FUND PERFORMANCE
Portfolio value over time
Fair valueStarting capital
AUTONOMOUS COPY FUND
Dashboard
BOT STATEOff
Doing nowWaiting to start
Last cycle—
Next check—
Heartbeat—
Real-time feed—
VIRTUAL FUND
The bot is ready to track TOP traders
A trade is only opened as a reaction to a new BUY/SELL by a tracked TOP wallet.
FUND STATE
—
AWAITING DATA
AI SUMMARY
What is happening now
Result breakdown is loading…
ACTIVITY
Live decision feed
EXPOSURE
Portfolio allocation
TRADER EVALUATION
Collecting data
Operational diagnostics and technical metrics
Open copies
0| Trader | Market | Entry | Mid / bid | Fair / liquidation P&L |
|---|
TOP wallet status
tracking → evaluation → copy → resultWAITINGCOPYINGVERIFIED
INFORMATION OVERVIEW
All detected trades of TOP wallets
The bot decides autonomously. Here you only see what it found, copied or rejected, and why.
AI QUALITATIVE VETO
Markets Claude flags to block — you approve
Claude reviews current markets only for concrete red flags it can BLOCK (ambiguous resolution, mis-scraped/duplicated data) — it never opens trades. Approved vetoes stop new copies of that market via a cheap blocklist; the trade hot path stays deterministic.
| Time | TOP trader | Market | Action | Trader price | Our price | Copy P&L | Decision | Reason |
|---|
AUTOMATIC SELECTION
Tracked TOP wallets
Ranked primarily by our Copy P&L. The trader's historical profit is only an entry filter.
TRADER INTELLIGENCE
Trader management
Traders are split into three clear tiers. Watchlist only collects data, Managed Portfolio holds actively copied sources, and Archived keeps disabled or paused traders.
ADAPTIVE LEARNING
Learning Center
PolyPilot does not learn to guess market outcomes. It learns from our closed copies which traders are truly copyable for us, when their edge fades, and how to adjust capital weight.
WHAT THE SYSTEM LEARNED
Current insights
SAFE LEARNING
How weights change
- A new trader starts in the Watchlist and only collects data.
- After enough closed copies, Copy ROI, win rate and drawdown are evaluated.
- A positively verified trader can move to Managed Portfolio.
- A negative or technically non-copyable trader is limited or archived.
- No change to trading logic happens on its own on live money.
LEARNING SAMPLE
Results by trader
| Trader | Sample | Copy ROI | Profit factor | AI score | Status |
|---|
WHEN TO COPY
Segment performance
| Segment | Sample | ROI | Win rate | AI score |
|---|
COUNTERFACTUAL BOOK
Shadow Lab
Every rejected BUY is recorded as a shadow trade and settled at resolution. That gives an honest answer to the only question that matters for each filter: is it protecting the fund, or is it throwing away money?
WHERE THE EDGE IS
Return by entry price
Favourite-longshot bias is the fund's central thesis. Executed copies and shadow (rejected) trades are kept apart on purpose — mixing them would flatter the record. Read "event groups" before trusting a row: a hundred copies of the same 5-minute market are one bet, not a hundred.
| Entry price | Shadow trades | Event groups | Shadow ROI | Win rate | Market calibration | Executed | Executed ROI |
|---|
FILTER CALIBRATION
What each filter costs or saves
A filter with a positive shadow P&L rejected trades that went on to win — it is costing you money. A negative one is doing its job. Verdicts only appear once the sample is large enough.
| Filter | Rejected | Event groups | Shadow P&L | ROI | Would have won | Verdict |
|---|
SOURCE DIAGNOSTICS
Live wallet feed
This table shows everything that actually arrived from the Polymarket Data API: checkpoints, new trades, delayed records and errors. If it is empty, monitoring has no data.
| Time | Trader | Status | Action | Market | Price | Age | Detail |
|---|
APP BASICS
Connect my Polymarket account
In virtual mode the account is used for a portfolio overview. Paper copy trading needs no secret credentials.
