 150 s. Bot loads early and late, with the late-window buys being the most efficient.

Burst Cadence

Hour-of-Day (UTC)

Win-rate spread: 52.59% (15 UTC) → 54.05% (07 + 20 UTC). Single losing hour by P/L: 05 UTC (−$67). All other hours net positive.

Day of Week

Weekday volume is roughly 2.5× weekend volume. P/L scales accordingly. No day is unprofitable.

Phase 7 - Filter Experiments

Headline: a single dominance-bucket filter (block underdog buys when skew ≥ 2×) lifts ROI from +2.59% to +12.69% (4.9× improvement). Full per-filter narrative in SLIP-ME_filters.md.

Phase 8 - Rolling Window Analysis

7-Day Rolling P/L

22 of 22 windows positive. Min window: +$23,162 (window ending 2026-04-27). Max: +$72,483 (window ending around 2026-04-15).

15-Day Rolling P/L

22 of 22 windows positive. Min window: +$57,359. Max: +$95,196 (window ending around 2026-04-19).

Drawdown

Longest drawdown streak: 1 day (2026-04-12, single losing day surrounded by winning days). No multi-day drawdown observed.

Phase 9 - P/L Decomposition

Spread leg loses $97K. Directional excess-shares leg gains $204K. Net +$107K on 2-side markets, plus +$5K from one-sided markets and the 15m bug.

Phase 10 - Top Markets

Top 10 by Volume

Top 10 Winning Markets

Top 10 Losing Markets

Largest single-market loss: −$614 (April 8, 1:40AM-1:45AM ET). Largest gain: +$495 (April 9, 10:00AM-10:05AM ET). The book's per-market P/L distribution is bounded - no single market can move the daily P/L by more than ~$500.

Structural P/L Decomposition (annual run-rate equivalent)

These extrapolations assume the activity profile of 2026-04-05 → 2026-04-27 holds, which it has not over the longer window (the bot tapered from ~13K trades/day in week 1 to ~5K in week 4). Treat the annual estimate as an upper bound assuming current intensity continues.

Footnotes

• Why no SELL accounting: The book has zero SELL trades. All P/L is from share settlement at expiry.
• Why 2 markets show 15m duration: likely a routing-bug edge case where the bot's universe-discovery code briefly grabbed 2 longer-duration markets. P/L impact: −$43.
• Why 1 day is inactive (April 19): the bot took a 24-hour gap. No P/L accrued; the surrounding rolling windows remained green.
• Why activity tapers in week 4: unknown. Likely a deliberate scale-back or operational decision; ROI per trade did not deteriorate.
• Resolution oracle latency: the 2,936 post-close trades represent matching-engine settlements happening in the 30–120 second window between the 5-min close and the resolution oracle firing. Net loss: −$528.

Purpose: Test whether single-rule filters on the resolved-BUY set would have improved realized P/L. Each row reports qualifying trades, win rate, total P/L, deployed capital, and ROI vs. the unfiltered baseline.

Headline: SLIP-ME's portfolio is NOT at filter terminal velocity - unlike LIL222 or other single-tick bots whose every fill is already optimized by the entry trigger, SLIP-ME deliberately funds two distinct loss-leader buckets (the underdog hedges and the spread-maker presence) that mask the underlying alpha. A single dominance-bucket filter would have lifted realized ROI by 4.9× without touching the rest of the strategy.

The Baseline

Filter Ledger

Filter-by-Filter Commentary

Price-band filters

Price 0.30–0.70 (the "fair-value middle") - keeps 60% of trades and 59% of capital, returns +2.83% ROI. Marginal lift only. The middle band is where the bot's spread-maker presence sits - paired cost is closest to $1.00 here, so the spread leg is the most expensive-per-share, and the directional alpha is diluted. Removing the tails (cheap hedges and high-conviction loads) doesn't help because the tails are the actual alpha sources.

Price 0.40–0.60 (tightest middle) - slightly better (+3.33%), still nothing dramatic. Same reason.

