🧪 Beta · Research platform

Structural dislocation research
for Polymarket

Detect, validate and simulate pricing dislocations between structurally related markets. A research-first platform for studying convergence opportunities before risking capital.

Request access How it works

Currently running in research mode. No real capital involved.

⚠️ Experimental software. Simulation results do not guarantee future performance. This is a research tool, not financial advice. Use at your own risk.
The challenge

Related markets don't always respect their structural relationships

On Polymarket, many markets are logically linked — complements, implications, exclusions. Their prices should obey specific relationships. When they don't, a dislocation exists. Studying those gaps at scale is hard without the right tooling.

Structural relationships
Pairs of markets should satisfy rules like A+B≈1 or A≤B. These constraints hold most of the time — but not always, and not uniformly across pairs.
Hard to monitor manually
Even a handful of pairs require continuous tracking of spread, z-score and duration. Dislocations open and close too quickly for manual review.
No native simulation workflow
Polymarket has no built-in way to test whether a convergence idea would have been profitable across historical dislocations — with realistic fees, per pair.
Evidence over intuition
A visible spread is not a tradable opportunity. You need measured data — convergence rate, duration, post-fee PnL — to decide if a dislocation is real or noise.
Core workflow

Pair dislocation detection — what this platform is built around

This is the center of the platform, not a side feature. Related markets on Polymarket should respect structural relationships — complement pairs sum to 1, implication pairs follow logical order, exclusion pairs cap at 1. When they don't, a dislocation exists. The platform measures the gap, validates it with Claude, tracks how long it lasts, and simulates what convergence would have been worth.

Democrat wins WH · Republican wins WH
Complement
Market A 0.52
Market B 0.47
Expected sum ≈ 1.00
Spread |A+B−1|1%
z-score: 0.4 Monitoring
Candidate wins primary · Candidate wins election
Implication
Market A (primary) 0.81
Market B (election) 0.64
Expected: A ≤ B violated
Spread max(0, A−B)17%
z-score: 3.1 · open 42 min Dislocation
Country A wins medal · Country B wins medal
Exclusion
Market A 0.61
Market B 0.44
Expected: A+B ≤ 1 +5% over
Spread max(0, A+B−1)5%
z-score: 1.8 Approaching
Within expected range
Approaching threshold
Dislocation detected

Research module — runs 24/7 in all phases. No capital involved.
Illustrative examples. Live values and thresholds depend on market conditions and configured pairs.

Research workflow

One workflow: observe pairs → simulate dislocations → optionally operate

Everything is organised around the same object — pair dislocations. Real-capital execution is optional and only becomes available after you explicitly decide the accumulated per-pair simulation data justifies it.

01
👁️
Observe
Structured data collection across 300+ markets every 2 minutes (general + sports). Builds a baseline of volatility, liquidity and correlations. AB Monitor tracks all pairs. Claude auto-discovers new structural pairs daily.
≥200 markets · ≥7 days
02
🎮
Simulate
Every dislocation detected by AB Monitor is validated by Claude and simulated with realistic fees (4 legs × 1%). Per-pair metrics accumulate so you can judge which pairs are profitable before committing real capital.
Claude validation · per-pair PnL
03
🚀
Operate (optional)
If you decide the simulation results support real capital, you send "ACTIVAR" via Telegram. Controlled live execution activates on your own Polygon wallet and API keys. Never automatic.
Human approval required
Capabilities

Everything in service of the dislocation workflow

Every capability here exists to detect, validate, simulate or measure pair dislocations. Nothing is a general-purpose market-analysis feature.

