Overview
👋
Welcome to Raven
Raven auto-discovers trading ideas from Twitter, backtests them against Polymarket history, paper-tests the survivors for 7 days, then deploys a small amount of real capital on the ones that pass. You don't build strategies by hand — you review what Raven finds.
Your command centre. Shows live stats across all your signal configs and a feed of recent pipeline jobs. Green numbers mean the system is working.
Heatmap of Alpha
System Pulse
ScraperActive
Triage EngineActive
Market MakerLoading…
Pipeline snapshot
Loading status summary…
Queued
—
Running
—
Pass
—
Fail
—
Consolidated into Overview in Phase 5b. Technical pipeline table remains as a hidden legacy view.
Recent pipeline jobs Live feed
| Name | Status | Hit rate | Verdict | Plugin | Ingest | What happened | Date |
|---|
Twitter / X — find ideas before you trade
Raven watches public tweets for you. You choose which accounts and which words matter. Matching posts land in the columns below so you can skim, sort, and send the good ones to backtesting (a historical “what if I had traded this?” check).
Watch List Settings
Keep it simple
- You do not need this page. If you only use My Signals and Run Backtest, you are doing it the straightforward way — ignore Twitter ideas.
- Use this page as an extra inbox. Same end goal (backtests on My Signals), but you skim tweets in columns first, then hit Promote. Link a signal when you save the watch list if you already have one; otherwise the first Promote can create a signal for you — no need to understand the plumbing.
Simple version: tweets show up as cards. Drag-style buttons move a card between columns.
Promote sends a tweet into the same backtest pipeline as My Signals (if you linked a signal to this watch list).
Technical: Stage-1 intake → triage → Kanban → optional Stage-2 pipeline_jobs.
Usage today
ⓘLoading…
Set up a new watch list
You are creating a saved recipe for what to pull from Twitter/X. You do not need to know handles by heart — use the suggested accounts or search below.
Words to watch for
ⓘKeywords = things people might tweet about that relate to Polymarket-style events (economy, sports, politics, crypto, etc.).
Selected keywords
Twitter / X accounts
ⓘTap + add on any row. You can still type a handle yourself under “Find or add” if it is not on the list.
Popular with other Raven users (updates over time)
Loading suggestions…
Selected @handles
Discovered Strategies
Raven reads Twitter every 15 minutes, passes candidate tweets through an LLM strategy generator, backtests the output, and auto-promotes passes to a 7-day paper soak. Each card below is one strategy in flight — no setup required. Use the Studio co-pilot to author custom strategies.
Auto-generate strategies
Runs every 15 minutes when on.
Loading…
Filter cards
Markets
Cached from Polymarket discovery. Run discover_markets to refresh.
Cached list of active Polymarket markets. Raven uses this to match your signals to tradeable markets. Refresh with the discover task if the list looks stale.
| Question | Volume | End | YES | NO |
|---|
Pipeline
Scanning
—
tweets being watched
Tested
—
strategies with an edge
Proving
—
running on live data
Trading
—
with real capital
Each discovered strategy flows left-to-right: tweets scraped from Twitter → LLM-generated strategies backtested → passes promoted to paper soak → profitable papers promoted to live.
Loading…
Backtest
—Loading…
Paper
—Loading…
Live
—Loading…
Backtest verdicts
Completed backtests per hour (last 7 days). Pass (green), fail (red), waiting on data (dashed amber), other (slate).
No completed backtests in this window yet.
Technical log (raw job rows)
| Expand row | Config | Job type | Status | Hit rate | Verdict | Plugin | Ingest | What happened | Actions | Created |
|---|
Market Maker
Set up the Market Maker, pick how much capital to risk, and monitor live activity. Paper trading only — no real money until you switch to a live stage.
🎛️ Live Floor
Inventory, quotes, fills, and skip reasons at a glance. Updates every cycle.
No MM config found. Configure one in .
Config
Loading…
Quote Ladder
P&L Decomposition
Live Fill Feed
Skip-Reason Feed
📑 Phase Passport
Where each strategy is in the Backtest → Paper → Live lifecycle.
No MM config found. Configure one in .
Strategy
Loading…
Backtest
⊘ N/A
MM strategies validated by paper soak, not historical backtest.
Paper
—
Loading…
Live
○ Not started
Eligible after 30d paper pass · requires ≥$50 realized.
Selected: PAPER
Backtest detail
Loading… (Step 3.6 — honest empty state for MM)
Live detail
Loading… (Phase 4 — when live trading lands)
🦝 Poeboy
Autonomous trading partner — what he's thinking and what needs your nod.
📝 Pending approvals · …
Tap Approve, then Apply. Whitelist on params.Loading…
📊 Today's scorecards
Rolling 7-day per running config.
