BTC/USDT 67 420 ▲ 1.84% ETH/USDT 3 512 ▲ 2.31% EUR/USD 1.0874 ▼ 0.12% GBP/USD 1.2731 ▲ 0.21% XAU/USD 2 358 ▲ 0.46% WTI Oil 78.92 ▼ 0.83% Brent 83.14 ▼ 0.55% SP 500 5 487 ▲ 0.38% Nasdaq 19 740 ▲ 0.62% DAX 18 320 ▼ 0.17% SOL/USDT 162.4 ▲ 3.9% USD/JPY 157.2 ▲ 0.09%
Bots online 24/7 Server-side strategy lab WAF · CAPTCHA · Bcrypt

AI Trading Platform

Proprietary trading terminal with AI and a strategy research engine: autonomous bots, 69 strategy families and an edge constructor with walk-forward validation and a forward gate, risk-based sizing with compounding — running even when the browser is closed. Crypto, forex, indices and commodities across 5 brokers: Binance, OANDA, IG, Saxo, Trading 212.

Learn more
AI trading robot
17
autonomous bots and executors
69
strategy families
516
edge constructor recipes
43
crypto edges proven
60
neural network inputs
61+
instruments in the lab
30
technical indicators
28
candlestick patterns
7
timeframes M5–W
24/7
server-side strategy lab
±
signal ensemble: entry + exit
5
brokers (OANDA·IG·Saxo·Binance·T212)
WF
walk-forward + forward gate
11
research laboratories
1:1
simulator mirrors the live bot
z
null gate + multiplicity correction

AI Trading Platform — a proprietary web terminal for trading crypto, forex, indices and commodities. Built on PHP 8.3 / JavaScript ES6+ / Python 3.8 with integration of Binance, OANDA v20, IG, Saxo and Trading 212 API. AI-driven automation, a server-side strategy lab with walk-forward validation and a null gate (runs with no browser open), a signal ensemble and a professional technical analysis toolkit. Charts — Lightweight Charts (TradingView), interface in EN / RU.

The platform was built as a single workplace for the systematic trader: instead of a dozen tabs and scattered scripts — one terminal where bots, indicators, technical analysis, the trade journal and risk management work together. Every bot keeps its own log, computes expectancy and win rate, and models are retrained on closed trades. You get transparent statistics on P&L for every bot and broker — with no black boxes.

Цикл 11 августа: данные, метки, инварианты

шесть находок, каждая меняла выводы, а не только код

Мы считали на 2% рынка

Данные по акциям брались с фида iex — это одна биржа из полутора десятков. Замер по PLTR на пятнадцатиминутках за месяц: iex дал 591 бар и 28.7 млн объёма, полный тейп sip — 818 баров и 426.6 млн. Треть баров отсутствовала вовсе, остальные сложены из одной пятнадцатой сделок: их максимум, минимум и закрытие не отражали рынок. Вывод «по бумаге эджа нет» оказался выводом о фиде: на полных данных тот же прогон нашёл семь стратегий вместо одной.

У данных теперь есть паспорт

Ошибка прожила месяцы, потому что данные никто не спрашивал, кто они. Перед каждым перебором ряд проверяется на источник, глубину, полноту, разрывы и битые бары — с поправкой на режим торгов, потому что круглосуточную крипту нельзя мерить меркой шестичасовой американской сессии. Паспорт кладётся рядом со стратегией: через месяц видно, на чём именно она доказана.

Победителей резали на пятой минуте

Профит-лок сравнивал пиковую прибыль в фунтах с плановым риском в фунтах. План расходится с фактом всегда — минимум лота, маржинальный кап, курс котируемой валюты. Порог уезжал, и все победители выходили ровно на +1.00R при цели 2.9R: победитель становился равен проигравшему. Теперь R берётся из цен, где валюта и объём не участвуют.

Выход судил по бару входа

Сигнал выхода считался по последнему закрытому бару — тому самому, на котором открылись. Форма бара входа выбивала сделку немедленно: медиана жизни позиции была пять минут при часовых и дневных стратегиях, три четверти закрывались за четверть часа. Выход теперь ждёт хотя бы один новый бар.

Проверки, которых не хватало

чтобы найденное не вернулось

Порог шума мог только расти

Калибровка случайности копилась через максимум со старым значением: один испорченный замер отравлял инструмент навсегда. По PLTR держалось 11.01, тогда как честный перезамер дал 1.77 — завышено вшестеро, и стратегии резались против числа, которого шум никогда не показывал. Таких ключей было 96 из 652. Перезамер теперь замещает, а файл ведёт список замен.

224 метки на сделку

Журнал вырос со 129 полей до 224 и подключён ко всем двадцати торгующим ботам вместо одного. Тренд, диапазон, волатильность, время, состояние счёта на входе; форма пути, смена режима под сделкой и КПД выхода — на закрытии. Всё считается из свечей, которые бот уже держит: ни одного лишнего запроса к брокеру.

