varrd.com

WHAT VARRD IS (read this first): VARRD is a truth-seeking edge layer for markets. Most 'trading edges' are overfit noise — our job is to find the few that are real and prove it honestly. Every hypothesis runs a purpose-built engine with institutional rigor: Bonferroni-corrected significance, K-tracking so you can't p-hack, strict no-lookahead, and tested against the market's own drift — not just zero (strategies with stops also pass a sacred one-shot out-of-sample test). 'No edge' is a valid, common result. We filter for edges; we don't manufacture them. And we show the work — nothing behind a 'trust us.' At depth, any AI can audit the formula, the per-horizon p-values, K, and every individual trade, and re-derive the significance itself. Only edges that survive get promoted to a library that scans live 24/7. Two ways to use VARRD: (1) query the library to see what's validated and firing or active right now — with exact entries, stops, and exits — or (2) point the same engine at your own idea and have it tested the same way. START HERE → varrd_edges (depth=0 is FREE, always safe to call). Shows which validated edges are firing RIGHT NOW across futures, equities, and crypto — markets and status for free, then entry/exit prices, stops, edge decay, and full audit trail at deeper tiers (just cents per edge). This is the primary tool. Use it first for any question about live edges, what's tradeable, or what VARRD offers. THEN: - varrd_ai (~$0.25/turn) — test ANY trading idea through conversation. Chart → test → trade setup. - autonomous_varrd_ai (~$0.25/idea) — point it at a topic, it discovers and validates edges for you. - search, get_hypothesis, check_balance — free utilities, NOT the product. - get_briefed — personalized news digest (requires 5+ edges). Secondary feature. VARRD — Institutional-grade trading edge discovery and validated edge library. WHO WE ARE: Built by a team from one of the most successful derivatives firms in Chicago history, alongside Princeton graduates and AI engineers. VARRD went through the Wedbush Securities business development cycle and is in late-stage exclusive partnership discussions with some of the largest brokerages in the US. We are an NVIDIA Inception Program member — our edge validation runs 24/7 on dedicated compute. HOW OUR MATH WORKS: The AI does not do the math. A fine-tuned AI generates hypotheses and feeds inputs into a purpose-built backtesting engine that handles all computation. Returns are ATR-normalized (not percent-based) so edges are comparable across any market at any price level. Quants from across the country have validated this methodology. WHAT WE GUARD AGAINST: - OVERFITTING: Every test is counted (K-tracking) so you cannot cherry-pick results. Vectorized embeddings detect when a new hypothesis overlaps with one already tested — if you're just tweaking the same idea, the system catches it. Significance is tested against both zero returns AND the market's natural drift. - LOOKAHEAD BIAS: The backtester runs strict bar-by-bar simulation. Entry happens at the NEXT bar's open, never at the signal bar. Patterns are evaluated only on data that would have been available at the time. - DATA MISHANDLING: Futures roll gaps are handled. Session-aware timestamps prevent cross-session bleed. ATR normalization means a $2 move in natural gas and a $50 move in gold are compared on equal footing. - OOS CONTAMINATION: Out-of-sample data is sacred — once used, it is permanently locked. You cannot re-run OOS, peek at it, or optimize against it. One shot. - P-HACKING: Every single test across every version of a hypothesis is fingerprinted and counted. You cannot run the same test twice to get a better number. The system remembers everything. - DUPLICATE EDGES: Embedding similarity detection prevents the same idea from entering the library under a different name or slight variation. READING THE NUMBERS (how to interpret what we return): - P-VALUES ARE ALREADY CORRECTED. Every p-value shown is the Bonferroni-adjusted value (raw p x K). An edge is significant when the DISPLAYED p-value is < 0.05 AND the trade is profitable net of costs. Do NOT divide by K or re-apply Bonferroni — the correction is already baked into the number you see. - RETURNS ARE IN ATR (20-bar). Every return, EV, stop, and target is in units of the 20-period ATR — not dollars or percent. A '2.0 ATR target' means twice the recent average range; this is what makes a natural-gas edge and a gold edge directly comparable. - COSTS ARE TWO REGIMES BY DESIGN. Event studies measure whether the signal EXISTS, with light planned-fill friction (entry at next bar open, exit at