Triangular Arbitrage Scanner (Trader Edition)

This project is not a "magic AI bot". It is a deterministic cycle scanner with a conscious decision layer.

Goal: only surface setups that still look positive after realistic trading costs.

What Changed

The codebase now includes a TradingLogicDirector in triangular_arbitrage/director.py.

It separates two responsibilities:

1. Detection (triangular_arbitrage/detector.py)
   Finds the best currency cycle by gross multiplier.
2. Decision (triangular_arbitrage/director.py)
   Applies taker fees, slippage assumptions, and a minimum net edge threshold before signaling TRADE.

This prevents "looks good on paper" cycles from being treated as executable opportunities.

Experimental Trading Logic

The director evaluates:

• gross_profit_multiplier from the detector
• cost model per leg: taker_fee_bps + slippage_bps
• compounded net multiplier across all legs
• execution gate: min_net_profit_bps

Decision rule:

TRADE only if expected net edge (in bps) is above threshold.
Otherwise: SKIP with an explicit reason.

Why Traders Care

Most public arbitrage demos fail because they:

• ignore fees
• ignore slippage
• report stale opportunities
• trigger on tiny edges that disappear on fill

This implementation explicitly addresses the first three at the decision layer and is parameterized for stricter filters.

Quick Start

Requirements

• Python 3.10+

Install

pip install -r requirements.txt

Run

python main.py

Example Output

Scanning...
-------------------------------------------
Best gross cycle on bitget: 0.32710%
Expected net edge after costs: 0.17180%
Decision: TRADE (Net edge clears threshold.)
1. buy BTC with USDT at 0.00003
2. sell BTC for ETH at 14.88123
3. sell ETH for USDT at 2060.20000
-------------------------------------------

If the net edge is below threshold:

Decision: SKIP (Net edge below threshold after costs (9.20 bps < 15.00 bps).)

Strategy Parameters

Edit main.py and tune DirectorConfig:

• exchange_name: ccxt exchange id (example: bitget, binanceus)
• max_cycle: set 3 for triangular cycles
• taker_fee_bps: fee assumption per leg
• slippage_bps: slippage assumption per leg
• min_net_profit_bps: minimum net edge required to trade

Experimental Analysis Template

Use this table while paper trading to validate profitability discipline:

Date	Exchange	Gross Edge (bps)	Net Edge Model (bps)	Realized (bps)	Decision	Notes
2026-02-20	bitget	41.0	19.6	14.2	TRADE	Partial fill on leg 2
2026-02-21	bitget	24.0	8.5	-	SKIP	Correct skip, edge too thin
2026-02-22	bitget	55.0	33.1	29.4	TRADE	Clean execution

Interpretation:

• positive realized bps on trades and avoided thin edges indicate the director logic is working
• if realized bps consistently trails model, increase slippage and threshold

Profitability Statement

This bot is designed to be profitable only when your market + fee regime supports positive net edge after costs.

It is not guaranteed profit. Treat it as an execution filter that enforces trading discipline, then validate with paper logs and live micro-size before scaling.