Use Case 05 — Branch Variant Comparison
Which management style earns its keep?
Layer 01 — The comparison
Three styles, one evidence standard.
Every variant gets attributed the same way, so preference can't hide behind vibes:
- Trade count and Avg R per variant — volume and quality, separated.
- EV share — which variant actually carries the expectancy.
- Checkpoint hit rates — whether the management logic performs as designed.
- Net-of-fees results — aggression pays more friction; the comparison charges for it.
Layer 02 — The permission matrix
Aggression must be earned, lens by lens.
The seven-lens permission matrix decides when an aggressive variant is allowed to run — evidence across multiple dimensions, not a good month. The doctrine: recent branch mix can flatter any variant; persistence over weeks and months is what changes the active profile.
How MARS uses this
MARS runs the modeled rules - branch probabilities, payoff structure, gate transitions, tier allocation - across 50,000 alternate histories and keeps the percentile bands as the reference envelope. Every review, live equity is plotted against that envelope, and the position is read together with drawdown bands, gate dwell, and tier behavior before any conclusion is drawn.
How it benefits you
You stop grading yourself by feel. Instead of 'I am behind' or 'this month feels slow', you know whether performance sits inside normal variance, is genuinely outperforming, or is drifting under the model - and whether that drift is edge decay or execution drag. It removes both panic below median and false confidence at a lucky P90.
Equity percentile fan across 50,000 simulated paths. Violet is P10-P90, blue is P25-P75, the gold dash is the median, and the green line is live equity drawn against the envelope.
BASELINE
LIVEcurrent governing profile
TIGHTER STOPS
REJECTEDEV cost exceeds drawdown saving
TNP WEIGHT +10
SANDBOXedge up, adverse tail deepens
FEE MODEL B
CANDIDATEfriction saving survives resampling
Worked example
Every proposed rule change - stop policy, branch weights, fee model - is cloned into a scenario profile and resampled against the same evidence. The grid renders the trade-offs, and only profiles whose edge survives without deepening the P10 tail earn candidate status for live promotion.
Four scenario profiles judged side by side: expectancy, adverse-tail cost, and a verdict. Changes graduate through this grid or not at all.
Layer 03 — Modules involved
The attribution stack.
The Variant Attribution Module carries the comparison, CP3's Journal supplies variant-tagged evidence, and Scenario Profiles let a challenger profile be evaluated before promotion.
Reference
The working pipeline
| Step | Instrument | Output |
|---|---|---|
| Define the candidates | Branch Variant Selection Matrix | Explicit variants with their management rules stated |
| Simulate each candidate | Monte Carlo Lab | EV and drawdown distributions per variant, same seed discipline |
| Compare risk-adjusted | RAER / RAPF lenses | Which variant converts risk into return most efficiently |
| Select and record | Decision log | The chosen variant with the evidence that chose it |
The governing idea
Connected inside MARS
This module doesn't work alone.
Go deeper
Operator briefs on this territory.
Deep dive — 01
A variant comparison is only as fair as what was held constant.
Different seed, different period, different sample. Then the winner means nothing.
Read the full brief →
Deep dive — 02
You are not only picking a variant. You are picking who has to run it.
The winning structure has requirements. Check that they are requirements you meet.
Read the full brief →
Deep dive — 03
The variant axes: time and exposure, and the aggression they trade.
Time = when protection starts. Exposure = how much size stays uncapped. Five locked variants per branch, zero improvisation.
Read the full brief →
Take it further
Comparing variants on live trades is the slowest and most expensive method available. The Foundry is the cheap version of the same comparison.
Open the AlphaRail Foundry Lab ↗
Every module ships in the complete MARS package.
One price. Eleven workbooks, three TradingView indicators, and the full manual library — $497.