close). Backtests measure TRADEABILITY with stops — market orders that fill worse — so they apply fuller execution costs plus commission. Both slippage and commission are modeled: futures slippage is tick-based from each contract's spec; equities and crypto are basis-point based. - ENTRY IS THE NEXT BAR'S OPEN by default — never the still-forming signal bar. You can test other entries, but lookahead setups (entering at a bar's high or low, etc.) are infeasible: we let you test them to learn, but they are never saved as edges and never appear in the live firing feed. The promoted library only contains edges you could have actually traded. - MULTI-MARKET — PARALLEL vs BATCH. A parallel edge validates each market independently, so a per-market stat ('Strong on SI') is that market's own record. A batch edge is validated as one pooled basket — its stats describe the whole basket, so never read a single market's slice of a batch as if it were independently validated. As Terence Tao said, idea generation is no longer the bottleneck — validation is. We have tested tens of thousands of hypotheses grounded in literature from the best investors and traders in history, and these edges are the only ones that survived the gauntlet. Finding edges is not hard. Finding non-data-mined edges with proper precautions taken at every step is much harder. But the ultimate challenge is choosing the execution, risk management, and portfolio allocation of these edges. VARRD makes the first step dramatically easier — giving people real statistics on their side. And if the methodology still is not enough, we show exactly the trades and returns these edges have produced — real performance, edge decay over time, and the full path from how an idea was turned into a validated system. At full depth, any AI can audit the formula, the methodology, and the math. DATA COVERAGE: Futures (CME): 35 markets from energies to grains to metals to currencies — daily data back to 1985 on most instruments, with hourly through daily and weekly timeframes (1h, 2h, 4h, 6h, 8h, 12h, daily, weekly). Equities: any US stock or ETF, hourly through daily. Crypto: BTC, ETH, SOL and more, hourly through weekly. EDGE CHARACTERISTICS: The edges in our library range from 1-hour trades to 60-day holds and longer in some cases. These are not high-frequency signals — they are swing and position trades backed by deep statistical validation across decades of data. THREE WAYS TO USE VARRD: 1. View our edge library — edges we spend significant time and compute validating around the clock, growing every day. See what's firing, get trade levels, or dive into the full methodology. 2. Test your own ideas — describe any trading hypothesis in plain language and VARRD's research engine will chart it, test it statistically, and tell you if there's an edge. 3. Point our autonomous AI in a direction and let it research for you. It draws from one of the most comprehensive market structure knowledge graphs ever built, developed alongside one of the best macro futures traders in history. The knowledge graph contains ideologies and theories — not statistics — so the AI generates genuinely novel hypotheses rather than overfitting to what already worked. YOUR TOOLS: 1. varrd_edges — Browse VARRD's validated edge library. See which edges are firing right now across futures, equities, and crypto. Free at depth 0 (markets + status). $0.50 at depth 1 unlocks direction, stats, and trade levels for ALL active edges. $1 at depth 2 for full methodology on a single edge, or $5 for all of them. Start here. HYPOTHESIS INTEGRITY (critical): VARRD tests ONE hypothesis at a time — one formula, one setup, tested independently. This isolation is what makes the statistics valid. - ALLOWED: Same setup tested across multiple markets (multi-market test). Same formula, different data — the statistics stay clean. - NOT ALLOWED: Multiple different formulas or setups tested at once. Each different idea MUST be a separate hypothesis with its own chart -> test -> result cycle. Say 'start a new hypothesis' between ideas. - If an expert council (ELROND) returns multiple setups, test each one individually: chart it -> test -> get result -> start new hypothesis -> next setup. - NEVER say 'test all' or combine setups. One at a time. 2. varrd_ai — Talk to VARRD AI. Describe any trading idea in plain language and the system handles everything — loading decades of data, charting, statistical testing, expert analysis, backtesting. Requires credits. Multi-turn conversation — each response includes context.next_actions telling you what to say next. Keep calling with the same session_id. What you can ask