Bonereaper (0xeebde7a0) - Strategy Reverse-Engineering Report

// 001 / Method

Ten-Phase Reverse-Engineering Protocol

What we searched for, executed programmatically over both CSVs in DATA_DROP/.

Every phase below was executed programmatically over both CSVs in DATA_DROP/ - 1.34M trades after deduping 3 duplicate transaction hashes across the two exports. Every market in this dataset is fully resolved; no “Open” remnants.

Phase	What we were looking for
Phase 1	Trader profile - scale, trade-size distribution, activity clock, inter-trade gaps, buy vs sell ratio.
Phase 2	Core strategy archetype - market-making, directional betting, latency arb, copy trading, DCA.
Phase 3	Dominance-ratio analysis - does the larger side of a both-sides market win more often as conviction rises?
Phase 4	Entry-price discipline - where does capital cluster, does win rate track implied probability, is paired cost below $1.00?
Phase 5	Category / market-type concentration - single vertical or distributed?
Phase 6	Timing and execution - burst patterns, second-side hedge lag, first-to-last trade span per market.
Phase 7	Filter experiments - which simple rules isolate the highest-edge subset?
Phase 8	Rolling-window consistency - is the edge durable across 7-day and 15-day slices?
Phase 9	P&L decomposition - where does the money actually come from?
Phase 10	Strategy specification - a reconstructable rule set describing what this wallet is doing.

// 002 / Executive answer

Fair-value market-making, with a real signal

The single sentence that summarises 1.34M trades.

This wallet is a fair-value-driven, both-sides market-making bot on 5- and 15-minute crypto up-or-down binaries.

It deploys small tickets (median $5.56), pairs both outcomes on 96.7% of markets within a median 10 seconds, almost never sells on the book, and holds every share to resolution. The evidence for a real fair-value signal is unambiguous: as the USDC imbalance between sides grows, the dominant side’s resolved win rate climbs smoothly from 59% at 1.0-1.5x to 97.7% at 5x+. Entry-price band win rates also track implied probabilities within ~2 points across the full $0-$1 range - a calibration signature you only get from a working model.

Across the 22-day window: 22 of 23 days profitable, every rolling 7-day and 15-day window green, best day +$39,767, worst −$2,507. The strategy earns ~$20K/day on a $1.3M/day trade flow.

In plain English: the bot decides which outcome is underpriced using an external fair-value reference, buys both sides to lock in spread where possible, and lets the dominant side run when conviction warrants it. The bigger the lean, the stronger the signal - and outcomes validate that lean almost perfectly at the top of the conviction curve.

// Phase 1 / Trader profile

High-frequency bot, tiny tickets, crypto-only

Scale, trade-size distribution, activity clock, inter-trade gaps, category and duration concentration.

Scale

1,342,304 trades over 21.99 days (~61,000/day, ~42/minute).
14,168 unique markets - ~644/day.
$29,091,910 buy notional. $22,783 sell notional (0.08% of gross; 2,866 trades).
Per-market density: median 83 fills, P95 209.

Trade-size distribution

Median ticket: $5.56. Mean: $21.69. P95: $54.78. Max: $15,000.
Top 5% of trades carry 57.3% of capital - strongly power-law.
Small probes dominate the count; large fills reserved for deeper books.

Timing (UTC)

Active all 24 hours - no human gaps.
Peak 13:00-17:00 UTC (US morning/midday ET, ~68K trades/hr).
Trough 20:00-22:00 UTC (~41K trades/hr).
Smooth distribution - automation, not a human operator.

Inter-trade speed (same market + outcome)

Median gap: 0 seconds. P90: 22s. Mean: 9.0s.
78.9% of consecutive fills under 10s; 97.5% under 60s.
Sub-human latency. Combined with small tickets → working liquidity in tight bursts.

Category & duration concentration

Underlying	Trades	USDC	% Capital
BTC	934,840	$24.997M	86.0%
ETH	403,291	$4.023M	13.8%
SOL	4,173	$72,554	0.2%

Duration	Markets	USDC	% Capital
5 min	10,224	$21.55M	74.1%
15 min	3,599	$6.85M	23.5%
4 hr	78	$99K	0.3%
other	267	$592K	2.0%

100% crypto directional binaries. No sports, politics, or event-driven markets. 98% of capital in 5-min or 15-min BTC/ETH windows.

// Phase 2 & 3 / Archetype + dominance

Hybrid market-maker with a real, monotonic signal

Both-sides participation, dominance ratio, and the directional-edge test.

Both-sides participation

96.74%

13,706 of 14,168 markets had both outcomes bought. Clearly in the market-making archetype (prompt threshold >60%).

Median dominance ratio

2.81x

Heavier side typically ~2.8x the lighter side. P90 ratio is 22.7x - heavy tail of high-conviction lopsided positioning.

