Polymarket Monitor

Quality pool entries at price >=0.85. Locked-in high-probability. OPT-49: lowered min from 0.90 to 0.85 AND raised max from 0.98 to 0.99 — previously the [0.90, 0.98] window missed BOTH borderline entries (skip 'price 0.490 outside elite bucket') and 0.99-priced settlements ('price 1.000 out of [0.9, 0.98]' n=12).

Quality pool, <2min lag, REVERSE the wallet's direction. Falsification arm. OPT-49c: price range 0.05-0.95 (was 0.10-0.90) — when we flip a high-prob entry (e.g. 0.945) the resulting price is 0.055 which got rejected at min=0.10. Widening to 0.05 catches these.

When market price moved >=3% in the last 30min, bet AGAINST the direction. Retail overreactions mean-revert in liquid markets. OPT-49: max_days 14->30 (466 'past end_date' was the dominant skip — 14d window excluded too many resolvable markets), threshold 5%->3% (5% in 30min is rare on Polymarket), edge_net 0.025->0.015.

Vegas line divergence. Bet Polymarket when it diverges >5pp from Vegas implied prob.

Last-60-minute orderbook inefficiencies. Walk the book on thin spreads when resolution is imminent. OPT-73: edge_net 0.04 -> 0.005. 133k evals / 0 fires post-deploy because realistic-cost engine eats 0.022-0.035 from the 0.05 raw edge, leaving net near zero or negative against the 0.04 wall.

Calibrated contrarian on Polymarket's BTC up/down 5m. When holder skew is extreme (>0.6) but BTC's 30s momentum doesn't confirm, fade them. Otherwise HOLD. Runs in polybot-claude-btc.service.

Quality pool, markets resolving within 6 hours. Late-info edge — the most informed money trades shortly before close. OPT-49: widened 60min -> 6h (360min). The 60min window was too narrow — most copy-trade signal events happen 1-6h before close. max_days 1->2.

Cross-venue arbitrage via pmxt (free tier). When the same event is priced >=3pp differently between Polymarket and Probable/Limitless/Kalshi/Myriad, bet the underpriced side on PM. Multi-venue arb is the most reliable edge in prediction markets.

Quality pool entries at price <=0.15. Extreme asymmetric payoff. OPT-49b: edge_net_threshold 0.020 -> 0.005 (we observed edge_net +0.0052 entries being rejected — still net-positive, ship them). OPT-65: price range [0.02,0.15] -> [0.01,0.20]. Skips were almost all 'price 0.015' (below low bound) or 'price 0.195' (above high) — both are valid deep-tail prices we should ship.

Lambda tail (price <=0.15 or >=0.85) with strict edge_net gate AFTER realistic costs. Quality-pool source only. Smaller bets, more of them.

Fade retail-driven price overreactions: >3% move in 60min where >5% of buy volume is from the retail pool. OPT-49c: min_retail 25% -> 5% — observed retail share consistently 0.05 (retail_pool of 500 wallets is small vs total trader population, so the share metric is naturally low even when retail is the marginal buyer).

Quality pool, markets with $5k-$50k volume — large enough for real liquidity but small enough that smart wallets still find edge.

Quote both sides on liquid markets. Spread + rebates. Market-neutral inventory.

Wallet must be in sharp_zone AND quality pool. Highest-conviction set. OPT-49c: max_open 30 -> 60 (saturated at 30 — was rejecting fresh signals).

Quality pool with hyper-tight 2-min lag cap. Speed beats size on signal decay.

Meta-agent. Fires only when 2+ of the above 4 produce same direction signal. Kelly-sized by combined confidence.

Calibration-weighted ensemble of 1,250 wallets. For each market, aggregate observed positions weighted by per-wallet calibration. Bet the gap to market price.

Buys when a sharp-zone wallet enters; previously needed 2+ to converge within 60min but that yielded only 1 fire in 45k evals. OPT-49b: 2 -> 1 wallet (now functions as a 'sharp follower' — its distinction from SHARP_PROVEN is the looser sharp_zone pool vs the PROVEN intersection with quality).

Follow lone elite sharps (n>=20 closed, WR>=80%) when they enter their proven price bucket. Different from SHARP_CONVERGENCE (which needs 3+); this trusts the wallet's per-bucket calibration solo.

Quality pool, Kelly-sized by per-bucket calibration. Same pool as TAIL_DISCIPLINED but no tail-price restriction. OPT-49c: price range 0.10-0.90 -> 0.05-0.95 (146 skips for price 0.945 in 5min — these are valid quality entries we shouldn't reject).

Copy whale only in their proven category. Per-whale x per-category calibration. Solves the -25% COPY_WHALE failure by gating on category-specific edge. OPT-49c: min trade $100 -> $20 (most whale entries are $10-50; $100 was blocking 79% of qualifying trades).

Any whale entry — looser counterpart to WHALE_PRECISION which requires per-category proven edge. Falsification arm. OPT-49: edge_net 0.020 -> 0.010 (skips showed -0.005 to -0.011 edge_net consistently — agent assumed_raw_edge=0.04 minus 0.02-0.03 fees crosses zero); min_source_trade lowered 150 -> 75 (more whale entries qualify).

Copy-side AGENT bot - was previously only on /copy-bots; unified onto /claude-bots per user request 2026-06-03. Full strategy detail on /copy-bots/AGENT_EDGE_NET_KELLY

Copy-side AGENT bot - was previously only on /copy-bots; unified onto /claude-bots per user request 2026-06-03. Full strategy detail on /copy-bots/AGENT_HOLD_TO_SETTLE

Copy-side AGENT bot - was previously only on /copy-bots; unified onto /claude-bots per user request 2026-06-03. Full strategy detail on /copy-bots/AGENT_LAMBDA_TAIL

Copy-side AGENT bot - was previously only on /copy-bots; unified onto /claude-bots per user request 2026-06-03. Full strategy detail on /copy-bots/AGENT_NOTHING_HAPPENS

Copy-side AGENT bot - was previously only on /copy-bots; unified onto /claude-bots per user request 2026-06-03. Full strategy detail on /copy-bots/AGENT_SPORTS_FLB