Инварианты во время работы

Сканер читает код — этот слой смотрит на поведение. Десять проверок стоят внутри журнала, через который проходит каждая сделка, поэтому покрыты и боты, которых ещё не написали. Главная из них: полный стоп обязан давать ровно −1.0R. Именно она вскрыла, что R считали делением фунтов на котируемую валюту.

Сканер: 8 классов → 29

Каждый новый класс добавлен по случившемуся на счёте, а не из общих соображений: время свечи через parseFloat (у всех баров выходило число 2026), выдуманная сетка объёма, риск на котируемую валюту, закрытие чужих позиций в цикле по всему счёту, частичный фид, данные без паспорта. Из 130 находок настоящими оказались меньше половины — остальные разобраны глазами и признаны ложными.

What changed in the last cycle

end-to-end audit of the engine and every bot: 1–2 August 2026

The simulator no longer sells at the peak

The profit lock was derived from the bar's high and could end up above the market — a broker rejects such a stop, yet the test booked an exit at the best price of the day. The tell was obvious: a lock fraction of 1.0 came out best at every arming level. The lock is now placed only if it is actually executable.

The trailing stop no longer peeks inside the bar

The stop was pulled up to the bar's high and immediately checked against that same bar's low — the test exited at a price that did not exist at that moment. On a single bar this conjured up to +2R out of thin air, and it inflated precisely the trend strategies the system leans on.

Risk measured in real account currency

Position size was computed by dividing pounds by a stop distance denominated in the quote currency. On yen pairs the real risk came out 210 times smaller than planned, while gold and indices were never traded at all — their size failed an invented minimum. Minimum size and size step now come from the exchange itself.

The lock threshold is now a fraction of risk

The threshold was set in absolute pounds: at £1 of risk that means 0.8R, at £2.10 it becomes 0.38R — the setting drifted silently. Measurement showed the value in place was the worst cell in the entire grid. It is now chosen by measurement rather than by eye.

New laboratories

each one settles a decision previously made by eye

Fidelity lab

Takes real closed trades from the broker and replays them through the model with identical parameters. It catches divergences nobody suspected. First run: for one bot the model promised +1.31R while the account delivered +0.05R — it turned out 86% of its trades were killed from outside, reaching neither stop nor target.

Portfolio lab

Every other lab measures a single strategy in isolation. This one trades the whole set simultaneously on a shared account and tests what used to be guesswork: how many positions to hold at once, at what correlation to stop adding another, and whether the daily loss limit helps at all.

Weak-point scanner

The bugs found were not isolated slips but eight classes multiplied by copy-paste — the same line sat in nine bots. The scanner describes each class as a check and runs it across every file, preventing it from being reintroduced.

Exit lab

Selects the profit-lock threshold and holding limit by measurement, tuned to a bot's actual setups. There is no universal rule: for one bot locking hurts at every level, for another it helps, for a third it makes no difference.

Observability

errors that look entirely plausible

Measurement at every step

Every bug found looked entirely plausible: the code ran, numbers came out, the logs were silent. Now each calculation step is checked against what is physically possible, and a violation is recorded with examples — even when nothing crashed.

Order checks before sending

One battery of checks: size is positive, stop and target are on the correct side, size is not below the exchange minimum, and risk matches the plan at least in order of magnitude. The last one would have caught the 210-fold miss instantly.

Dead strategy detection

A strategy that produced not a single signal across its whole history is flagged as dead. That is how a bot whose indicator was identically zero passed for working for half a year.

A network failure no longer means everything closed

An empty broker response on timeout was indistinguishable from “no positions”: trades were counted as closed, slots were freed, and positions stayed with the broker unmanaged. A failure is now distinguished from an empty list.

Calculation accuracy

what changed in the calculations themselves

Costs from history, not from a snapshot

The spread was taken from a single quote and applied to the entire history — on weekends that meant Friday's close, overstating costs fourfold. Now the median of the actual spread across the whole sample is used.

The rollover hour is closed

Measured by hour of day: at 21:00 UTC the spread widens 2.6–5.8 times. That hour is now excluded from entries everywhere — it hurts fast timeframes the most.

Risk scale decoupled from the timeframe

The stop was hard-wired to the signal's timeframe, which made fast intervals look untradeable. Entry timing and trade weight are different things: on the same data 47,682 variants yielded zero survivors under the old rule and ten under the new one.

Selection six times faster

Each strategy's signal was recomputed from scratch forty-five times in a row. Removing that redundancy cut a run from 185 seconds to 33 with fully identical results.

Platform sections

Each section on its own page

Features

What the platform does.

Bots

Autonomous trading bots.

Strategies

Strategy families.

Engine

Laboratory and walk-forward.

Flow chart

How it all works.

Markets

Markets and indicators.

Security

Account and data protection.

History

Update timeline.

FAQ

Frequently asked questions.

Access

Plans and access.

Ready to launch autonomous trading?

Connect your brokers, configure the bots and hand the routine to the algorithms. Transparent statistics, round-the-clock server-side learning and capital protection — from your first login.

Secure access · DEMO and LIVE · 5 brokers · 80+ instruments + crypto