Architecture
Variance Markets is built as a layered system on top of Polymarket. The protocol never interferes with Polymarket execution, liquidity, or resolution. It only reads public market data, derives variance signals, and exposes those outputs to products and integrations, with settlement anchored to the underlying resolved outcome.
High-Level Layers
Base Layer (Polymarket)
Event creation and outcome definition
Liquidity + trading activity
Market price discovery (implied probability)
Final resolution / settlement of the outcome
Data Layer (Ingestion)
Collects public market time-series data for selected markets
Produces consistent, timestamped snapshots for downstream computation
Computation Layer (Signals + AI)
Extracts raw divergence signals from market dynamics
Detects and classifies belief regimes via the AI layer
Computes regime-conditioned probability estimates and divergence metrics
Normalization Layer (Scoring)
Normalizes signals across markets (time, liquidity, activity)
Produces comparable variance scores and confidence-weighted metrics
Distribution Layer (Outputs)
Exposes standardized variance signals to:
protocol-native products
automation and strategy systems
external integrations (dashboards, bots, analytics)
Settlement Layer (Outcome-Anchored)
Anchors all resolution to the final Polymarket outcome
Uses recorded pre-resolution signals as the only behavioral input
Stops signal computation once the market resolves
Data Flow
Market Selection
The protocol selects a set of active Polymarket markets to monitor (criteria can include liquidity, activity, maturity, and relevance).
Ingestion
For each market, the system streams time-series data:
price p(t)
price velocity dp/dt
volume v(t)
liquidity/depth proxy l(t)
Data is stored as immutable snapshots for reproducibility and auditing.
Signal Extraction
Raw variance signals are derived from observable dynamics:
reaction asymmetry
persistence vs mean reversion
delayed belief updates
regime transitions in behavior
AI Regime Detection
The AI layer classifies each time step into a regime:
R(t) ∈ {r1, r2, …, rk}
Outputs regime labels + confidence scores
Computes regime-conditioned probability estimates p_ri(t)
Normalization & Scoring
Signals are normalized to avoid bias from:
long-duration markets
thin liquidity
high-noise environments
Produces standardized variance scores usable across markets.
Distribution
Outputs are published to consumers through:
internal protocol interfaces
automation/strategy hooks
external APIs / integration endpoints
Outcome-Anchored Settlement
When the underlying market resolves:
signal computation stops
settlement references the resolved outcome
variance-based interactions are settled deterministically using:
final outcome
recorded pre-resolution signals and scores
Trust Assumptions
Polymarket is the source of truth for outcomes and resolution.
Variance Markets assumes access to public, verifiable market data.
Variance outputs must remain:
reproducible
auditable
deterministic given the same inputs and model parameters
The protocol does not rely on private trader data, identity inference, or discretionary settlement logic.
Determinism & Reproducibility
To ensure verifiability:
each signal is computed from recorded snapshots
model versions and parameters are versioned
outputs can be recomputed from the same input dataset
This enables independent verification of:
regime classification
variance score computation
settlement outcomes
Failure Modes
If the underlying market:
is canceled
fails to resolve
resolves as invalid
Variance Markets mirrors that state and applies predefined fallback handling (e.g., cancel/neutral settlement) for any dependent variance-based interactions.
If ingestion data is missing or incomplete:
signals for the affected window are flagged
scoring excludes unreliable segments where required
settlement relies only on recorded pre-resolution signals that meet validity constraints
Last updated