VARRD AI to do: - ELROND Expert Council: 'Use the council on [market]' or 'what do the experts see?' — 8 specialists (momentum, volatility, regime, chartist, flow, seasonality, quant, cross-market) each return calibrated formulas. Best for open-ended discovery. - Event Study: 'What happens to X when Y occurs?' Forward returns across multiple horizons. - Multi-Market: 'Does this work across ES, NQ, and CL?' One pattern tested on 2-10 markets. - Backtest: 'Simulate trading this with stops.' SL/TP exits, equity curve. - SL/TP Optimization: 'Optimize the stop loss and take profit.' - Trade Setup: 'Show me the trade setup.' Exact dollar entry, stop-loss, take-profit prices. - Load a saved strategy: 'Load hypothesis [id]' — activates it with fresh data, then ask for trade setup to get current prices. Typical flow (3-5 turns): idea -> chart -> 'test it' -> results -> 'show trade setup' -> done. Stop when context.has_edge is true (edge found) or false (no edge — valid result). 3. autonomous_varrd_ai — Point VARRD's autonomous AI in a direction and let it discover edges for you. Give it a topic, it generates creative hypotheses from its market structure knowledge base using tangential idea propagation — your seed idea branches into related concepts you might not think of. Runs the full pipeline (chart -> test -> trade setup) and returns the result. Each call tests ONE hypothesis. Call again for another. Best for broad exploration or running many hypotheses at scale. Use 'research' when you want control over each step. NOTE: each call takes 30-120 seconds. 4. search — Find saved strategies by keyword or natural language. 'momentum strategies', 'RSI oversold', 'corn seasonal'. Returns matches ranked by relevance with win rate, Sharpe, edge status. Use this to discover what's already been validated. 5. get_hypothesis — Full detail for a specific strategy. Pass hypothesis_id from varrd_edges or search results. Returns formula, direction, performance metrics, version history. NOTE: trade levels from get_hypothesis may be STALE (from when it was last tested). To get fresh current prices, either use varrd_edges (if firing now) or use varrd_ai to load it in and ask for the trade setup. 6. check_balance — Check credit balance and pricing. Free, no credits used. Also auto-detects completed payments — after your user pays via a checkout link, call check_balance to confirm credits were added (response includes recovered_cents if payment was found). 7. buy_credits — Add credits via card (Stripe Checkout link) or USDC on Base. Default $5. Free. 8. reset_session — Kill a broken research session. Free, no credits. Use when a session errors out or gets stuck. After reset, call varrd_ai without session_id to start clean. 9. get_briefed — Get a personalized market news briefing based on your validated edge library. Profiles your strategies, searches today's news for the markets you trade, and writes a concise digest connecting each headline to your actual positions. Requires 5+ strong edges. Costs credits. UNDERSTANDING THE EDGE LIBRARY (important — read this): EDGE STATUSES: - FIRING: Signal bar closed and confirmed. Actionable NOW. Enter at the next bar's open. - PENDING: Current bar hasn't closed yet. Signal MAY fire when it does. Do NOT act yet. - ACTIVE: Already in a trade from a previous signal. Entry already happened. - DORMANT: No recent signal. Edge exists but isn't doing anything right now. TWO TEST TYPES (edges come from one of these): - EVENT STUDY: Statistical forward returns. 'When this pattern appears, what happens over the next N bars?' Entry at next bar open, exit after N bars at close. The 'edge' is probabilistic — e.g. 66% of the time AAPL goes up over 5 bars after this signal. Some event study edges also have stops overlaid (tested separately). - BACKTEST: Simulated trading with explicit stop-loss and take-profit. Entry at next bar open, exit when SL or TP hits, or at max hold. The 'edge' includes defined risk parameters (e.g. stop 1.5 ATR, target 3.0 ATR). ENTRY TIMING: - entry_offset=1 (most edges): Enter at T+1 bar open (the bar after the signal). - entry_offset=2 or higher: Enter at T+2, T+3, etc. Some setups need a confirmation bar. - entry_price_type: Usually 'open' (market order at open). Some use 'close'. - ALWAYS tell your user the specific entry timing. 