Why Journal Trades at All? Because Net P&L Cannot Name the Strategy That Is Losing

This is the first post in a short series on the question underneath our entire product: why log trades at all. We are starting here rather than with features because the objection is a good one, it is currently doing the rounds again, and most of the replies to it are bad. "All professionals journal" is unfalsifiable and, in the form people usually mean it, untrue. So this article makes one specific argument instead, with our own numbers attached, including the ones that do not flatter us.

The Objection, Stated Properly

The case against journaling arrives in three forms, and they are not equally strong.

The first is that the balance is the scoreboard. If the number goes up you are doing something right, and if it goes down you are not, so the reporting layer is decoration. The second is that journaling is a diary, not work: screenshots, mood ratings, and a paragraph about discipline, filed somewhere and never read again. The third is that a mechanical system does not need a journal, because the rules are fixed and the historical testing already told you what to expect.

We concede the second one entirely. A log that is never queried is a hobby. If what you have is a folder of chart screenshots and some notes about feeling impatient, the people calling that a waste of time are correct, and adding software to it does not change the verdict. We would also concede a piece of the third: if the system genuinely is fixed, mechanical, and singular, and you have already validated it out of sample with walk-forward analysis, then the incremental value of a manual log is small. That is a narrower case than the people making the argument usually occupy.

The first one is where the argument breaks, and it breaks on arithmetic rather than on psychology.

A Sum Cannot Be Decomposed After the Fact

Your balance is an aggregate. Aggregation is a one-way operation: given the parts you can always compute the total, but given the total you can never recover the parts. That is not a limitation of your broker's reporting, it is what a sum is.

For a trader running one setup on one instrument at one size, this costs nothing. The total and the part are the same object, and the balance answers every question you could reasonably ask. Almost nobody trades that way. The moment you have a primary setup and a secondary one, or a breakout approach you took from a course and a mean-reversion approach you drifted into, or a size you use when confident and a size you use when you are not, the balance stops being a scoreboard for a strategy and becomes a scoreboard for a portfolio of strategies whose membership you never wrote down.

And a portfolio of strategies can print a perfectly unremarkable number while containing two violently opposed results. That is not a hypothetical, so here is ours.

What This Looked Like on Our Own Scoreboard

On August 27, 2026, our founder pulled up the Strategy Performance screen on a live stream and went through his own forward-tested strategies on camera. These are strategies defined inside SignalDeck and scored on live signals as they fire each morning, rather than a hand-kept discretionary log, but the mechanic is identical to the one that applies to your trades: every result carries a label saying which strategy produced it, so performance can be asked per strategy instead of per account. The figures below are read from those cards.

Read from SignalDeck's Strategy Performance screen on August 27, 2026, and shown in full on the stream. The third row reports net P&L as a percentage rather than a dollar figure, which is a presentation inconsistency on our side, not a different kind of measurement.

The first row is the one that matters. High-Tight Flag Breakout was the flagship: the pattern the whole approach was built around, the one worth naming a strategy after. Over 343 trades it lost $11,516 with a profit factor of 0.85, meaning it returned 85 cents of gross profit for every dollar of gross loss. His summary on the stream was that it "has not done well, and there is no way to argue with 343 trades."

The maximum drawdown is worth a second look: at $69,057 it is roughly six times the size of the final net loss. The strategy did not bleed quietly down to minus eleven thousand. It went a very long way underwater and came most of the way back, and a trader watching only a period-end figure would have experienced that path without ever having a number for it. Whether you could have survived that sequence with real size on is exactly the question Monte Carlo simulation exists to ask, and it needs the trade sequence to ask it.

Now put the first two rows together, which is what an account balance does automatically. Minus $11,516 and plus $9,748 net to minus $1,768. A quiet stretch. Slightly disappointing, nothing alarming, the sort of number that produces a resolution to be more disciplined and no change to what you actually trade. The blended figure is not wrong. It is just the one presentation of these results that contains none of the information.

The third row makes the same point from the other direction. The Bollinger Band Fade was not part of the plan and was not being watched; it turned up in the list at +71.4% over 30 trades with an expectancy of $524 per trade. His reaction on camera was "it's all of a sudden taken off. That's interesting." You cannot notice a strategy you are not measuring, and a balance has no mechanism for surfacing one.

