Methodology

How EdgeVisor turns market noise into a usable thesis

EdgeVisor is a research system, not a blind signal feed. The goal is not to predict everything. The goal is to identify where the narrowest pipe in the workflow is: information quality, price quality, or execution quality, and only press when the bottleneck is acceptable.

1. Ingestion and normalization

We ingest live prediction markets, normalize contract metadata, and attach market structure context such as price, spread, depth, time-to-close, and category. That keeps the system anchored in observable state rather than story-first speculation.

2. The 16 virtual analysts

Each market passes through 16 independent analytical lenses. Every analyst produces a confidence score and a directional signal. The Multiplicative Weight Update (MWU) adjuster learns which analysts are reliable per category and adjusts weights accordingly.

Core Signal Analysts

  • Base-Rate — checks plausibility by category historical base rates
  • Liquidity — evaluates whether depth supports reliable pricing
  • Momentum — detects 20-bar price trends and acceleration
  • Favorite-Longshot — corrects crowd bias at price extremes
  • Anomaly — flags unusual price or volume deviations

External Intelligence

  • Smart Money — whale wallet tracking via Falcon and on-chain
  • News Sentiment — monitors RSS feeds for event-relevant news
  • NLP Sentiment — semantic scoring of news content via NLP
  • Preddy Deviation — compares with Preddy.trade consensus
  • Metaculus Consensus — cross-references Metaculus forecasts
  • Crowd Aggregation — weighted average of public prediction sources
  • Wallet Behavior — on-chain wallet pattern analysis

Structural Edge Analysts (Arbitrage v2)

All 16 analysts run in parallel on every ingestion cycle (every ~30 seconds). Individual weights are tracked per market category and adjusted by resolved outcomes through the MWU learning loop.

3. Research summary

EdgeVisor compresses the 16-analyst layer into a human-readable explanation: likely direction, confidence notes, supportive evidence, counter-evidence, and explicit warnings. If the evidence is weak, the output may remain informational rather than actionable.

4. Decision rules and sizing

The engine gates on liquidity, timing, evidence quality, and category-specific caveats before any tier logic runs. When a market clears the gate, the decision tree assigns a trade class such as arbitrage, aggressive, standard, fade, or bond. Sizing is still separate from the thesis. On the math side, the educational calculator uses f* = (p - c) / (1 - c) for a YES contract under full-Kelly assumptions.

5. Learning loop

Resolved outcomes feed back into the online adjuster. The model learns which signal families helped and which ones drifted. In physics terms: the system is damped, not free-running. Equal-weight baseline stays available as a stable fallback when learned weights should not dominate yet.

6. Trust boundaries

  • No hidden human analyst layer behind the scenes.
  • No claim that +EV on paper equals executable alpha after slippage.
  • No anonymous public preview writes into track record.
  • No silent fallbacks when mandatory market data is missing.

Input Data

  • Sources: Polymarket CLOB, Gamma API, Preddy consensus, Falcon smart-money signals, whale trackers, news feeds.
  • Signals: price, volume 24h, spread, depth, momentum, cross-venue consensus, on-chain whale activity, sentiment.
  • Constraints: minimum volume threshold, liquidity filters, category-specific rules, time-to-close gates.

Scoring & Confidence

  • Confidence: composite score from 16 virtual analysts, weighted by MWU online learning.
  • Uncertainty: explicit uncertainty bands — system reports what it does not know.
  • Update cadence: every 30 seconds for prices, every 5 minutes for signals, weekly for weight adjustments.

Validation & Audit Trail

Disclosure

Primary references

Next step

Use the method on a real market

The calculator teaches the math. The public Tool shows how the thesis looks on a live Polymarket link. The dashboard shows the full pick workflow.