Loading…
🦝 Recent cycles
Heartbeat 4h · daily review 06:30 · orchestrator 6h · optimizer Monday.
Loading…
✓ Recently applied
Decisions that already moved a mm_config column.
Loading…
🧪 Studio
Describe a strategy in plain English. Raven writes the code, runs a backtest, and shows you the result.
Start mode
Market context
Select a signal to load context.
1 Saved & co-pilot
→
2 Python
→
3 Results
Strategy code
Tip: ⌘K or CtrlK opens inline edit
Version history
Restore loads the saved snapshot for that accepted edit (newer accepts keep their snapshots).
Save history
Portfolio
Follow strategies from paper soak → live on the Pipeline & trades tab, then switch to Positions for the execution ledger (open, closed, awaiting price).
Trade Flow
Dashboard view of strategies moving through the pipeline. The equity curve below covers both paper soaks and live positions. Empty rows here mean no strategy has reached paper or live yet — check the Discovered Strategies feed for earlier stages.
Equity Curve
Live
Paper
Chart loading or no data available...
| Market | Strategy | Stage | Side | Entry | Current | P&L | Actions |
|---|
Positions
All trades Raven has placed or is tracking. Open = currently in market. Closed = resolved. Awaiting Price = signal fired, order not yet executed.
Open / Paper / Live
0 | Paper: 0 | Live: 0
Won / Lost / Win Rate
0 / 0 (0%)
Unrealised / Realised / Total P&L
$0 | $0 | $0
P&L Dashboard
Your running profit and loss across all strategies and time periods. Equity curve shows cumulative performance. Daily bars show day-by-day results. The per-strategy snapshot ranks closed-trade P&L visually; the table below keeps exact averages and profit factor.
Equity Curve
Daily P&L (Last 30 Days)
Per-strategy P&L snapshot
Bars share one scale (largest win or loss in this list). Solid green extends right from the centre line; striped red extends left — colour plus pattern encodes sign.
Per-Strategy Breakdown
| Strategy | Trades | Win Rate | Total P&L | Avg Win | Avg Loss | Profit Factor |
|---|
Drawdown
Max Drawdown (All Time)
—
Current Drawdown
—
Daily DD vs Limit
$0 / $0
Inline edit
Describe the change. Nothing applies until you accept the diff.
Precise
Balanced
Bold
Precise keeps edits surgical. Balanced is the default. Bold allows larger rewrites — always read the diff before accepting.
Current: Balanced
Preview diff
Rename strategy
Display name is safe anytime. Changing the plugin slug updates how signals reference this code.
Replay Trade
Strategy Performance
Detailed analysis of a backtest run. Select a job to see its equity curve, drawdown profile, win/loss distribution, and Kelly sizing. These are historical results — they don't guarantee future performance.
🧪
No backtest for Market Maker strategies
Market Maker is validated by live paper-trade soak rather than a historical backtest — the Avellaneda-Stoikov quote path doesn't have a one-shot historical equivalent. Look at paper P&L + fills instead of a verdict curve.
→ Open Phase Passport
Verdict
Equity Curve
Drawdown
Win/Loss Distribution
Kelly Sizing
Recent step risk: absolute change in equity between backtest steps (last 40). Shown next to Kelly for context — not a second Kelly estimate.
Signal Decay Curve
Market Regime Heatmap
Live vs Paper Comparison
No live trading data yet. Positions will appear here once trades are closed.
Equity Curve (Live)
Risk Parameters
Per-signal guardrails. These limits control how much capital Raven can deploy on a single trade, how many positions can be open at once, and when to stop trading. Set these before activating any signal.
Signal Config
Select a signal to view and adjust its risk parameters.
Position Sizing
$1$500
$1$500
120
Risk Controls
Kelly fraction
0.25
0.05 (conservative)0.50 (aggressive)
Fraction of Kelly criterion to apply. Quarter-Kelly (0.25) is the standard conservative choice.
1% (tight)50% (loose)
Exposure Preview
With a $1,000 bankroll: max $25 per trade, $75 total exposure across 3 positions.
Conversations
Thinking…
Operator Health
Test Center
Automated health checks across platform infrastructure and pipeline stages. Results are saved for 30 days.
Recent history
Alerts
Market Maker
About MM paper vs pipeline paper
MM "paper" simulates post-only CLOB quotes and fills (api/mm/paper_fills.py)—it is not the same as dashboard pipeline backtest verdicts (Postgres historical engine) or the CLI 7-day paper gate.
Inventory in the table is YES shares; the cap max_inventory_usd is enforced on USD notional (≈ shares × mid). See docs/STRATEGY-PARITY.md and docs/STRATEGY-HARNESS.md in the repo.
