TradingClaw — AI algorithmic-software company: trading software licence & cloud server runtime

Tradingclaw Limited (BVI) is a company limited by shares incorporated in the British Virgin Islands under the BVI Business Companies Act, 2004. BVI Company No. 2214672. Registered office: c/o CCS Trustees Limited (registered agent), Mandar House, 3rd Floor, P.O. Box 2196, Johnson’s Ghut, Tortola, British Virgin Islands. Support: support@tradingclaw.live.

What we sell

We are an AI-driven software and IT services provider. We license automated trading software for the MetaTrader 5 platform and provide the cloud server (VPS) it runs on. Pricing is a flat annual subscription priced by the visitor's region (CNY 49,999 in Mainland China · HK$59,999 on Traditional Chinese pages · SGD 9,999 in Singapore · US$7,999 in the United States); the final charge currency and amount are confirmed by the team in writing before activation. It covers the software licence and the cloud runtime. Billing is annual and does not auto-renew. A 90-day refund applies to the software subscription fee and is entirely independent of trading results.

What we are not

We are not a broker, not a registered investment advisor, and not a fund or asset manager. We do not give customers investment advice, do not recommend any specific trade to a customer, and do not send customers trading calls or instructions for them to act on. The customer starts the licensed software in a brokerage account they open and control; the software then follows its built-in product rules. That is a software licence, not an agency to accept orders. This company is not a licensed financial institution in any jurisdiction and does not carry on regulated financial activity. We take no profit share and charge no performance fee. The software holds order-execution permission only and has no deposit or withdrawal access. We never touch, hold, or move customer funds, and the customer can switch the software off at any time.

Risk disclosure

All performance figures published on this website are hypothetical historical backtest results. Past performance does not indicate future results, nothing here is a promise of any return, and real trading results are almost always worse than hypothetical results. The published backtest period includes two losing months (about -2.4% in Sep 2025 and again in Apr 2026) and a maximum drawdown of approximately 32.0% (closed-trade balance, after trading costs) (current site baseline: 461 trades, Sep 1, 2025 – Sep 19, 2026, report leverage 1:500; final equity about USD 764,733 / +7547.3%). Trading leveraged products such as gold CFDs carries the risk of losing your entire principal. Customers are solely responsible for their own trading decisions and outcomes.

Jurisdictions

This service is not offered to residents of the United States, nor to residents of any jurisdiction where such software services are prohibited.

Policies

This page: anti-overfitting robustness testing

Engineering quality-assurance documentation. It describes the L1–L4 test programme we use to check the software's strategy logic — out-of-sample split, rolling test months, transaction-cost stress, weekly block bootstrap, random benchmark, time blocks and weekly stability, plus parameter perturbation and stop-noise injection — together with the items that passed and the one that did not pass. Version 9.9, installed in the software on 2026-10-11 and trading from 2026-10-12, was evaluated on 2026-10-11 over Sep 1, 2025 – Oct 9, 2026 (recorded real ticks, USD 30-per-lot costs, compound and fixed 0.1-lot forms, 0.01-lot orders excluded): 13 of 14 checks passed and 1 did not pass. Stop-noise injection (L3.2) did not pass: in the compound form, with every order's stop moved 10% tighter (all in the same direction, the strictest form), net profit fell 24.5%, above the 20% line; at 5% tighter, 5% wider and 10% wider the changes were -17.5%, +5.5% and -11.9%; at 20% tighter net profit was USD 425,108.08 (-82.6%, true max drawdown 28.28%) and at 20% wider USD 2,362,829.21 (-3.2%, true max drawdown 21.10%), so all six levels stayed profitable. Each level was run once; a control run with no shift matched the original 9.9 run exactly. L2.2 parameter sensitivity PASSED: risk percents and the C/D stop/target multipliers each moved 20% down and up (the owner's values); all 28 cells profitable, no cliff; worst deviations -15% (compound ratio, C stop 0.8x) and -21% (single-lot ratio, D stop 0.8x), both still profitable. 9.9 went into the software before the stop-noise and parameter-sensitivity results were in. The 9.9 evaluation used our own recorded ticks only (no cross-broker run). Its rules were chosen on the same year of data, so this is an in-sample health check, not a forecast. The site's headline performance figures are the 9.0 baseline (461 trades), which was not part of this evaluation; 9.0 is no longer the version deployed in the software. Earlier rounds (a 2026-07 sample and version 7.6 in Aug 2026) passed 13 of 14 each; on 7.6, noise injection fell short under the strictest definition and every level was published.

Risk disclosure (applies to every figure on this page)

All performance figures shown are hypothetical historical backtest results, not live account statements and not a promise of any return. Past performance does not indicate future results. Trading leveraged products such as gold CFDs carries the risk of losing your entire principal.

This page requires JavaScript for the full experience. The statements above are identical in substance to the rendered site.