# The Composable Pre-Trade Verification Stack for Autonomous Trading Agents

**Version**: 1.0 | **Status**: Draft Specification | **License**: Apache 2.0

## Abstract

Autonomous trading agents need layered verification before executing trades.
Capability without verification is liability. This specification defines a
composable pre-trade verification stack — five independent layers that an
autonomous agent MUST pass through before executing any financial transaction.

## The Problem: Agents With Money and No Safety Rails

An autonomous trading agent with exchange access, a funded wallet, and no
pre-trade verification will execute trades during market closures, circuit
breaker halts, and based on unverified signals. Each failure mode is
independently dangerous. Together, they represent systemic risk.

## The Stack

```
┌─────────────────────────────────────────────────┐
│  Layer 5: Trade Execution                       │
├─────────────────────────────────────────────────┤
│  Layer 4: Payment (x402 or equivalent)          │
├─────────────────────────────────────────────────┤
│  Layer 3: Signal Verification (VeroQ or equiv.) │
├─────────────────────────────────────────────────┤
│  Layer 2: Spend Authorization (Ampersend or eq.)│
├─────────────────────────────────────────────────┤
│  Layer 1: Market State Gate (Headless Oracle)   │
│  Ed25519-signed, fail-closed. UNKNOWN = CLOSED. │
└─────────────────────────────────────────────────┘
```

If any layer fails, all subsequent layers are skipped and the trade is halted.

### Layer 1 — Market State Gate (Headless Oracle)

**Question**: Is the target exchange currently open for trading?
**Protocol**: Ed25519-signed receipt, 60-second TTL.
**States**: OPEN, CLOSED, HALTED, UNKNOWN. Only OPEN proceeds.
**Coverage**: 28 global exchanges (equities, derivatives, 24/7 crypto).
**Endpoints**: MCP (`npx headless-oracle-mcp`), REST (`/v5/status`), x402 ($0.001 USDC).
**Why Layer 1**: Every subsequent layer depends on the market being open.

### Layer 2 — Spend Authorization (Ampersend or equivalent)

**Question**: Is the agent authorized to spend this amount?
**Protocol**: Policy-bound authorization. Human-in-the-loop for high-value actions.
**Composable**: Layer 1 receipt signature serves as evidence in authorization request.

### Layer 3 — Signal Verification (VeroQ or equivalent)

**Question**: Is the trading signal factually accurate?
**Protocol**: Claim verification against live market data.

### Layer 4 — Payment (x402 or equivalent)

**Question**: Can the payment be executed with cryptographic proof?
**Protocol**: On-chain USDC on Base with transaction-level proof.

### Layer 5 — Trade Execution

Execute the order with all layer proofs attached for audit trail.

## Why Layer 1 Must Be Fail-Closed

1. Markets have objective state — open or closed is a fact, not a judgment.
2. All other layers depend on market state.
3. Fail-closed prevents the worst outcomes (capital loss vs. opportunity cost).
4. 60-second TTL forces re-verification before every execution window.

## Reference Implementations

| Layer | Implementation | Protocol |
|-------|---------------|----------|
| 1 | [Headless Oracle](https://headlessoracle.com) | Ed25519 receipts, MCP, REST, x402 |
| 2 | [Ampersend](https://github.com/edgeandnode/ampersend) | A2A, policy-bound auth |
| 3 | [VeroQ](https://veroq.ai) | AI claim verification |
| 4 | [x402](https://www.x402.org/) | HTTP 402, on-chain USDC |

## Machine-Readable Discovery

JSON: `GET https://headlessoracle.com/v5/pre-trade-stack`