Skip to content
Montex Alpha Rail System
$497

Scenario Manager · Promotion Workflow

From idea to production, with a paper trail.

Gates before production

5

Rejected scenarios

KEPT

Rollback

PRE-WRITTEN

How MARS uses this

Promotion is earned in stages, and the final stage is this picture: the candidate's modeled envelope with live shadow-tracking drawn inside it. A change that behaved in resampling must also behave in forward observation before it touches production rules — the overlay is the last gate on the paper trail.

How it benefits you

You stop grading months by feel. A flat quarter inside the band stops causing panic edits, a hot streak at P85 stops justifying size increases, and genuine model breaks get caught by classification instead of by pain.

ABOVE MEDIAN · NORMALTRACKING P60live equity (green) read against the simulated envelope — never in isolation

The final promotion gate: forward-observed performance tracked inside the candidate's own modeled envelope.

Version, then compare

A promoted change makes the previous benchmark unreadable unless it is versioned.

Benchmark assumptions are never modified without versioning the workbook and updating the operating standard alongside it. The reason is mundane and expensive: once the ruler changes without a version number, every prior comparison in the review history silently becomes ambiguous, and nobody can say whether last quarter's below-median month was measured against the same model as this one. Promotion is therefore a documentation event before it is a configuration event.

The rollback clause

Every promotion ships with its own undo condition.

A change graduates carrying a pre-written rollback trigger: the specific live evidence that would prove the scenario wrong, and the exact prior state to restore. Deciding the exit before the entry is the same doctrine trades follow, and it converts a failed promotion from a crisis requiring judgement under pressure into a scheduled, unemotional reversal that was agreed to while everyone was calm.

  • Persistence windows span multiple weeks and months; one flattering sample promotes nothing.
  • Rejected scenarios stay in the library with their verdicts, because negative knowledge is still knowledge.
  • Promotion updates weights or rules explicitly, so production changes are visible diffs rather than silent drift.

Shadow first

Forward observation is the last gate, not the first.

A change that behaved well in resampling has demonstrated one thing: it survives the sample it was tested on. Tracking the candidate forward inside its own modelled envelope asks the harder question, whether it behaves in evidence it has never seen. Only after forward observation stays inside the candidate's own bands does the change touch production rules, and the observation window is defined in advance so it cannot be shortened by an encouraging first fortnight.

Post-promotion QA

04

The first live run after a promotion is treated as a test run.

The checklist that precedes any benchmark use applies with extra weight immediately after a change: confirm live data is complete, confirm period labels match, confirm drawdown sign convention is consistent, confirm target-capped values are understood, confirm locked paths are being interpreted correctly, and confirm tier usage is compared against the gate cap rather than raw frequency. A promotion that was correct in the sandbox can still be mis-read in its first live month, and a mis-read first month is how a good change gets reverted.

Reference

What gets refreshed on promotion, and how often thereafter

Refresh itemFrequencyWhy it matters
Completion probability and lock rateQuarterly or after a major model changeConfirms survival-adjusted compounding is still intact
Hit-week percentilesQuarterlyDetects whether target speed changed
Monthly equity bandsMonthlyThe primary live equity comparison band
Monthly return bandsMonthlyDetects live return drifting under or over the model
Monthly drawdown bandsMonthlyShows whether live drawdown is normal or adverse
Gate dwell and worst gate reachedMonthlyFeeds regime classification and structural review
Tier usage distributionMonthlyChecks whether the throttle over- or under-uses the ladder
Smart exposure compressionMonthlyChecks whether carryover is blocking throughput
Override stress metricsMonthlyPrevents full-tier override from becoming routine
Throttle efficiencyMonthly to quarterlyMeasures conversion of deployed risk into return

Doctrine

Connected inside MARS

This module doesn't work alone.

Connected — 01

Scenario Manager (full page)

Open system page →

Connected — 02

Scenario Discipline

Open system page →

Go deeper

Operator briefs on this territory.

Deep dive — 01

Sandbox → persistence window → documented verdict → promotion.

Four gates between a promising idea and the production rail — and the order is the protection.

Read the full brief →

Deep dive — 02

Every promotion ships with the condition that would undo it.

Decide the exit before the entry — the rule trades follow, applied to the system itself.

Read the full brief →

Deep dive — 03

Resampled evidence and forward evidence are not the same evidence.

Backward-looking evidence can be overfitted. Evidence generated after the prediction cannot.

Read the full brief →

Every module ships in the complete MARS package.

One price. Eleven workbooks, three TradingView indicators, and the full manual library — $497.

Get MARS Pro — $497Explore the Platform