USER ACCOUNT
Google sign-in
Checking configuration…
Google OIDC is for the cloud/SaaS deployment. Local mode stays safely without required sign-in.| Market | Outcome | Value | P&L |
|---|
SIMPLE LIMITS
Autonomous fund settings
The research feed records every new event. The virtual fund only trades fresh events with a verified order book; delayed and unpriceable events never change equity.
AI CONFIG ADVISOR
Claude-proposed adjustments — you approve
Claude proposes bounded changes to safe filter/risk knobs based on the fund's state and the backtest findings. Nothing is applied until you approve each one; the deterministic engine then validates and clamps it.
Wipes the fund and starts fresh with the backtest-validated favourite-band strategy (entry 0.60–0.90, wider 60s freshness so it actually copies).
PROFESSIONAL PLAYBOOK
How to actually profit copying Polymarket
A trader-grade guide, not a sales pitch. It states where the edge is measured, where it is not, and how to run PolyPilot toward a small but real forward edge — never the viral fantasy.
THE HONEST THESIS
Naive copy-trading loses. A narrow, validated edge survives.
Copies only fill when the price has NOT run away after the trader bought — so you systematically get their weaker trades (adverse selection). Add latency, slippage and fees ("copy tax") and blind copy-all is negative in backtest (about −4.6%).
What DID survive walk-forward validation (positive in 6/6 time folds) is the favourite band: entering at prices 0.60–0.90, avoiding short-horizon crypto. Ranking wallets by past profit or win rate did NOT reliably persist (train→test copy-ROI correlation only ~+0.11) — that part is largely survivorship.
Realistic goal: a small, forward-persistent edge from WHERE you copy (segments/band) far more than WHO you copy. No 127x/month.
PROFIT CHECKLIST
Run it like a professional
- Apply the favourite strategy (Settings → "Apply the favourite strategy"): entry band 0.60–0.90, short-horizon crypto blocked.
- Keep the trade-age window tight — stale trades are exactly the adverse-selected fills that lose.
- Judge traders by OUR realized Copy P&L and Copy ROI, never by their Polymarket historical profit (that is only an entry filter).
- Let risk sizing bind: a copy can only shrink to fit book depth and your max-per-trade, never inflate.
- Enable auto-disable so wallets that lose money for us get archived after a real sample.
- Treat AI layers as advisors: they explain, propose and can veto — they never open a trade.
VALIDATION TOOLS
Prove the edge before trusting it
Offline, reproducible scripts (held-to-resolution, honest time splits). Run them before believing any strategy.
scripts/backtest_copy.pyDoes copying have positive EV at all, and in which segments (wallet / price band / category)?scripts/walkforward.pyDoes the favourite-band + wallet selection persist across expanding time folds?scripts/wallet_screener.pyDoes a wallet's past copy-ROI predict its future copy-ROI (or is it luck)?scripts/validate_score.pyDoes the production wallet SCORE (from period T) predict real forward return (T+1)? Honest out-of-sample, no look-ahead.
SCREEN BY SCREEN
What each page is for
DashboardLive fund state, equity curve, decision feed and exposure. Your at-a-glance control room.
OpportunitiesEvery detected TOP-wallet trade with why it was copied, rejected or is waiting. Plus the AI qualitative veto.
Top walletsThe tracked pool, ranked primarily by our Copy P&L — not by the trader's headline profit.
WatchlistThree tiers: Watchlist collects data, Managed Portfolio holds actively copied sources, Archived holds paused ones.
AI LearningWhat the system learned from CLOSED copies: which traders are truly copyable and when their edge fades.
Live wallet feedRaw truth from the Data API: checkpoints, new trades, delays and errors. Empty here means no data upstream.
SettingsBounded, safe knobs: profile, band, freshness, sizing, sample thresholds. The engine validates and clamps every value.
SAFETY
PolyPilot is a paper/virtual fund. It never sends a real order. The AI layers are human-in-the-loop and never trade. Nothing here is financial advice — it is a research and decision-support tool.