Price < 0.30 (cheap-side hedges) - +11.13% ROI on $276,919 deployed, with a 20.3% win rate. This is the lottery-ticket leg working as designed. When SLIP-ME buys the underdog at $0.10 and that side actually wins, the share pays $1.00 - a 9× return per share. Even at a 20% win rate, the math is 0.203 × $0.90 + 0.797 × −$0.10 = +$0.103 per dollar ≈ +10% expected ROI per ticket. Realized matches expectation almost exactly. Filter does not help - it is the correct strategy for that price band. Keeping it is correct. The note here is that this band's +$30,811 of P/L on $277K of capital punches well above its weight: it's 7.1% of capital and 27.5% of P/L.

Price > 0.70 (high-conviction late loads) - 41,750 trades at 82.5% win rate, but only +0.62% ROI. The win-rate looks great in isolation but the per-ticket margin is razor-thin. Buying at $0.85 returns $0.15 per share when right and loses $0.85 when wrong: 0.825 × $0.15 + 0.175 × −$0.85 = +$0.0238 per dollar. The realized number lands at +0.62% because the actual entry price within the band is somewhat lower than $0.85 (mostly $0.70–$0.80). Filtering this out doesn't help - it's marginally profitable and provides the late-window presence that signals to other bots that SLIP-ME is committed. Removing it would also remove the dominance-skew payoff from the markets where SLIP-ME's late loads cement his ≥ 2× skew.

Dominance filters (the headline)

Dominance ≥ 2×, dominant side only - +36.08% ROI on $1,310,221 deployed = +$472,765 realized P/L on a 96.7% win rate. This is the single sharpest filter in the entire experiment. By taking only the trades on the side SLIP-ME ended up overweighting in markets where the final skew was ≥ 2×, you keep all the directional alpha and dump every losing-side hedge in those markets. The filter is large enough to matter (22% of all trades, 30% of all capital).

Dominance ≥ 3×, dominant side only - +32.76% ROI on $603K deployed. Smaller universe but cleaner. Win rate 98.8%. Almost identical ROI to the ≥ 5× filter.

Dominance ≥ 5×, dominant side only - +33.08% ROI on $175K deployed. Win rate 99.8% - only 1 of 520 markets in this bucket had the dominant side lose. As filters narrow, win rate climbs but absolute P/L falls.

Dominance lift is real but with a caveat - these filters are retrospective. The dominance ratio is observable only at the end of the market, after all fills have happened. To use the dominance filter live, you need a way to predict (or observe in near real-time) which side will be dominant by close. SLIP-ME's bot already does this implicitly via its loading mechanic - by minute 4 the dominance ratio is largely determined and the bot has already loaded most of the dominant side. But a clean implementation would need to either (a) re-run SLIP-ME's loading logic and gate the underdog tickets, or (b) use a real-time orderbook+tape signal as a proxy for the eventual dominance.

Underdog blocking - the operationally clean filter

"Underdog blocking @ ≥ 2× skew" is the implementable version of the dominance filter: keep every fill the bot would normally take in low-skew markets (where there's no clear winner), but block the underdog-side fills in markets that have already developed ≥ 2× skew. This does not require predicting the dominant side - it just requires knowing the current skew at the moment of the prospective trade.

In plain English: by declining to throw cheap hedges into markets where the bot has already committed to a side, the operator would have saved $437K in capital deployment and added $381K in P/L. The hedges in those markets are insurance against the rare 0.2–4.4% reversal in the dominant side; in net they cost more than they pay.

The tradeoff: without those hedges, the worst-case single-market drawdown gets uglier. In a market that goes 5×+ skewed toward Up at minute 4 and then reverses on a final-30-second BTC spike, the underdog-blocked book takes the full hit on the dominant Up shares. With the hedges in place, the bot recoups some of the loss via the cheap Down lottery tickets paying $1.00. The hedges are insurance; this filter cancels the insurance. Whether the operator should make the change depends on their tolerance for the 1-in-200 worst-case market.

Symbol / duration filters (no signal)

BTC only / 5m only - these are the entire universe (188,932 of 188,932 trades and 188,835 of 188,932). They aren't filters; they're the universe definition. ROI is identical to baseline because the universe is identical to baseline.

ETH only / SOL only - zero qualifying trades in this date window. The bot is BTC-only inside 2026-04-05 → 2026-04-27. Filter rejected on insufficient sample, not lack of edge.