New MM Config
300 bps
0.10
Config Detail
Active Markets
| Market | Mid | Bid | Ask | YES sh | Notional | Cap % | Buy ID | Sell ID |
|---|
Recent Fills
| Time | Market | Side | Price | Size | Mode |
|---|
Settings
Market cache
Open markets cacheMarket discovery is now accessed from Settings. This opens the cached active-markets table and pin-to-signal tools.
Integration tests
Run all integration checks at once. Individual test buttons are on each card below.
Polymarket
—
First time? Read the wallet setup guide →
PolyBackTest Data API
—Optional: connect your PolyBackTest key for crypto Up/Down + Binance research data. Keys are encrypted; Raven never displays them after save.
Connection status
Click Refresh to check PolyBackTest (uses server-side key: yours or deployment fallback).
LLM Key
—
API key: ••••••••
Billing
Subscription and Stripe customer (source of truth for your tier after checkout).
- Current tier
- —
- Next billing
- —
- Stripe customer
- —
Webhook URL
Receive alert payloads via POST JSON when sentiment spikes fire.
Twitter search
Confirms your deployment can reach the configured Twitter data path. This does not change intake settings.
API Keys
Programmatic access to the API. Keys are shown only once when created.
| Name | Created | Last used | Expires |
|---|
Help
How Raven Works
🐦
Twitter Signal
Detect market-moving tweets
📊
Backtest
Validate against history
🧪
Paper Test
7-day simulated trading
🚀
Live Trading
Real money on Polymarket
🐦 Twitter Signal
What happens: Raven monitors specific Twitter accounts and keywords. When a matching tweet is detected, an LLM scores its market relevance and extracts a directional signal (BUY/SELL).
Pass criteria: The LLM confidence score must exceed the configured threshold.
Duration: Seconds — signals fire in near real-time.
What can go wrong: Twitter API rate limits, LLM hallucination, stale account lists, or keyword noise from unrelated tweets.
📊 Backtest
What happens: Raven replays the signal against historical Polymarket price data. It simulates what would have happened if the signal had been traded over the lookback period.
Pass criteria: Hit rate ≥ 55%. The signal must correctly predict price direction more often than not.
Duration: 10–60 seconds depending on data volume.
What can go wrong: Insufficient historical data, overfitting to past patterns, or survivorship bias if the market has already resolved.
🧪 Paper Test
What happens: Raven trades the signal in simulation mode for 7 days using live market prices but no real money. Positions are tracked as if they were real.
Pass criteria: Positive P&L over the 7-day window and hit rate remains above threshold.
Duration: 7 days.
What can go wrong: Market conditions during the test period may not be representative. Slippage and liquidity aren't fully simulated.
🚀 Live Trading
What happens: Raven places real orders on Polymarket using your connected wallet. Positions are sized according to your Risk Parameters (Kelly fraction, max position, stop-loss).
Pass criteria: N/A — this is live. Monitor P&L and drawdown continuously.
Duration: Ongoing until you deactivate the signal or a stop-loss triggers.
What can go wrong: Slippage, liquidity gaps, API failures, market manipulation, or model degradation. You can lose money. Past backtest/paper results do not guarantee live performance.
Frequently Asked Questions
Raven is a signal-to-live trading pipeline for Polymarket prediction markets. It scrapes Twitter for market-moving signals, backtests them against historical price data, runs a 7-day paper test, and — if the signal passes both gates — executes live trades on your behalf. It automates the entire flow from hypothesis to execution so you don't have to watch markets manually.
You define a signal config with Twitter keywords, tracked accounts, and an entry side (BUY or SELL). When a matching tweet appears, Raven scores it with an LLM for relevance and direction. If it passes, the signal enters the pipeline: first a historical backtest, then a 7-day paper test, then live trading. Each stage has pass/fail criteria — a signal that fails at any stage is stopped from progressing.
It is an optional inbox for tweets before they become trades. You save a watch list (a friendly name + Twitter accounts + keywords). Raven fetches public posts, sorts them into columns, and you can send the promising ones to the same backtest step as My Signals. You do not have to type raw @handles — use the suggested list or search, same as the first-time setup wizard.
A backtest replays a signal against historical data. Raven looks at past instances where the signal would have fired and checks whether the market price actually moved in the predicted direction within the lookback window. The hit rate is the percentage of correct predictions. A hit rate of 55% or higher is the default pass threshold — anything below that suggests the signal isn't reliable enough to risk real money on.
Paper trading is a 7-day simulation using live market prices but no real money. It exists because backtests use historical data, which can be misleading — markets change, and past patterns don't always repeat. Paper trading tests the signal in current market conditions. If a signal passes backtest but fails paper, that's a strong warning that something changed and the signal shouldn't go live.
