Real World Backtest Data

We have spent four lessons building a scorecard and never once filled it in.

This lesson fills it in.

Four screens, all of them live and public, all running on the same universe of US equities. We put the scorecard from chapters 6 and 7 against each one, read what it says out loud, and then take a single screen from the moment it fires to the moment the order is placed.

Nothing here is a suggestion to trade anything. These are examples of reading, and reading is the skill this whole course is trying to hand you.

What these numbers actually are

Before the figures, the boring paragraph that keeps the rest of the lesson honest.

A screen is backtested per instrument, not per screen. Every stock in the universe gets its own entry history, its own exits, and its own set of numbers. So the single win rate you are about to see printed next to a screen name is an average across many instruments, which is a different object from any one of them. No stock in that list behaves like the average.

The averaging rule matters just as much. An instrument only counts if it took at least ten trades and showed a positive expected value, and each one is weighted by how many trades it took. That is a selection, and you should hold it in your head for the rest of the lesson: these are the instruments where the screen worked, not every instrument it touched. The warning in chapter 5 about the difference between a real edge and a flattering sample applies to us exactly as much as it applies to anyone else.

Every figure below quotes the long side.

Four screens, side by side

Each card links to the live screen page, so everything below can be checked against the current version of the real thing.

Numbers as read on August 4, 2026. Historical backtests are not predictions. Not investment advice.

Four different ideas sit on that board: trend continuation, a breakout out of compression, momentum at a new high, and mean reversion from an extreme. Two timeframes, both of them slow enough to run around a job. If you did chapters 3 and 4, your own taxonomy is already staring back at you.

Now read it properly, because the obvious reading is the wrong one.

Win rate does not rank them

Chapter 6 argued that win rate on its own is close to meaningless. Here are four live screens making that argument better than any formula did.

New 52-Week High wins most often, 52.3% of the time, and it is not the best line in the block. RSI Oversold Extreme wins less often, 49.0%, and returns nearly twice as much per trade. Consistent Uptrend wins least often of all and is still comfortably profitable.

Rank those four by win rate and you get one order. Rank them by average return and you get a different one. Neither order is the answer, because neither number is the strategy. Read them as a pair or you have learnt nothing.

A 44% win rate is not a broken screen

Consistent Uptrend loses more often than it wins. That fact makes most people close the page, which is exactly why it is the most useful line in the block.

Run chapter 6's break-even formula backwards on it. At a 44.0% win rate, the reward you need for every unit of risk is:

Break-even R = (1 − win rate) / win rate
             = 0.560 / 0.440
             = 1.27

So any exit that pays more than 1.27 times the stop distance turns that screen profitable. Now put a real target on it: 2:1, meaning the target sits twice as far from the entry as the stop does. Winners pay +2R, losers cost the stop, which is 1R by definition. The expectancy per trade is what the wins pay, minus what the losses take:

Expectancy = (win rate × reward) − (loss rate × risk)
           = (0.440 × 2R) − (0.560 × 1R)
           = 0.88R − 0.56R
           = 0.32R

Every trade, on average, pays you about a third of whatever you risked on it. A screen that is wrong 56 times out of every 100 has a healthy positive expectancy. Chapter 6 said this was possible. This is what it looks like when it has stopped being a hypothetical.

The catch is the one chapter 6 warned about, and it is human rather than mathematical. A 44% screen produces long stretches of losers as a matter of routine. Most people abandon it during an ordinary run and never find out it was the better thing they owned.

Timeframe changes the units, not the quality

Look at RSI Oversold Extreme again. It returns +8.34% per trade against +4.97% for the 20-Day Consolidation Breakout. That is a 68% bigger number, and it invites an obvious conclusion.

The obvious conclusion is wrong, because the two figures are not measured over the same thing. The weekly screen held its average trade for about 56 days. The daily breakout held its average trade for about 39 days. You are comparing a longer rental of your capital against a shorter one.

Put them on the same footing and most of the gap disappears:

RSI Oversold Extreme     8.34% / 56 days = 0.149% per day
20-Day Consol. Breakout  4.97% / 39 days = 0.128% per day
New 52-Week High         4.54% / 36 days = 0.128% per day
Consistent Uptrend       3.97% / 38 days = 0.104% per day

Sixty-eight percent better becomes about sixteen percent better. Still better, and a far more modest claim than the headline made.

This is not a trick of arithmetic. It is what a longer timeframe is. A weekly screen makes bigger moves because it waits longer for them, and the waiting is paid for out of capital that could have been working somewhere else. Whenever you see a per-trade return, the very next question is how long the trade was open. Without that, the number has no units.

The long short split is a regime fingerprint

The last column is the one nobody reads, and it carries chapter 8 inside it.

92% of the instruments where 20-Day Consolidation Breakout worked, worked on the long side. For RSI Oversold Extreme it is 96%.

