Alpha Quant Pro AQP Bundle Development Guide

This page follows the current production contract: AQP Bundle Manifest v2.

---

1. Canonical Contract

| Concept | Current standard | |---|---| | Research object | Strategy | | Delivery/install object | AQP Bundle | | Production artifact | signed *.aqp | | Bundle ID | aqp_bundle_* | | Bundle kind | bundle_kind: strategy | | Strategy kind | rules | ml | hybrid | | Manifest version | format_version: 2.0 | | Python entrypoint | strategy.py | | Entry class | TradingStrategy | | Optional factory | load_strategy() |

Manifest v1 names exist only behind the explicit read-only compatibility boundary for historical signed packages. New code, examples, tests, and releases must use Manifest v2 terminology.

---

2. Source and Release Layout

aqp_bundle_example/
|-- manifest.json
|-- strategy.py
|-- config.json
|-- risk_profile.json
|-- requirements.lock
|-- models/
`-- README.md

A production .aqp additionally carries manifest.sha256 and aqp_signature.json. Production Desktop accepts signed .aqp packages; folders, ZIP files, and loose Python sources are development-mode inputs only.

---

3. Manifest v2 Example

{
  "bundle_id": "aqp_bundle_gold_example",
  "bundle_kind": "strategy",
  "strategy_kind": "hybrid",
  "format_version": "2.0",
  "runtime_version": "aqp-runtime-v1",
  "name": "Gold Example",
  "version": "1.0.0",
  "entrypoint": "strategy:TradingStrategy",
  "supported_symbols": ["XAUUSD"],
  "supported_timeframes": ["M15"],
  "model_files": ["models/model.onnx"],
  "permissions": {
    "trade_request": true,
    "mt5_direct": false,
    "network": false,
    "filesystem": "bundle_readonly"
  }
}

The Manifest must not contain account-level execution settings such as lot size, risk percentage, maximum daily loss, maximum spread, magic number, deviation, or an auto-trading switch.

---

4. Strategy Interface

class TradingStrategy:
    def on_bar(self, ctx):
        confidence = 0.72
        if confidence < ctx.config.get("min_confidence", 0.65):
            return ctx.intent.hold(reason="confidence_filter")
        return ctx.intent.market(
            side="BUY",
            confidence=confidence,
            sl=ctx.price - 5.0,
            tp=ctx.price + 15.0,
            reason="model_signal",
        )

The Strategy returns a TradeIntent. It does not call MetaTrader 5 directly. load_strategy(bundle_path) may be supplied as a factory when needed.

---

5. Ownership Boundary

AQP Bundle owns: strategy logic, feature/model inference, signal confidence, intent, and strategy-internal configuration.

Alpha Quant Pro Runtime owns: account/subscription/device/broker authorization, final position sizing, account risk, spread/news filters, circuit breakers, and MT5 order execution.

A Bundle cannot bypass Runtime authorization or account-level risk controls.

---

6. Dependencies and Security

• Production dependencies belong in requirements.lock, with pinned versions and SHA-256 hashes.
• Production Workers do not install dependencies automatically.
• Network access must be explicitly declared and routed through ctx.http; raw socket, requests, and urllib access is not an execution shortcut.
• Direct MetaTrader5.order_send(), dynamic MetaTrader5 imports, subprocess/system commands, and file access outside the Bundle are prohibited.
• Protected .aqp packages must pass Manifest, hash, signature, dependency-policy, and entrypoint validation before installation.