Hit rate is the percentage of trades where the market price moved in the direction the signal predicted. A 60% hit rate means 6 out of 10 trades were correct. The default pass threshold is 55% — above coin-flip odds, but not by much, because prediction markets are efficient and consistent edges are rare. You can adjust this threshold per signal if you want stricter or looser gates.
The LLM (large language model) interprets tweets to extract market-relevant signals — it reads the text and scores whether it's bullish, bearish, or irrelevant for a specific market. However, the LLM does not make trading decisions. The actual gate is the data: a signal must pass a historical backtest and a live paper test before any trade is placed. The LLM is the ears; the backtest is the brain. If the data says the signal doesn't work, the LLM's opinion is overridden.
When a live-approved signal fires, Raven calculates the position size using Kelly criterion (scaled by your Kelly fraction setting), checks it against your max position and max concurrent position limits, and places an order on Polymarket via the CLOB API. The position is then tracked in the Positions view. If the price moves against you past your stop-loss threshold, Raven will attempt to close the position automatically.
Yes, you can lose money. This is real trading with real funds. Specific risks include: Signal risk — the LLM may misinterpret a tweet or the pattern may stop working. Execution risk — orders may not fill, or fill at worse prices than expected (slippage). Model risk — backtest results don't guarantee future performance; markets change. System risk — software bugs, API outages, or network failures can cause unexpected behaviour. Total loss risk — in extreme scenarios, you could lose your entire deposited amount. Never trade with money you can't afford to lose. The risk parameters exist to limit exposure, but they cannot eliminate risk entirely.
Max position size caps the USD amount in any single trade. Max concurrent positions limits how many trades can be open at once, controlling total exposure. Kelly fraction determines position sizing — full Kelly is mathematically optimal but volatile; quarter-Kelly (0.25) is the conservative standard. Stop-loss automatically exits a position when the unrealised loss exceeds this percentage. These are guardrails, not guarantees — set them conservatively, especially when starting out.
Common fixes: Adjust lookback hours — too short and you catch noise, too long and the edge decays. Try 4h, 8h, and 24h. Narrow your keywords — broad keywords fire on irrelevant tweets. Check your accounts list — not all Twitter accounts move markets equally. Flip the entry side — if your signal is consistently wrong, the opposite direction might be right. Try a different market — some markets are too efficient for Twitter signals. If nothing works, the signal might genuinely not have an edge — that's useful information too.
Raven is a software tool — it does not custody funds or act as a broker. However, it interacts with Polymarket, which has its own jurisdictional restrictions. You are responsible for ensuring that prediction market trading is legal in your jurisdiction. Raven's Terms of Service require you to confirm this at registration. If you're unsure, consult local legal counsel before using the platform.
For technical issues or account problems, reach out via the support channels listed in Settings. For bugs, include: what you did, what you expected, and what actually happened. Include any error messages you saw. Response times vary — Raven is a small team and we prioritise issues that affect live trading.
Glossary
Watch list
A saved set of Twitter/X accounts and keywords Raven should search. Shown under Twitter ideas. Not the same as a full signal — link a signal if you want one-click backtests from promoted tweets.
Signal
A Twitter-based trigger that Raven watches for. When a matching tweet is detected, the signal fires and enters the pipeline.
Backtest
A historical simulation that replays a signal against past market data to measure its predictive accuracy.
Hit Rate
The percentage of trades where the market moved in the predicted direction. 55% is the default pass threshold.
Paper Test
A 7-day simulated trading period using live prices but no real money. The final validation before live trading.
Live Trade
A real trade placed on Polymarket with actual funds from your connected wallet.
Equity Curve
A line chart showing cumulative portfolio value over time. An upward slope means you're making money.
Drawdown
The decline from a portfolio's peak value to its lowest point. Measures the worst losing streak.
Kelly Fraction
A mathematical formula for optimal bet sizing. Quarter-Kelly (0.25) is standard — it sacrifices some growth for much lower volatility.
Stop-Loss
An automatic exit trigger that closes a position when unrealised loss exceeds a set percentage of entry value.
Entry Side
The direction of the trade: BUY (bet YES, price will rise) or SELL (bet NO, price will fall).
Lookback Hours
How long after a signal fires to check whether the market moved. Shorter is more reactive but noisier; 4h is a good default.
Profit Factor
Gross profit divided by gross loss. Above 1.0 means profitable; above 1.5 is good; above 2.0 is excellent.
Adverse Selection
When you consistently trade against better-informed counterparties. A sign your signal is known to the market and priced in.
Awaiting Price
A position state where the signal fired but the order hasn't been executed yet. May need manual attention if stuck.