📡
Pair monitoring
Tracks structurally related market pairs every 2 minutes. Computes spread and rolling z-score per pair and relation type (complement, implication, exclusion, lag).
🤖
Dislocation validation
Every detected spread anomaly is reviewed by Claude, which judges whether the divergence is genuine or justified by fundamentals before anything is simulated.
🔬
Convergence simulation
Simulates entering and exiting convergence trades with realistic fees (4 legs × 1%). PnL is recorded per event, per pair — no assumptions, just measured outcomes.
🔎
Structural pair discovery
Claude scans active markets daily and proposes new pairs that satisfy a structural relationship. Accepted pairs enter the monitor automatically.
📊
Per-pair reporting
Convergence rate, average duration, post-fee PnL and Sharpe — tracked individually for every pair. Daily briefing summarises active events and recent outcomes.
🔔
Dislocation alerts
Telegram notifications when a dislocation opens, when it converges, and when a new pair is auto-discovered. No generic market noise.
🎯
Execution (optional, OPERATE phase)
After you decide the simulation data supports it, a controlled execution workflow can be enabled for specific pairs. Requires your wallet, your API keys, and your explicit confirmation.
Control

Research discipline, not a sales pitch

Seven fixed limits apply to every decision — simulated or real. They exist to enforce research discipline: simulation control, execution safeguards, bounded exposure. None of them prove profitability.

🔒
15% max
per position
📋
8 max
open positions
📉
5% daily loss
stops all activity for the day
🛑
20% drawdown
pauses everything + alert
💧
Min liquidity
threshold required
⏰
1 day min
to market close
⚖️
25% max
per pair
Dashboard

What you see in the dashboard

A structured view of what the platform is monitoring, validating and simulating across pairs — at every stage of the research workflow.

Overview
Current phase, active dislocations, recent events with Claude validation status, and active monitored pairs at a glance.
Dislocations
Full pair table: live spreads, z-scores, open events with validation status, simulated PnL and convergence rate per pair.
Trades
History of simulated convergence trades — entry/exit spread, duration, post-fee PnL, converged/non-converged outcome.
Reports
Daily briefing focused on dislocation activity: active events, 24h PnL, convergence stats, auto-discovered pairs.
Markets & pairs
Underlying market data — price, volume, liquidity. Admin panel for curating monitored pairs and reviewing auto-discovered ones.
Methodology

The goal is not to automate conviction. The goal is to test whether repeatable structural dislocations exist at all.

The platform does not assume edge exists. It is built to measure — pair by pair, event by event — whether structural dislocations on Polymarket converge often enough, fast enough, and net of fees to be worth anything.

Measurement over belief
No pair is trusted because it looks logical. It is trusted because its measured convergence rate, duration and post-fee PnL say so.
Simulation before capital
Every dislocation runs through the simulator before anyone proposes touching real money. The simulation is the evidence, not a marketing exercise.
Human in the loop
The transition from simulation to live execution is never automatic. You decide per pair, based on accumulated data, when (if ever) real capital is justified.
Transparent metrics
Convergence rate, Sharpe, drawdown, post-fee PnL — visible per pair. If a pair underperforms, you see it clearly, not a smoothed version of it.
Resilient by design
Every external call has retry logic and graceful fallback. API outages are logged and skipped — the research process continues without interruption.
Who built this

A serious experiment, built by one engineer

This is not a hedge fund product, a trading service, or a guarantee of any kind. It is a personal research project built with rigor.

👨‍💻
Alberto Rodríguez
Software Engineer & Data Engineer
Professional with extensive experience in data engineering, backend development and automated information systems. Specialised in data pipelines, heterogeneous source integration and data-driven platforms.

Built this platform to test a specific question: do repeatable structural dislocations exist on Polymarket, and can they be detected, validated and simulated with enough rigour to decide — with evidence, not intuition — whether real capital is justified. No assumption of profitability. The project exists to measure, not to sell.

Follow the research on X

Structured findings from the dislocation research: per-pair convergence stats, notable events, platform updates. No hype, no noise.

@agentpolymarket

Dislocation research · live from Polymarket

Contact

Interested in the beta?

Questions about the research, access requests, or collaboration proposals — send a message.