Second-side hedge lag

10s

93.2% of both-sides markets pair within 60 seconds. Hedge is part of the entry, not an afterthought.

Dominance-ratio buckets - the directional-edge test

If the trader has a real signal, the dominant side’s win rate should rise with the ratio. Prompt thresholds: ≥70% at 2.0x+ and ≥85% at 3.0x+. This wallet exceeds both decisively.

Ratio bucket	Markets	Dom-side win rate	% markets profitable	Avg P&L / mkt	Total bucket P&L
1.0-1.5x	2,988	59.0%	40.6%	−$5.23	−$15,638
1.5-2.0x	1,879	72.2%	46.6%	+$9.04	+$16,984
2.0-3.0x	2,343	81.0%	51.3%	+$10.54	+$24,704
3.0-5.0x	2,078	91.2%	67.1%	+$59.38	+$123,390
5.0x+	4,418	97.7%	76.2%	+$65.62	+$289,916

This is a textbook fair-value-driven inventory trader. The dominant-side win rate rises smoothly 59% → 72% → 81% → 91% → 97.7% as the ratio climbs - exactly what the prompt predicts for a bot that (a) builds an external probability estimate, (b) sizes the two sides inversely to perceived value, and (c) hedges around a paired-cost ceiling. Per-market P&L scales from −$5.23 to +$65.62 - the ratio is a genuine conviction dial, not a byproduct.

Paired-cost discipline

Median paired cost: $1.006. Mean: $1.027. 47.3% below $1.00; 35.2% below $0.97.
Sub-$1 markets average +$53.72 P&L / market; above-$1 markets still average +$12.58.
Both regimes profitable. Below parity, spread capture works on matched shares; above parity, the directional edge carries the P&L.

// Phase 4 / Entry price

Win rate tracks implied probability within ~2 points at every band

Capital allocation, win rate, and proxy ROI per share across the full $0-$1 entry-price range.

Price band	Capital	% of total	Trades	Win rate	Implied prob	Proxy ROI / share
$0.00 - $0.10	$205K	0.70%	83,323	5.9%	5%	+$0.0012
$0.10 - $0.20	$451K	1.55%	120,521	14.5%	15%	−$0.0011
$0.20 - $0.30	$712K	2.45%	132,373	25.3%	25%	+$0.0066
$0.30 - $0.40	$1.10M	3.77%	147,902	36.0%	35%	+$0.0135
$0.40 - $0.50	$1.85M	6.36%	183,012	46.7%	45%	+$0.0177
$0.50 - $0.60	$2.71M	9.32%	210,152	55.9%	55%	+$0.0166
$0.60 - $0.70	$2.37M	8.14%	152,350	66.1%	65%	+$0.0173
$0.70 - $0.80	$2.25M	7.73%	115,565	75.4%	75%	+$0.0108
$0.80 - $0.90	$2.43M	8.36%	86,374	85.7%	85%	+$0.0140
$0.90 - $1.00	$15.02M	51.6%	107,866	97.5%	95%	+$0.0068

Win rate hugs the entry-price implied probability within ~2 percentage points at every band - strong evidence of a well-calibrated internal fair-value model. The $0.30-$0.90 range shows the highest proxy ROI per share (1.1-1.8¢). The $0.90+ band captures the most capital (52%) at the thinnest margin (0.7¢/share); volume compensates.

// Phase 5 / Category

100% crypto up-or-down. Single-vertical focus.

Distribution across underlying assets, durations, and market types.

All 14,168 markets are crypto price binaries - BTC (86% of capital), ETH (14%), SOL (trace). No sports, politics, or event-driven markets. 97.6% of markets are either 5-minute or 15-minute windows. This strategy is built for one market family only.

Implication

A category filter cannot rescue this wallet’s P&L - the category is the strategy. Anyone replicating this needs to accept the same vertical constraint.

// Phase 6 / Timing & execution

Dense bursts per market, instant pairing

Per-market accumulation span and second-side hedge lag.

Per-market accumulation

Median span first-to-last fill: 274s (4.6 min).
P95 span: 882s (14.7 min).
Median fills per market: 83. P95: 209.
Matches the 5-min / 15-min market windows - the bot saturates each market during its entire life with small child orders.

Second-side hedge lag

Median lag (first-side → first opposite-side fill): 10s.
93.2% of both-sides markets paired within 60s.
Hedge is essentially simultaneous; the ratio is chosen at entry, not reactive.

// Phase 7 / Filter experiments

Dominance ratio is the cleanest edge; 1.0-1.5x is the only losing bucket

Simple filter rules tested for incremental P&L and per-market economics.