15m only - 97 trades returning −$43. A trivial subset, almost certainly a routing bug. Filter eliminates 99.95% of the book.

Hour filters (no signal)

Hourly win rate spreads only 52.59% → 54.05% end-to-end. No 4-hour window stands out as significantly better or worse. The bot is time-of-day agnostic - every hour produces 6,500–9,700 trades and 52.5–54.1% win rate. No hour filter improves the book by more than ~0.3 pp ROI, and the experiment was not run on individual 4-hour blocks because the spread is too narrow to justify the model complexity.

What This Filter Analysis Reveals About the Strategy

1. The bot's edge is concentrated in the high-skew markets, not in the low-skew ones. 22% of trades carry 423% of the P/L (in the ≥ 2× dom-side-only band). The other 78% of trades are net-positive but barely.

1. The losing-side hedges in already-skewed markets are net negative. The bot pays for insurance it nearly never collects on - and when it does collect (the 4.4% of ≥ 2× markets where the dominant side lost), the lottery payout is small relative to the cumulative cost of the hedges. Removing those specific hedges is the single most impactful filter intervention.

1. The cheap-hedge band ($p < 0.30$) is its own profit center, separate from the dominance filter. Those bets earn +11.13% ROI as a standalone strategy. They should not be removed even though they have a 20.3% win rate.

1. The mid-band and high-band trades (p $0.30–0.95$) are essentially flat-to-slightly-profitable on their own but they generate the orderbook presence that sets up the high-skew dominance markets. Without them, the bot would not be in position to develop the 99% win-rate dominance loads. They are infrastructure cost; their P/L should be measured net of the dominance alpha they enable, not standalone.

1. No naive single filter (price, hour, category) lifts the book by more than ~1 pp ROI. The dominance filter is the only one that produces dramatic lift, and it requires either retrospective skew measurement or a real-time skew tracker.

Recommendation for an Operator Replicating This Strategy

Run the strategy as SLIP-ME runs it, with one specific addition: a real-time current_dominance_ratio tracker per market. When the ratio crosses 2.0×, gate underdog buys for the remainder of that market's lifecycle. Continue allowing dominant-side and small-skew (< 2×) trades unchanged. Expected lift: from +2.59% to ≈ +12% ROI on the same execution machinery.

Do not filter out the cheap-hedge band ($p < 0.30$) globally - it is a profit center on its own. Do not filter out the late-window high-conviction band ($p > 0.70$) - it is where the dominance gets cemented. Do not filter by hour - there is no edge there. Do not add SELL logic - the bot's economics depend on hold-to-expiry settlement and adding SELLs would introduce timing and inventory risk that the current spec sidesteps.

The single non-obvious risk of the proposed filter: it concentrates the worst-case market loss. SLIP-ME accepts a small per-market hedge tax in exchange for a softer worst-case. Removing the hedges makes the average market more profitable but the 1-in-500 worst market noticeably worse. Backtest the filtered version on out-of-sample data with the worst-case loss expectation in mind before running it live with size.

Wallet: 0x476639d9845d7a0261cb005dae6473f089ff5a03 Window analyzed: 2026-04-05 → 2026-04-27 Realized: +$111,822 P/L · +2.59% ROI · 22 / 22 weeks green · 188,932 BUYs · 0 SELLs

Operator Brief (read this paragraph if nothing else)

SLIP-ME is a directional market-maker on Polymarket's btc-updown-5m book that fires roughly 8,500 BUY orders per active day, ~40 fills per market, with a fixed clip size of $10–$60 and a hold-to-expiry exit. It enters both legs of every market within seconds of open, then accumulates the side that the BTC tape is moving toward, ending with 2× or greater dominance toward the eventual winner in 60% of all 5-minute markets traded. Edge does not come from spread capture - paired cost averages above $1.00 and the spread leg is structurally a $4K/day drag - it comes from the excess shares on the dominant leg, which paid +$204K of directional alpha vs. the −$98K spread loss. The strategy is profitable, exceptionally consistent (every 7-day rolling window green), and bottlenecked at modest scale ($4.3M of cumulative deployed capital over 23 days, max single fill $117). To replicate, you need: a sub-second BTC price feed, a CLOB-API-connected executor that can fire 8,500 orders/day, ~$50K of working capital, and the discipline to hold every share to expiry without any take-profit logic.

