~/polyedge/data/snapshot_*_pre_clean.json if we need to restore.Polymarket markets moving with no matching news in our feed. Where the market knows something we don't. Runs hourly.
Highest-signal accounts / keywords we aren't watching yet. Copy the value, add to the indicated config, reload the agent.
Data-driven recommendations generated from positions, scenarios, audit findings. Rebuilds every scan. Act on P1 first.
One card per play. This is where we decide to scale, tighten, or cut. Mistakes + lab trades excluded from these numbers.
Config changes + position housekeeping applied during the audit remediation.
Each play type has its own entry rules, take-profit target, and stop-loss threshold. Rules are defined in polyedge.py and synced here every 15 minutes.
News headlines matched to Polymarket markets. Score 5+ = strong match. 2+ different sources = convergence alert (emails + auto-trades).
Markets that meet the criteria for Time Decay or Crypto plays right now. Not all get auto-traded - some may be too close to thresholds or already traded.
Markets with the biggest price swings in 24h. Big movers >20% on >$500k volume trigger auto-trades.
Every Polymarket market we're scanning. Color-coded by our relationship. Top 80 by 24h volume. Click any tile to open on Polymarket.
Chronological narrative: what worked, what we changed, what we added, why.
Every rule change, simulated against historical trades. Shows what would have happened WITH the change vs WITHOUT. Lets us validate changes against 2-3 weeks of real data instead of waiting for new trades.
Per-play readiness score. Green = commit full size. Amber = small/half-size. Red = park. Based on evidence, sample size, recent performance, and how well the rule is validated.
Week-by-week P&L per play. Shows when each play was working and when it stopped. Rule change and event timeline annotated so you can tie cause to effect.
Claude reviews the full portfolio once a day and produces a strategic audit: portfolio health, new strategy ideas grounded in the last 24h of data, alternative ways of positioning, and early-warning watch-outs. Runs via launchd at 09:00 UK time.
Dynamically ranked ideas based on what's actually working in your portfolio right now. Rebuilds every scan.
Celebrity, music, awards and sports markets. Half-size stakes ($50), separate P&L from the main pot. These trades don't affect the Overview or Plays numbers. Here to collect data on whether these categories produce edge.
Multi-outcome events where sum of YES prices != $1. Buy all YES (if sum < 1) or all NO (if sum > 1) for a guaranteed return after fees. Scanner runs every 15 min. Log at data/neg_risk_opportunities.jsonl.
Claude estimates the true probability of each liquid, non-extreme market. When its estimate diverges from the market price by 15 percentage points or more, we flag an edge. Capped at 5 calls per scan (~$0.03/day).
Every $5,000+ trade on Polymarket, logged with wallet address and trader handle. Leaderboard ranks wallets by total notional traded. Research says copy-trading whales has turned $313 into $438,000 in documented cases.
Every headline that matched a Polymarket market, captured with its published time and the Polymarket YES price at the moment of match. Data at data/news_timing.jsonl.
Three categories: (1) Solo sources - 1 headline matched a market with score ≥5 but convergence required 2 (if we'd widened the rule, we'd have caught these). (2) Converged but didn't trade - 2+ sources hit but the market filter blocked (volume, price extreme, watchlist). (3) Would have lost badly - markets where looser rules would have burned us. Simulated P&L uses current live Polymarket price.
Replays historical trades with alternative rule sets and shows what P&L they would have produced. Use to test hypotheses (drop a play, exclude a day of week, raise a score threshold) before making rule changes live. Auto-runs after every scan.
Historical P&L recomputed under different rules. Forward rule changes (surgical BIG_MOVER, wider stops, NEWS_EDGE scope) are live - historical trades stay untouched so you can compare like-for-like.
Trades caused by bugs or unintended behaviour, not genuine strategy outcomes. Filtered out of main Overview P&L so we analyse the strategy we INTEND to run. Each entry tagged with the reason we caught it so we remember the lesson. Auto-flagged by ~/polyedge/flag_mistakes.py. Manual flag: python3 ~/polyedge/flag_mistakes.py --flag SLUG "reason"
Automated checks across the whole stack: data integrity, filter logic, stale slugs, launchd agent health, feed responsiveness, P&L reconciliation, Polymarket market state. Runs after every scan. Anything red needs fixing.
Live snapshot of scanner configuration. To change: edit polyedge.py and reload launchd.