Filter	Qualifying markets	Total P&L	Avg / market	% profitable
All resolved markets (baseline)	14,168	+$443,754	+$31.32	59.7%
Dominance ratio ≥ 1.5x (skip near-balanced)	10,718	+$454,995	+$42.45	63.8%
Dominance ratio ≥ 2.0x	8,839	+$438,011	+$49.55	67.5%
Dominance ratio ≥ 3.0x	6,496	+$413,306	+$63.62	73.3%
Dominance ratio ≥ 5.0x	4,418	+$289,916	+$65.62	76.2%
Sub-$1.00 paired cost only	6,488	+$348,545	+$53.72	58.2%

Key takeaways: (1) the 1.0-1.5x bucket is the only bucket that loses money - skipping it adds $11K to baseline P&L while retaining 76% of markets. (2) Higher-conviction buckets have materially better per-market economics but fewer markets - the real-money sweet spot is ratio ≥ 3.0x (avg +$63/mkt, 73% profitable). (3) The sub-$1.00 paired-cost filter is strong but orthogonal to the ratio filter; stacking them should be tested in a follow-up.

// Phase 8 / Rolling windows

22 of 23 days green. Every rolling 7-day & 15-day window profitable.

Daily realized P&L (UTC) and rolling-window consistency.

Profitable days

95.7%

22 of 23 calendar days green. Only 2026-03-29 red (−$2,507).

Rolling 7-day windows

100.0%

Worst 7-day: +$19,204. Best: +$164,888.

Rolling 15-day windows

100.0%

No drawdown periods observed.

Daily realized P&L (UTC)

Date	P&L
2026-03-25	+$19,204
2026-03-26	+$23,099
2026-03-27	+$14,884
2026-03-28	+$30,161
2026-03-29	−$2,507
2026-03-30	+$28,695
2026-03-31	+$34,137
2026-04-01	+$6,019
2026-04-02	+$17,704
2026-04-03	+$14,751
2026-04-04	+$21,688
2026-04-05	+$10,298
2026-04-06	+$39,767
2026-04-07	+$19,018
2026-04-08	+$30,219
2026-04-09	+$22,564
2026-04-10	+$21,334
2026-04-11	+$21,413
2026-04-12	+$3,897
2026-04-13	+$19,507
2026-04-14	+$31,032
2026-04-15	+$20,599
2026-04-16 (partial)	+$2,028

// Phase 9 / P&L decomposition

Many small wins, occasional large losses - net positive by a wide margin

Where the money actually comes from.

Component	Amount	Note
Total buy notional (22 days)	$29,091,910	1,339,438 buy trades.
Total sell notional	$22,783	2,866 sells - 0.08% of flow. Effectively resolution-hold.
Net USDC deployed	$29,069,126	Buys minus sells.
Realized payouts from resolution	$29,512,880	Winning shares × $1.00.
Realized P&L	+$443,754	+1.527% gross return in 22 days.
Per-market avg / median P&L	+$31.32 / +$15.60	Mean > median → occasional big wins skew right.
Best / worst single market	+$3,917 / −$15,861	Worst loss exceeds best gain - market-maker absorbing occasional adverse-selection in exchange for steady small wins.
Hedge tax (non-dominant USDC deployed)	$5,001,226	Capital on the lighter side of both-sides markets. Offset by spread capture + dominant-side wins.

Annualized extrapolation (simple): +1.527% × (365/22) ≈ +25.3% annualized on gross deployed capital - thin per-market, but high velocity (~644 markets/day) scales the edge into six-figure monthly P&L.

// Phase 10 / Spec

What this bot is doing, stated precisely

A reconstructable rule set describing what this wallet is doing.

One-sentence summary

A fair-value-driven, both-sides market-making bot on 5- and 15-minute crypto up-or-down binaries that scales directional conviction into an asymmetric USDC allocation across outcomes, holds every share to resolution, and compounds a thin per-market edge across ~644 resolved markets per day.

Market universe:     Short-duration crypto up/down binaries.
                     BTC (86%), ETH (14%), trace SOL.
                     98% of capital in 5-min or 15-min windows.

Signal source:       External fair-value / implied-probability model.
                     Not visible in CSV; inferred from the near-perfect
                     alignment between entry-price implied probability
                     and resolved win rate across every band, and from
                     the monotonic ratio-vs-win-rate curve.

Entry trigger:       Market opens for trading AND fair probability
                     diverges enough from the book to clear the
                     paired-cost ceiling at target position size.

Sizing model:        Power-law probes (median $5.56 ticket).
                     Top 5% of trades carry 57% of capital.
                     Size scales with available book depth and edge.

Both-sides logic:    96.7% both-sides rate.
                     Second side enters within median 10s of the first.
                     Ratio is chosen at entry, not reactive.

