Analytics Labs — Variant Attribution
How MARS uses this
MARS never trusts a single blended figure. Weekly branch probabilities produce a separate EV per branch, weighted by the live profile (45 / 33.75 / 11.25 / 10 by default) into the blend. Reviews read the decomposition first: which branch produced the edge, which one leaked it, and whether the weights still match the evidence.
How it benefits you
You stop repairing the wrong thing. When results dip, the decomposition shows whether the whole system weakened or one branch is dragging - so you fix a branch instead of overhauling a working system, and you spot a single strong branch masking weakness everywhere else.
Branch-level EV decomposition with blend weights. The blended number alone would hide exactly this picture.
The key idea
The seven-lens permission test
From the Branch Variant Selection Matrix: EV lens (does expectancy justify the management style?), Drawdown/Gate lens (gate always wins), SDE lens (is structure improving?), Volatility/ATR lens (clean continuation or chaos?), Structure lens (trending, compressing, rotating?), Execution-Quality lens (do MAE/MFE/capture/fees support it?), and Regime/Variance lens (is the machine stable enough for expansion logic?).
The checkpoint ladder
0.75R and 1.25R test Time-Conservative banking. 1.3R tests whether delaying first protection actually works. 1.4R tests the Exposure-Conservative bank. 1.6R is the trend unlock. 1.8R tests TNP's delayed trail activation. 2R/2.5R/3R+ diagnose right-tail preservation — are aggressive variants preserving upside or destroying expectancy?
Failure modes
03Using Time-Aggressive because the setup 'feels strong' replaces evidence with feeling. Turning Normal into a trend trade after entry is identity drift. Overusing Time-Conservative in clean trends under-monetizes the right tail. TNP aggression must be earned through quota, EV, SDE, gate, and ATR evidence — never through emotional acceleration.
Worked example
The MARS operating loop: execute, capture, score, gate, throttle, deploy - every cycle feeding evidence into the next.
Before you go deeper
Inside this module
Connected inside MARS
Go deeper
Deep dive — 01
Truth, source, explanation, sustainability, context — five layers, each one deep enough for exactly one kind of question.
Read the full brief →
Deep dive — 02
Every variant is a bet on which rungs get reached. The hit-rate battery is where the bet gets graded.
Read the full brief →
Deep dive — 03
Counts, shares, and the vs-Standard ratio — the audit that keeps variant selection honest about itself.
Read the full brief →
Take it further
Attribution across seven lenses is demanding on sample size. The Foundry is where a variant can be examined before it is given real trades to spend.
Open the AlphaRail Foundry Lab ↗One price. Eleven workbooks, three TradingView indicators, and the full manual library — $497.