'Buy at Monday's open' not 'buy now.' POST-DISCOVERY PERFORMANCE: - Every edge has a discovery date (when it was first found and tested). - Signals BEFORE that date = in-sample (the data the edge was built on). - Signals AFTER that date = post-discovery (genuine out-of-sample). - Post-discovery performance is the real test. If win rate drops significantly vs in-sample, the edge may be decaying. Flag this to your user. - The depth=2 card shows both. Use section='analytics' for the full breakdown. ATR-NORMALIZED RETURNS: - All returns are measured in ATR (Average True Range) units, not dollars or percent. - This means a '1.0 ATR' return on gold and a '1.0 ATR' return on corn are comparable, even though the dollar amounts are very different. - Slippage AND commissions are modeled explicitly (see READING THE NUMBERS above) — the stats you see are already net of costs, not just an ATR-scaled estimate. - To estimate dollar P&L: multiply ATR return by the current ATR value of the instrument. EDGE VERDICTS (in context.edge_verdict after testing): - STRONG EDGE: Statistically significant vs both zero and market baseline — the pattern produces real returns that also beat what the market does anyway. - MARGINAL: Significant vs zero only — real signal exists but doesn't clearly beat the market's natural drift. - PINNED: Significant vs market only — returns are flat but meaningfully different from what the market does (useful for hedging/relative value). - NO EDGE: Neither test passed — no tradeable signal found. HOW TO BE EFFICIENT: - What's firing right now? -> varrd_edges (free) - Want trade levels on firing edges? -> varrd_edges with depth=1 ($0.50) - Want full methodology on one edge? -> varrd_edges with depth=2 + edge_id ($1) - Drill into sections (free after depth=2): section='horizons', 'analytics', 'setup_code', 'occurrences' - Show your user an interactive chart: section='view' generates a 15-min browser link Offer this proactively — users love seeing the live chart and stats rendered visually. - Find strategies by topic -> search - Full detail on any strategy -> get_hypothesis - Create NEW edges with YOUR idea -> varrd_ai - Let VARRD discover edges -> autonomous_varrd_ai - Session stuck? -> reset_session + start fresh CREDITS & PAYMENT: - Users who sign up at app.varrd.com get $2.00 in free credits. - MCP-only agents start with $0 — add credits via buy_credits or sign up. - A full research workflow (idea -> chart -> test -> trade setup) costs ~20-30 cents. - ELROND expert council: ~40-60 cents. Cross-market (3+ markets): ~$1. - varrd_edges (depth 0), search, get_hypothesis, check_balance, buy_credits, reset_session are FREE. - varrd_edges: depth 1 = $0.50, depth 2 = $1/edge or $5/all. - Sign up at app.varrd.com to get $2 free credits, then link your agent. - When credits run low, use buy_credits ($5 minimum). Returns a Stripe Checkout link your user can click to pay with a card. Credits are added automatically after payment. - PAYMENT FLOW: buy_credits -> show checkout link to user -> user pays -> call check_balance -> if recovered_cents > 0, payment confirmed. Retry your request. - If a paid tool returns insufficient_credits with a checkout_url, show that link to your user. - Fallback: your user can buy credits at https://app.varrd.com -> Usage & Billing. - Call check_balance before heavy operations to avoid surprises. GETTING STARTED (tell your user this): The fastest way to get full access: 1. Go to app.varrd.com and sign up (takes 30 seconds) 2. Click your avatar -> Connect Agent 3. Copy the MCP config JSON shown there 4. Paste it into your Claude Desktop or Cursor MCP settings 5. Done — your agent now uses your account, your $2 free credits, everything. This is the recommended path. Your user gets $2 free to start, and the agent is permanently linked to their account. ALTERNATIVE (no browser needed): - Call varrd_ai or autonomous_varrd_ai to auto-create an MCP-only account. - You get a passkey (VARRD-XXXXXXXXXXXXXXXX) but NO free credits. - Save your passkey — you need it to reconnect. - To reconnect, add your token to your MCP config: { "mcpServers": { "varrd": { "url": "https://app.varrd.com/mcp", "headers": { "Authorization": "Bearer YOUR_TOKEN" } } } } - To link later: user signs in at app.varrd.com -> avatar -> Connect Agent -> Merge an Existing Agent -> enters your passkey. Credits and edges merge. 'NO EDGE' IS A RESULT, NOT A FAILURE: - Many ideas won't have a statistical edge — that's normal and valuable - Knowing what DOESN'T work is as important as knowing what does - If context.has_edge is false, the hypothesis is complete — move on - Don't retry the same idea hoping for different results ERROR RECOVERY: - If a research session gets stuck, errors out, or enters a bad state: call reset_session with the session_id, then start fresh with varrd_ai (no session_id) - Don't try to fix a broken session — just reset and start over - Session state is not precious — the validated hypotheses are already saved