Dominance pattern:   Median 2.81x. Ratio IS the conviction dial.
                     Higher ratio -> dominant side wins more:
                       1.0-1.5x : 59.0%   -$5.23 / mkt
                       1.5-2.0x : 72.2%   +$9.04
                       2.0-3.0x : 81.0%   +$10.54
                       3.0-5.0x : 91.2%   +$59.38
                       5.0x+    : 97.7%   +$65.62

Paired-cost target:  Median $1.006. 47.3% below $1.00.
                     Both regimes profitable; sub-$1 earns ~4.3x
                     more per market ($53.72 vs $12.58).

Exit strategy:       None on-book (<0.1% sells).
                     Every position held to resolution.

Edge source:         (1) Well-calibrated external fair-value model
                         prices both sides across $0-$1 range.
                     (2) Ratio asymmetry lets the bot earn on
                         conviction without forfeiting spread on
                         balanced markets.
                     (3) High velocity (644 markets/day) converts
                         thin per-market edge into durable P&L.

Known weakness:      1.0-1.5x ratio bucket (near-balanced) loses
                     money on average (-$5.23/mkt). Skipping it
                     would improve aggregate P&L by ~$11K without
                     dropping more than 24% of market count.

Rebuild parameters:  - Only crypto up/down 5-min & 15-min binaries.
                     - Target paired cost < $1.00 ideal, < $1.03 max.
                     - Dominance ratio >= 1.5x to avoid loss bucket.
                     - Fill with small probes, ~10s second-side pairing.
                     - Hold to resolution; no book-level exit logic.
                     - Activity clock: 24/7 with soft peak 13-17 UTC.

// 003 / Conclusion

Most honest final conclusion

What we observed, what we inferred, and what would still need live data to confirm.

This wallet is a well-executed reference for a fair-value inventory bot on crypto-price short-duration binaries. Every phase of the 10-phase protocol produces a consistent picture:

Observed 1.34M trades, 99.8% buys, 96.7% both-sides, 10s median hedge lag - confirmed bot behavior.
Observed 8,461 of 14,168 markets profitable (59.7%). Median per-market P&L +$15.60, mean +$31.32.
Observed Dominant-side win rate climbs 59% → 97.7% as ratio grows - clean directional signal.
Observed Win rate tracks entry-price implied probability within ~2 points across every price band.
Observed Realized P&L +$443,754 on $29.07M net deployed (+1.527% in 22 days).
Observed 22 of 23 days green; 100% of rolling 7-day and 15-day windows profitable.
Inferred External fair-value engine (likely a BTC/ETH reference price feed) drives the ratio decision.
Inferred Fill-and-burst execution; no resting orders worth exiting before resolution.
Live-Data-Required The specific fair-value source and order-book snapshot at entry are not in the CSV. A shadow-trade study against a BTC reference feed would be needed to prove the exact model.

Recommended follow-ups: (a) bankroll playbook shadowing this wallet at 1% scale with ratio ≥ 1.5x filter; (b) replicate the fair-value layer using a BTC reference feed and backtest against this wallet’s entry timestamps; (c) stack ratio-and-paired-cost filters to isolate the highest-edge subset; (d) investigate the 2026-03-29 red day to characterize the regime in which this bot bleeds.

Source data: two CSVs in DATA_DROP/ combining to 1,342,307 raw rows → 1,342,304 after deduping 3 duplicate transaction hashes. Analysis window: 2026-03-25 07:11:51 UTC through 2026-04-16 06:59:54 UTC (21.99 days). All markets fully resolved. Computation: analyze.py. Raw metric blob: REPORTS/_analysis_blob.json.

Bonereaper

Ten-Phase Reverse-Engineering Protocol

Fair-value market-making, with a real signal

High-frequency bot, tiny tickets, crypto-only

Scale

Trade-size distribution

Timing (UTC)

Inter-trade speed (same market + outcome)

Category & duration concentration

Hybrid market-maker with a real, monotonic signal

Both-sides participation

Median dominance ratio

Second-side hedge lag

Dominance-ratio buckets - the directional-edge test

Paired-cost discipline

Win rate tracks implied probability within ~2 points at every band

100% crypto up-or-down. Single-vertical focus.

Dense bursts per market, instant pairing

Per-market accumulation

Second-side hedge lag

Dominance ratio is the cleanest edge; 1.0-1.5x is the only losing bucket

22 of 23 days green. Every rolling 7-day & 15-day window profitable.

Profitable days

Rolling 7-day windows

Rolling 15-day windows

Daily realized P&L (UTC)

Many small wins, occasional large losses - net positive by a wide margin

What this bot is doing, stated precisely

Most honest final conclusion

Other trader breakdowns & guides.

LIL222

KCH123

MEAN REVERSION GUIDE