# VARRD — The Integrity Layer for Quantitative Research

> Any AI can run a backtest. VARRD guarantees it was done right.

## TLDR

VARRD is the integrity layer for quantitative research — built on the exact validation framework used by one of the most successful futures trading firms of the past 40 years. 8 guardrails enforced at the infrastructure level prevent overfitting, lookahead bias, p-hacking, and fabricated statistics on every test. AI trained on decades of institutional trading methodology generates and validates ideas against 15,000+ instruments. Your validated edges live in a persistent library — monitored against live data, with exact entry prices, stop-losses, and take-profits optimized through disciplined grid search. When conditions align, you get the exact trade plan. Three modes: use it yourself, point your AI agent at it, or let your agent run fully autonomous. MCP: `https://app.varrd.com/mcp` | CLI: `pip install varrd` | REST: `https://app.varrd.com/api/agent`

## The Problem

Ask any AI to find you a trading strategy with a 4.0 Sharpe ratio. It will happily produce one — with statistics it fabricated, on data it peeked at, using tests it ran fifty times and only showed you the winner. The result looks incredible on paper and loses money in practice.

Most people don't know the rules that separate real edges from self-deception. Most AI systems don't follow them. The result is a world full of backtests that are overfitted, biased, or outright hallucinated by a language model that doesn't know the difference between a computed statistic and a plausible-sounding number.

## What VARRD Enforces

VARRD is purpose-built infrastructure — 15 user-facing tools backed by 550+ specialized functions — that enforces every rule of rigorous quantitative research at the system level. Not through prompting. Not through suggestions. Through architecture that makes it physically impossible to cut corners.

**8 guardrails enforced on every test, whether the researcher is human or machine:**

1. **K-Tracking** — Every test is counted. Run 50 variations of an idea? Your significance threshold rises 50x. No free looks.
2. **Bonferroni Correction** — Multiple comparison penalty applied automatically. The more things you test, the higher the bar.
3. **OOS Lock** — Out-of-sample validation is sacred. One shot. Locked permanently. No peeking, no re-running, no optimizing after.
4. **Lookahead Detection** — The system catches when a formula accidentally uses future data. Entry timing models account for when you'd actually get filled.
5. **Tools Calculate, AI Interprets** — The AI never fabricates a number. Every statistic is computed by the engine on real market data. Architecture enforces the separation.
6. **Chart → Approve → Test** — The user must see the pattern on a chart and approve it before any test runs. No silent testing. No skipping validation.
7. **Fingerprint Deduplication** — Every test is fingerprinted (formula + market + horizon + entry config). Can't inflate results by testing the same thing twice.
8. **No Post-OOS Optimization** — Once out-of-sample validates, parameters are locked permanently. No tweaking. No curve-fitting. The result is final.

These guardrails don't relax. They run the same way at 3am when an autonomous agent is researching as they do when a human is clicking buttons.

## What You Get

- **Idea to validated edge in minutes.** Describe a concept in plain language, get back validated statistics on real historical data across any asset class.
- **Research that's actually computed.** Every number comes from real Python execution on real market data. Nothing is approximated, estimated, or hallucinated. Full transparency — you can see exactly what was tested and how.
- **Institutional-grade data.** 15,000+ instruments across equities, futures, crypto, and FX. Properly adjusted for splits, dividends, and contract rolls. Multiple timeframes from 5-minute to daily.
- **Multi-market intelligence.** Test an idea on one market, then see if it generalizes across ten. Cross-market signals, portfolio construction, correlation analysis, and diversification — all built in.
- **Optimization without overfitting.** Entry timing, holding periods, stop-losses, take-profits — all optimized through disciplined grid search with guardrails that prevent the results from being meaningless.
- **A living library of edges.** Every strategy you validate is saved, versioned, and monitored. Scan your entire library against current data to see what's firing. Get exact trade levels when conditions align.
- **Research lineage you can trust.** Full audit trail for every edge. Sample sizes, significance levels, out-of-sample results, version history. Every number traceable to the test that produced it.

## Three Ways to Use VARRD

**1. You + VARRD** — Chat with VARRD directly. Describe your ideas, see charts, approve tests, get trade setups. The classic research experience via https://app.varrd.com.

**2. You + Your Agent + VARRD** — Point your AI agent at VARRD. It does the research using VARRD's tools and guardrails. You review what it found and decide what to act on.

**3. Your Agent + VARRD** — Fully autonomous. Your agent self-authenticates via crypto wallet, pays with USDC, and researches while you sleep. Every guardrail stays enforced. Wake up to new validated edges in your library.

