Edge Spotter: Technical Documentation
Pipeline architecture, edge scoring math, and system invariants
System Overview
Edge Spotter fetches publicly available market data from Kalshi.com and processes it through a signal discovery pipeline. It integrates with the QSE v1.0 Confidence Allocation engine to identify markets where model assessments diverge meaningfully from market-implied positions.
The system is designed around a fail-closed philosophy: when in doubt, suppress rather than publish.
Core Integrity Guarantees
- No lookahead bias. Signals are graded only after market settlement, never retroactively.
- No internal data exposure. Weights, scores, and calibration data are stripped before storage.
- Fail-closed publishing. Uncertain or low-quality candidates are suppressed, never published.
- Server-side enforcement. All tier, eligibility, and quality decisions are resolved on the server.
Data Isolation Guarantees
Mock / Production Isolation
All data is tagged with is_mock. Public surfaces enforce .eq("is_mock", false)via a dedicated customer query layer. Mock data never appears in track records.
No Lookahead Bias
Track record calculations enforce settled_at > published_at. Signals cannot be retroactively graded with information available before publication.
QSE Internals Never Stored
Only final state, confidence, and top 2–3 factor names are persisted. Weights, raw scores, calibration data, and diagnostic traces are stripped before storage via validatePublicFields().
Why Uncertainty Matters
High uncertainty reduces signal strength, preventing overconfident outputs. When the model's Unknown state allocation is high, the uncertainty penalty (λ) shrinks the edge toward zero, making publication less likely.
This is deliberate: the system would rather miss a potential signal than publish one it cannot support with sufficient confidence.
This system prioritizes integrity over volume and clarity over false precision.
Tier-Based Access Control
| Tier | Signal Access | Market IDs | CSV Export |
|---|---|---|---|
| edge_free | Up to every 60 min | Redacted | Blocked |
| edge_pro | Up to every 60 min | Redacted | Blocked |
| edge_power | Up to every 60 min | Visible | Allowed |
The pipeline runs approximately every 60 minutes and publishes any signals that clear the publication rules; a run may publish none. Published signals are visible to every tier immediately, subject only to field redaction. All tier decisions are resolved server-side. Missing or expired tiers fail closed to edge_free. Client-side cannot override entitlements.