Polymarket signal analysis · interactive summary

Yes — historically better after capital drag. But only if live execution survives.

This page combines the trading opportunity and the monitoring system: what the edge looked like after realistic VWAP/capacity/opportunity-cost haircuts, and how to turn it into alerts for upcoming active markets without fooling ourselves.

Investment — default conservative prefunded net ROI
Historical alert rule — TP/(TP+FP) on metadata-locked holdout
Continuous warning mode — median lead at <1 false warning/year
Main failure mode Execution spread/depth/slippage/live decay, not KKM

Investment opportunity

Capital-cost adjusted edge calculator

Default: index 7% + KKM 5.18% = 12.18% annual drag.

Formula prefund cost = peak capital × annual drag × 1.76y active cost = stake × annual drag × holding time

Active-only cost is tiny because positions resolve fast. Prefunded-wallet cost is the conservative view if capital sits ready on Polymarket waiting for alerts.

Loading…

Execution realism ladder

Bars compare gross ROI versus conservative prefunded net ROI at your selected annual drag.

What this means

Loading…

  • Do not trust the +80–100% toy returns as deployable without live order-book depth.
  • The realistic VWAP proxy is the anchor: lower return, but much more believable.
  • Event-deduped view matters because several markets can be one real-world event.

Rate sensitivity table

selected strategy

Capacity and fill realism

How much money could the signal plausibly absorb?

Historical proxy: post-alert matched flow, not true alert-time L2 order book.

Capacity proxies

Total stake and gross ROI by sizing rule.

Largest post-alert YES-buy liquidity examples

Alerting system

Historical warning performance and false-warning budget

Alert = investigate now; not proof the event will happen.

Choose a false-warning budget

Frontier from continuous warning backtest. Lower false alarms means shorter/less complete warning.

Alerting interpretation

Loading…

What historical alerts looked like at ~4 false positives/year

Upcoming / current markets

Latest local scan: active alerts and deadline families

Loading…

Top active term rows

Calendar package check: buy later-YES + earlier-NO only locks profit if cost is below 1 and wording is truly nested. Current rows are mostly near-misses/diagnostics, not automatic trades.

Operational runbook

How to turn the research into a live alerting workflow

The workflow posts only meaningful alerts; no raw progress dumps.

1

Poll active markets

Every 5–15 minutes: Gamma metadata, CLOB prices/books, Data API recent trades.

2

Compute alert-time features

p(24h/6h/1h), volume surge, trade count, net Yes pressure, spread/depth, theta-adjusted hazard for deadlines.

3

Group siblings

Collapse related by-date markets into one event family so Iran/Ukraine clusters do not spam ten “independent” alerts.

4

Escalate only unexplained moves

Attach market link, feature snapshot, book depth, sibling family summary, and a public-news snapshot.

5

Paper-trade first

Log best ask, fill simulation, rejected fills, slippage, and post-alert VWAP before risking real money.

6

Review misses/false alarms

Weekly: re-score false positives, missed Yes events, and changes in market microstructure.

Recommended alert text shape


        

        

          
Red-team guardrails

          
    
            
  • Close/resolution time is not always real-world event time.
    • 
            
  • Near-certainty markets can be settlement mechanics, not useful alpha.
    • 
            
  • Backtest rows are correlated by event family.
    • 
            
  • Live edge can vanish if other people copy the signal or books get thinner.
    • 
            
  • No real trading until paper-fill logs prove spread/depth is acceptable.
    • 
          

        

      

    


    

      

        

          
Auditability

          
Data files behind this page

        

        
—

      

      

        
Source files

        
Verification summary