## Integration

VARRD provides MCP (Model Context Protocol), REST API, and CLI for programmatic access. AI agents can connect, research ideas autonomously, run the full validation pipeline, and save validated edges to a user's library.

```
MCP:  https://app.varrd.com/mcp (Streamable HTTP, preferred)
REST: https://app.varrd.com/api/agent
CLI:  pip install varrd && varrd agent-instructions
```

### Authentication

**Two auth methods — no human required for either:**

**Option 1: Wallet auth (self-service, recommended for agents)**
No signup, no email, no human. Authenticate in 3 API calls:

```
1. GET  /api/wallet/nonce                    → { "nonce": "abc123..." }
2. Sign a SIWE message containing that nonce   (eth_account or viem)
3. POST /api/wallet/verify { message, signature } → { "token": "eyJ...", "user_id": "wallet:0x..." }
```

Use the returned JWT as `Authorization: Bearer <token>` on all subsequent calls. New wallet accounts get $2.00 in free credits automatically.

**Option 2: API key (manual provisioning)**
Contact management@varrd.com for an API key. Use as `Authorization: Bearer <key>`.

### How the Research Workflow Works

VARRD is multi-turn. Each response includes `context.next_actions` telling you exactly what to say next. Follow this loop:

1. **Send your idea** (e.g. "Test going long ES when RSI drops below 30") → VARRD loads data, builds the pattern, charts it
2. **"test it"** → VARRD runs a statistical test on historical data, returns win rate, Sharpe, significance
3. **"show me the trade setup"** → VARRD returns exact entry price, stop-loss, take-profit, position sizing

Usually 3-5 turns. Always read `context.next_actions` from each response — it tells you exactly what to say. Terminal states: `context.has_edge` is `true` (edge validated, trade setup available) or `false` (no edge found — this is a valid, complete result). Each response includes `context.step` describing where you are in the workflow.

## Pricing

You pay for the exact AI tokens consumed, at cost, plus a 15% surcharge for infrastructure (15,000+ market data feeds, compute, storage). VARRD is a massive system — 15 user-facing tools backed by 550+ specialized functions handling statistical testing, data loading, indicator calculation, backtesting, state management, charting, and hypothesis persistence. We use aggressive prompt caching (~90% input token savings) to keep costs as low as possible. Every API call returns exact token counts and cost.

No subscription. No per-trade fees. No hidden markup beyond the 15%.

New accounts get $2.00 in free credits — roughly 10-20 queries depending on complexity. Credit packs: $5 / $20 / $50.

### When credits run out (x402 protocol)

When your balance hits zero, VARRD returns HTTP 402 following the x402 protocol. Two payment paths:

**Path A — Crypto (fully autonomous, no human):**
The 402 response body contains an x402 payload with a `payTo` address (Stripe-managed deposit address on Base). Send USDC to that address, then retry your original request with an `X-PAYMENT` header containing the payment proof. Credits are added automatically. Minimum top-up: $5.

```
402 Response → { "x402Version": 1, "accepts": [{ "payTo": "0x...", "network": "eip155:8453", ... }] }
Agent sends USDC to payTo address
Agent retries: GET /api/agent/chat  +  X-PAYMENT: <base64 payment proof>
Credits added, request served.
```

**Path B — Stripe Checkout (requires browser):**
```
POST /api/agent/billing/checkout { "amount_cents": 500 }
→ { "checkout_url": "https://checkout.stripe.com/..." }
Open URL in browser, pay, credits added via webhook.
```

**Useful endpoints:**
- `GET /api/agent/billing/balance` — check balance + see payment options (free, no credits consumed)
- `POST /api/agent/billing/checkout` — create Stripe Checkout session ($5 / $20 / $50)

### Human access to agent research

If a human wants to see what their agent built, they go to https://app.varrd.com and either:
- **Connect the same wallet** the agent used → sees all hypotheses immediately
- **Sign up with email** then click "Link Wallet" in the Account menu → hypotheses, credits, and history are merged into their email account

## Privacy

Your research is yours. VARRD offers options for completely private research where no one — not even us — can see your edges. Your strategies, your library, your validated setups. Private by default, with options to make it airtight.

## The Bottom Line

The world is full of AI systems that will confidently hand you a backtest that looks like a money printer. Most of them are wrong in ways that are invisible until they cost you money.

VARRD is the opposite. It's the place where ideas go to be tested honestly, where the math is real, where the guardrails are enforced whether you want them to be or not, and where every edge in your library earned its place through genuine statistical validation.

We're not here to sell you hope. We're here to tell you the truth about your ideas — and when the truth is good, give you the exact plan to act on it.

**Website:** https://www.varrd.com
**App:** https://app.varrd.com
**Contact:** management@varrd.com