That is not a claim that buying beats selling. It is a description of the history the test ran on. The stretch of market these screens were measured against went up, so setups betting on things going up had more to work with. Read those splits as a regime fingerprint, which is to say as a fact about the past rather than a rule about the future.

It also tells you where the fragility sits. A screen at 96/4 has almost never been tested in an environment that punished longs. That is not a reason to discard it. It is a reason to know which way it leans before you find out the expensive way.

One trade, start to finish

Everything so far has been about reading screens. This section is about what a single trade looks like when a screen produces one. We take New 52-Week High on the daily and walk one hypothetical trade through five steps, with round numbers so the arithmetic stays visible. Nothing here is a recommendation. It is the checklist.

Step 1: the screen hands you a list. On the morning this was written, New 52-Week High held 6 stocks. You did not search for them. The list was produced at bar close by the same rule the backtest tested, and your job starts at reading it. Some mornings the list is empty. An empty list is a normal output, not a failure; the 20-Day Consolidation Breakout held nothing at all that same morning.

Step 2: read the stock's own card before its chart. Pick one stock off the list. Before looking at its chart, look at its backtest card, in the order chapters 6 and 7 built:

• Is this evidence? Enough trades to mean something. The screen overall has 648 across 47 stocks; the individual stock's card shows its own count, and chapter 6's ten-trade floor applies to that number.
• Is there an edge? Win rate and average return, read together, the way the scoreboard above taught.
• Can I sit through it? The losing streak and drawdown on that stock's card. These are the numbers that will actually happen to you, and they are the ones most worth a minute of staring.

Only after those three answers do you look at the chart. That ordering is the discipline; everyone does it the other way around.

Step 3: fix both exits before the entry exists. Suppose the stock broke out at $80.00, and suppose the setup is invalid if price falls 5% back below the breakout. That places the stop at $76.00, so the risk is $4.00 per share. In chapter 6's language, $4.00 is your 1R for this trade.

Does a target 1.5 times the risk clear the bar for this screen? Check it against the win rate, same formula as before:

Break-even R at 52.3% win  = 0.477 / 0.523  = 0.91
Chosen target              = 1.5R           (clears 0.91 with room)

Expectancy = (0.523 × 1.5R) − (0.477 × 1R)
           = 0.78R − 0.48R
           = 0.31R

The 1.5R target puts the exit $6.00 above the entry. Both levels exist on paper before any order does:

Entry    $80.00
Stop     $76.00    risk   $4.00/share   (1R)
Target   $86.00    reward $6.00/share   (1.5R)

Step 4: size it, and be willing to walk away. Chapter 10 does sizing properly; here is the short version. A $5,000 account split into 20 equal chunks gives each trade $250 of capital:

Shares     $250 / $80.00    =  3 shares
Risk       3 × $4.00        =  $12.00
Account    $12.00 / $5,000  =  0.24% of the account at risk

Comfortably inside the 1% to 2% ceiling chapter 10 will argue for. And the step everybody skips: if this stock had been trading at $400, the $250 chunk buys nothing, and the honest move is to skip the trade rather than quietly enlarge the chunk for it. Skipping is a legitimate outcome. Breaking the rule once is how people discover their rules were decoration.

Step 5: place the bracket and close the app. Entry, stop and target all go to the broker together, as one bracket order, so the trade can finish either way without you watching it. From this point there is nothing left to decide. That is the entire point of deciding everything in steps 2 through 4, and chapter 11 is about what you do with the rest of your day.

One more habit belongs in the walkthrough: check sample size before returns, every single time. Some screens are so specific they barely ever fire and carry no backtest data at all. A number without a sample behind it is not a number.

What the numbers do not tell you

The scoreboard leaves four things out, and each one changes how you should read it.

Those returns belong to a specific exit. Each figure comes from one exit rule, chosen per instrument, not from the screen on its own. Change the target, change the stop, hold for longer, and you no longer own the number you read. A screen is an entry. Half of the result comes from the other end of the trade.

The average contains no stock. Averaging across 47 instruments produces a number that no single instrument produced. Before you take a trade, read that instrument's own card rather than the screen's headline.

The selection is visible in the figures. Only instruments with ten or more trades and a positive expected value are averaged in. That flatters every number above, which is exactly why the method sits in a paragraph near the top of the page instead of in small print.

Costs are not in there. Commission, spread and slippage all come out of that +4.54% before it reaches your account. On a screen holding trades for 36 days that is a small bite. On a fast screen it can eat the edge whole.

None of this makes the numbers useless. It makes them numbers, with a method attached, which is more than almost anything you will be shown by somebody trying to sell you a signal.

And notice how little of the work above was about choosing. Discovery took a minute, reading the scorecard took five, and then the entire question became how much, where does the stop go, and what happens when six of these go wrong in a row. Picking is the part everyone talks about, and it is the smaller half of the job.

The next lesson is the bigger half.