Pseudocode (the entire bot in 30 lines)

# Per-market loop. Run one of these per active 5-min BTC market.
          while market_seconds_remaining > 0:
              yes_price, no_price = polymarket_clob.bbo(market.condition_id)
              btc_tick           = btc_tape.last_tick()
              yes_share_count, no_share_count = position_state(market.condition_id)
              skew = max(yes_share_count, no_share_count) / max(min(yes_share_count, no_share_count), 1e-9)
          
              # Phase 1: open both legs in the first 30 seconds of the market
              if market_seconds_remaining > 270 and (yes_share_count == 0 or no_share_count == 0):
                  # Lift whichever side has cheap depth first; second side follows in <60s
                  side = "Yes" if yes_price <= no_price else "No"
                  place_buy(market, side, shares=60)         # ~$25–$40 clip
                  continue
          
              # Phase 2: trend-follow during minutes 1–4
              elif market_seconds_remaining > 60:
                  # Add to whichever side the BTC tape is moving toward
                  if btc_tick.direction == "up" and yes_price < 0.95:
                      place_buy(market, "Yes", shares=60)
                  elif btc_tick.direction == "down" and no_price < 0.95:
                      place_buy(market, "No", shares=60)
                  # Cheap-hedge insurance: load the cheap underdog when it gaps below $0.20
                  elif yes_price < 0.20:
                      place_buy(market, "Yes", shares=120)   # lottery ticket
                  elif no_price < 0.20:
                      place_buy(market, "No", shares=120)
          
              # Phase 3: high-conviction final-90s loads on the dominant side ONLY
              elif market_seconds_remaining > 0:
                  dominant_side = "Yes" if yes_share_count > no_share_count else "No"
                  dom_price = yes_price if dominant_side == "Yes" else no_price
                  if dom_price < 0.97:                       # avoid silly-tight quotes
                      place_buy(market, dominant_side, shares=60)
          
              sleep(2)  # median inter-fill gap
          
          # At market close: hold every share. No SELL. Resolution oracle settles.

The five-minute outer loop runs for every active btc-updown-5m market and exits when the market closes. The inner work is two pieces of state (own position + current BBO) and one external signal (BTC tape direction). No model. No regression. No prediction. The "edge" is the loop's reaction time to BTC tape ticks.

Filtered variant (recommended): add one gate before the Phase-2 underdog cheap-hedge block:

if skew >= 2.0:
              # market has already declared a side; don't fund underdog hedges
              continue

This single addition lifts realized ROI from +2.59% to ~+12.69% on identical execution machinery. See SLIP-ME_filters.md for the full justification.

Bankroll Math

This strategy is highly capital-efficient per market ($30–$890 per market) but scales linearly with the number of markets you're willing to participate in concurrently. SLIP-ME participates in ~140 markets per day (≈ half the available 5-min BTC inventory). The total deployed capital at any single moment is bounded by the number of open markets the bot is sitting in (max ~5 at a time, since markets are 5-minute non-overlapping windows) times the per-market clip cap (~$890).

The scaling cliff is between $250K and $1M working capital. At ~$1M deployed, the bot would consume noticeable orderbook depth on every fill, and other bots would start adversely-selecting against the larger clip size. The 5-min BTC orderbook is shallow - it cannot cleanly absorb a market-maker running $500/clip without observable price impact, which would shrink the realized ROI from ~12% to closer to ~2–3%. This strategy is not a $10M strategy. It is a $50K–$250K strategy.

The +2.59% ROI baseline scales linearly within the $50K–$250K band; the +12.69% filtered variant likewise. Beyond $250K of working capital, expect ROI compression as the bot becomes the marginal price-setter on its own fills.

Daily turnover (relevant for gas/fees): SLIP-ME averaged ~8,588 trades / active day. At Polymarket's USDC stablecoin gas profile (≈ $0.02 / trade in worst-case L2 fees), that is ~$170/day in transaction costs. Already netted out of the +$5,100/day P/L. At higher trade frequencies (>20K trades/day), gas would start eating meaningfully into the margin.

