A multi-asset trading platform running an autonomous strategy testing arena across US equities, Indian markets, crypto and forex. This guide walks through every surface of the product and states plainly what is live, what is simulated, and what is still being built.
Every trade in this platform is paper or simulated. No real money is at risk,
and live trading is blocked at the infrastructure level (autotrade.allow_live = false).
Nothing in this guide should be read as live investment performance.
That is a deliberate design choice, not a limitation we are hiding. The arena's purpose is to accumulate honest, statistically meaningful evidence about which strategies work — under real market data, real spreads and real risk limits — before a single dollar is exposed. Prices are never fabricated: when a market or vendor does not publish a value, the interface shows a dash rather than a placeholder number.
Every figure below is read directly from the production database at the time this guide was generated. None are projections.
The arena has been running continuously since 25 July 2026. The gap between 7,999 decisions and 267 fills a day is the point: the system evaluates constantly and acts rarely, and every rejection is recorded with its reason.
Most trading products show you one blended equity curve. That makes it impossible to tell whether a good month came from a genuinely strong strategy or from one lucky position masking four losers.
Hybrid Forex gives every strategy its own isolated book. There are currently 29 separate accounts — one per strategy per venue — each with its own cash, its own positions and its own P&L. A fill in vwap_wave on US equities can never be confused with one from ichimoku_cross on crypto.
This means we can answer the question that actually matters to an allocator: which strategy works, on which market, and with what consistency? — with months of tick-by-tick evidence rather than a backtest.
Sub-accounts are modelled as broker connections under a single owner, not as separate logins. One user can operate dozens of independent books while every fill remains traceable to exactly one strategy, one venue, and one rule.
The landing surface after sign-in: account state, recent activity and what the system is currently watching.
Multi-asset charting with drawing tools, indicator overlays (VWAP, anchored VWAP, Ichimoku, Donchian, Bollinger, MACD, RSI) and an order ticket supporting market, limit, stop and bracket orders. Charts render real market data; orders route to simulated venues.
Every open position across all venues, with live mark-to-market P&L. Two views are kept deliberately: a merged per-symbol table, and a By strategy table showing each isolated book ranked by unrealized P&L.
Indian books are denominated in INR and every other venue in USD; the interface never sums across currencies. It also leads with gross exposure and a long/short split rather than a single net figure, because net market value goes sharply negative whenever shorts dominate and misleads at a glance.
Where strategies are configured: symbol universe, timeframe, minimum signal strength, position sizing, stop loss, take profit, trailing stop, maximum concurrent positions and daily order caps — per rule.
Market discovery across four asset classes: top gainers, losers, most active, volume shockers, near 52-week highs and 52-week lows, over a curated universe of 203 liquid instruments.
On crypto and forex the two volume-ranked boards are intentionally replaced with a note, because our data vendor does not publish volume for those instruments. We would rather show an honest absence than an invented number.
Trades, written notes, and the auto-trader's full reasoning log — every decision it made, including the ones where it declined to trade and why.
A panel of AI analyst personas that reach a verdict on a symbol. It operates on veto semantics: the panel does not generate trades, it can only block one a strategy already wants to take.
Market intelligence: 25,600 news items and 30,738 economic calendar events, plus sentiment.
System health, daily reconciliation, email delivery, strategy scoreboard and the self-improvement engine's findings.
User management with search, filters, sorting and bulk actions; audited login-as-user support that is strictly read-only — an administrator viewing a member's account cannot place trades or change settings on their behalf.
Being precise about this matters, because "AI trading" is often oversold.
| Component | What it genuinely does |
|---|---|
| Strategy engine | Computes deterministic technical signals. No model inference — it is indicator mathematics. |
| Scanner | Sweeps the symbol universe hourly and emits signals with a strength score. |
| Round Table | LLM analyst personas that can veto a pending trade. They never originate one. |
| Improvement engine | Reviews hourly/daily/weekly outcomes and proposes changes. Material changes require human approval. |
| Risk engine | Hard-coded limits. Deterministic, and not overridable by any model. |
The AI's role is judgment and veto, layered on top of deterministic signal generation and non-negotiable risk limits. It is not a black box that invents trades.
| Control | Behaviour |
|---|---|
| Live trading | Blocked at infrastructure level. Requires explicit sign-off to change. |
| Concentration cap | No single symbol may exceed a set share of account equity, measured across all rules — many individually-compliant rules cannot quietly pile into one name. |
| Time stop | Positions are force-exited after a maximum holding period. |
| Directional filter | Rules may be constrained long-only or short-only. An unrecognised value refuses to trade rather than trading unconstrained. |
| Short capacity | Shorts are rejected when the venue cannot support them, preventing un-closable positions. |
| Funded-account rules | Prop-firm style guardrails: daily loss limits, drawdown ceilings, flatten windows. |
| Per-rule risk | Stop loss, take profit and trailing stop configured per rule. |
Risk controls fail closed. If a limit cannot be evaluated, or is configured with a value the system does not recognise, the trade is refused rather than allowed. A guardrail that silently does nothing is worse than no guardrail, because it manufactures false confidence.
Market data comes from commercial vendors with a documented fallback chain. Crypto candles are ingested directly from exchanges. Indian instruments resolve through a dedicated pricing path.
The rule we do not break: if no vendor returns a real price, the system records no price. Positions are then valued at cost and the interface shows a dash. There is no synthetic pricing anywhere in the platform.
| Layer | Technology |
|---|---|
| Web | Next.js 15, React 19, server components |
| API | tRPC gateway, end-to-end typed |
| Services | 17 Python services: strategy engine, scanner, broker agent, risk engine, round table, improvement engine, reporting |
| Data | PostgreSQL with TimescaleDB |
| Security | TOTP two-factor, stateful revocable sessions, trusted devices, encrypted vendor credentials, full audit log |
Included because a guide that claims everything is finished is not credible.
| Area | State |
|---|---|
| Testing arena, US & crypto | mature Running continuously since July 2026 |
| Indian (NSE) arena | new Live since August 2026; long-only by design, as Indian cash shorts are intraday-only |
| Alpaca paper integration | active Connected, trading paper |
| Live trading | blocked Intentionally disabled pending sign-off |
| Options | paper Single and multi-leg on simulated venue |
| Market discovery | live 203 instruments; volume unavailable for FX/crypto |
| Community | rebuilding Previous chat platform retired |
| Offsite backups | pending Tooling built, awaiting provider credentials |