Use Case 03 — EV & Edge Validation
The validation loop
Deployment was justified by an expectancy model. Validation asks, continuously, whether live evidence still supports that model:
The validation loop
Deployment was justified by an expectancy model. Validation asks, continuously, whether live evidence still supports that model:
The core idea
Why continuous
An edge rarely announces its death — it drifts, and P&L variance covers the drift for weeks. Structural stability and drift reads catch expectancy decay while the P&L still looks fine, which is the entire difference between retiring an edge and being retired by it.
Modules involved
CP3's EV Scorecard computes the layers, the Weekly Summary Engine sets the tags, and the SDE's drift and z-score pipelines provide the early-warning read.
Reference
| Step | Instrument | Output |
|---|---|---|
| Estimate checkpoint probabilities | Weekly Scorecard inputs | P(1R) and conditional conversion rates per branch |
| Compute branch expectancy | Branch EV formulas | EV per trade in R, branch by branch |
| Stress the estimate | EV Sensitivity Lab | Decay tolerance — how many points of hit-rate the edge can lose |
| Confirm across thousands of futures | Monte Carlo Lab | Distribution verdict, not a single lucky sample |
Connected inside MARS
Go deeper
Deep dive — 01
The number is not the defendant. The claim that justified deployment is.
Read the full brief →
Deep dive — 02
The blend is green. One branch is carrying it. Both facts are true and only one is visible.
Read the full brief →
Deep dive — 03
The cumulative average is the last to know. Rolling windows are where edge decay first becomes visible.
Read the full brief →
Take it further
Validating an edge means asking how much has to go wrong before it stops paying. The Foundry answers that in an afternoon rather than a quarter.
Open the AlphaRail Foundry Lab ↗One price. Eleven workbooks, three TradingView indicators, and the full manual library — $497.