Operational Requirements

What This Playbook Deliberately Does NOT Include

A replicator's instinct is to add complexity. Resist it. The following were considered and deliberately excluded because the data shows SLIP-ME's strategy works without them and adding any of them would either (a) reduce ROI by adding constraints to a working loop, or (b) introduce risk surface that the current spec sidesteps.

• No SELL logic / no take-profit. SLIP-ME placed 0 SELL orders. Adding take-profit logic ("sell my dominant-side shares back at $0.99 instead of holding to $1.00") sounds like free money but it isn't - the orderbook depth at $0.99 is thin and inconsistent, and selling locks in the spread loss instead of recovering it via the $1.00 settlement. Hold to expiry.

• No directional model on BTC. The bot does not have a "BTC will go up in the next 5 minutes" prediction. It reacts to the BTC tape. There is no ML model, no regression on prior ticks, no calibration. The tape itself is the signal. If you build a directional BTC model and overlay it, you will trade against your own reactive engine and degrade the win rate.

• No Kelly sizing / no edge-weighted bets. The clip size is fixed at ~60 shares. SLIP-ME's max single fill across 188,932 trades was $117 - meaning he never sized up on a single market regardless of how confident the dominance signal was. Don't introduce confidence-weighted sizing; the per-market position cap (~$890 cumulative) is the entire risk-management envelope.

• No category expansion (yet). SLIP-ME runs btc-updown-5m only. The 97 trades on btc-updown-15m lost money (−2.78% ROI on $1,543). Don't expand to ETH, SOL, or longer-duration markets without re-running the full analysis on those instruments - the orderbook microstructure differs and the loop's parameters are tuned to the 5-min BTC tape rhythm specifically.

• No copy-trading or signal-following. SLIP-ME does not lag any other wallet. The strategy is self-contained. Don't try to follow him by mirroring his fills - by the time you observe a SLIP-ME fill on-chain (≥ 1-block latency), the orderbook depth he just lifted is gone and you'll be paying the next-tick price with no edge.

• No reinforcement learning / no adaptive parameters. The clip size, skew threshold, and underdog-hedge price band are constants. SLIP-ME does not learn within or across markets. Adding adaptation introduces non-stationarity that breaks the consistency profile (22/22 green weeks).

• No stop-loss. The 5-minute time horizon and ~$890 max per-market exposure cap function as a hard time-based stop. The bot cannot lose more than ~$890 on any single market and the worst observed market loss in the analysis window was −$614 (April 8, 1:40AM-1:45AM ET). Adding an explicit stop-loss within a single 5-min market would only consume the late-window high-conviction load slot, where the highest ROI is realized.

• No hour-of-day filter. Hourly win rate spreads only 52.59% → 54.05% across the 24-hour cycle. There is no edge from time filtering. Run 24/7.

• No SELL-side hedging on related markets. SLIP-ME does not run pairs trades against btc-updown-15m, btc-1h, or any other instrument to hedge his 5-min exposure. Each market is a standalone payoff. Don't introduce cross-market hedging without a full analysis of the correlation structure on btc-updown-* markets.

• No position roll-overs. Each market resolves and the cash returns to the wallet. Don't pre-commit capital to "the next market" before the current one settles.

Implementation Checklist

[ ] Polymarket CLOB API key and signed wallet (≥ $50K USDC)
          [ ] BTC price feed at sub-second latency (multiple-source redundancy)
          [ ] Per-market 5-minute outer loop with 2-second inner tick
          [ ] Position state tracker per condition_id
          [ ] Skew calculator (yes_shares / no_shares, real-time)
          [ ] BBO subscription per active market
          [ ] Order placement function with retry + nonce management
          [ ] Optional: skew >= 2.0 underdog gate (recommended for +10pp ROI)
          [ ] Hold-to-expiry settlement reconciliation
          [ ] Per-day P/L and per-market dominance dashboard
          [ ] Daily reconciliation with on-chain settlements
          [ ] Killswitch for BTC feed loss / RPC outage

Build the inner loop first; verify it can place 60-share BUY orders in a paper environment without errors. Then verify the position state and skew calculator stay coherent across simultaneously open markets. Then dial up capital from $5K → $25K → $50K over a 2-week period, watching the daily P/L curve match expectation.