Backtests grade the past. Balances grade luck. Expectancy grades the machine.

The cadence test

A grade you receive annually cannot steer anything.

Every candidate measurement can be judged on one practical axis before accuracy is even discussed: how often does it produce a reading you can act on? A backtest is thorough and produces one verdict, historically, about a system that has since changed. An annual return is precise and arrives far too late to influence the year it describes. Both are accurate. Neither is available at the frequency decisions actually get made, and a grade that cannot be consulted when a decision is being taken is not part of the operating loop. The cadence requirement also explains a common frustration: traders who have done thorough historical validation and still feel they are flying blind are not missing rigour, they are missing a reading that updates.

Why the alternatives fail

Each competitor fails on availability, not on correctness.

Set the usual candidates against the cadence requirement and the field thins quickly:

• Backtest — grades a historical system, not the current one. No live reading.
• Account balance — available constantly, and reports luck and skill mixed.
• Win rate — available weekly, and says nothing about payoff or cost.
• Annual return — accurate, arrives once, too late to steer.
• Expectancy in R, net — available weekly, comparable, and about the method.

What continuous requires

Comparability across everything the account does.

For a grade to work weekly it has to be comparable across instruments, position sizes, account balances, and time — otherwise each reading is a fresh quantity rather than a point in a series. Expressing expectancy in R rather than currency is what supplies this: a +0.3R month is the same statement about the method whether the account is five thousand or five hundred thousand, and whether the trades were in one pair or six. Currency-denominated measures fail here quietly, because they change meaning as the account grows while appearing to be the same number.

Net, or it does not count

A gross grade measures a system nobody is trading.

The qualifier in the name's promise is doing real work: expectancy net of friction. A gross figure describes an idealised version of the method operating in a market without spread, commission, or slippage, and the account never experiences that version. Reporting gross would make the grade systematically optimistic in exactly the situations — higher frequency, thinner edges — where the operator most needs it to be blunt, and would hide the entire category of systems that are profitable in theory and a slow leak in practice.

The reporting relationship

Every other number in MARS ultimately reports to this one.

The claim in the page copy is structural rather than rhetorical. Drawdown asks what this expectancy costs to harvest; RAPF asks whether the profit behind it is quality; RAER asks how efficiently it is converted; acceleration asks how it is aging; the benchmark asks how it distributes. Each is a question about the same underlying quantity. That is what makes it the master metric — not that it is the most important number in isolation, but that the other four are all interrogating it from different angles.

The limit

A master metric is still an estimate with a confidence interval.

Being the grade that steers does not make expectancy precise. It is an estimate from a finite sample, and a weekly reading on a small number of trades carries an uncertainty band wide enough to contain most of the interpretations an anxious operator might reach for. This is why the sample count is displayed beside it and why thin windows are flagged rather than averaged in. The metric grades continuously; it does not grade confidently at every cadence, and conflating those two properties is how a weekly reading gets over-trusted. The correct response to a thin window is therefore not to distrust the metric but to widen the window, which is the same discipline the sample-size guards enforce automatically rather than leaving to the operator's patience.