So the direction of the surprise is the actual finding here, and it is the reason this article exists. The strategy he believed in was the loser. The strategy he was ignoring was the winner. Conviction and results pointed in opposite directions, and the only instrument capable of detecting that was a record that tagged each result with the strategy that produced it.

The Winner in That Table Is Not Proven Either

It would be easy to end the section there, and it would be a misuse of the data. Attribution tells you which strategy produced which result. It does not tell you that the result will persist, and the three rows above have wildly different claims to being believed.

343 trades at a profit factor of 0.85 is close to a verdict. A losing sequence that long is difficult to attribute to variance, and the correct response is to change something. 22 trades at a profit factor of 2.00 is not a verdict, it is a hint. A run of 22 results is comfortably inside the range that a strategy with no edge produces by accident, and it is worth remembering that a small sample of winners is precisely what a lucky streak looks like from the inside. The asymmetry is worth stating plainly: it takes far fewer trades to establish that something is broken than to establish that something works, because losing consistently is harder to do by chance than winning briefly.

The third row deserves the most scepticism despite having the best-looking numbers. A profit factor of 36.91 comes from an average win of $1,243 against an average loss of $25.76, over 30 trades that were all long, with an average hold of 33 days. That is not a description of a strategy with a huge edge. It is a description of a payoff profile whose losses have not happened yet: a rule set that exits losers almost immediately has, so far, never been caught in the gap-through-your-stop scenario that defines its real risk, and 30 long-only trades have not yet met a market that goes down. We are not saying it is bad. We are saying that 30 trades cannot tell you, and a number like 36.91 is a signal to check the sample rather than to allocate.

One honest note about our own product while we are here: SignalDeck computes these metrics on whatever sample exists and does not currently suppress or flag them below a minimum trade count, so a Sharpe ratio calculated on three trades will render as confidently as one calculated on three hundred. Read the trade count column first. That guard is a known gap on our side rather than a design choice, and until it ships the discipline has to come from you.

For the general version of this problem, our post on why win rate lies covers which statistics stabilise quickly and which ones do not, and SQN is explicitly a sample-size-aware alternative to reading raw returns.

The Four Fields That Make Attribution Possible

None of the above requires a journaling practice in the diary sense. It requires four fields per trade, and the rest is optional.

• A strategy label, applied at entry. The load-bearing field. Without it there is nothing to group by and every other number in this article is uncomputable.
• Planned risk at entry — your intended stop distance expressed in money. This is what converts every outcome into an R-multiple, so that results are comparable across instruments and position sizes instead of being dominated by whichever trade happened to be biggest.
• Entry and exit, with timestamps. Available from your broker, and the one part of this a statement genuinely does give you.
• The realised result. Also from the broker.

Two of those four come free from an import. The two that do not are the two that carry all the analytical weight, which is a reasonable summary of why broker data alone does not answer these questions. That distinction is worth a post of its own and will get one later in this series.

The timing clause in the first bullet is not pedantry. A label applied after the outcome is known is not data, it is a story. Tag a trade before you know how it ends and you have a genuine classification. Tag it afterwards and a losing trade quietly becomes "not really my setup, I forced that one," while winners are generously admitted to whichever strategy you currently like. Do that consistently and you will build a beautifully organised dataset showing that all your setups work and only your discipline fails, which is both flattering and useless. Entry-time tagging is the entire difference between a log that can correct you and a log that agrees with you.

Everything else — screenshots, emotional state, market context, a note on what you were thinking — is a genuine addition for some traders and dead weight for others. It is not the part under discussion here, and it is not what makes a log analysable.

Where the Sceptics Are Right

Three cases where the argument against journaling holds, stated without qualification because a page that cannot name its own exceptions is advertising.

You genuinely run one strategy. One setup, one instrument class, one sizing rule, no discretionary overrides. Attribution is trivial because your account is your strategy, and the balance already reports it. Your remaining questions are about risk and decay rather than about which bucket the money came from.

You are under about 20 trades. There is nothing to attribute yet. Log them, because you cannot retroactively create the record later, but do not compute anything from them and do not make decisions off a per-strategy breakdown of a handful of results.

The logging is displacing the trading. If the record has become the ritual — an hour of formatting, a screenshot library, notes nobody rereads — then the sceptics are describing you accurately, and the fix is to cut the practice down to the four fields rather than to defend it. Our earlier post on whether a profitable trader needs a journal at all works through that decision in more detail, including a ten-minute test you can run on a broker statement without adopting anything.

And the largest concession, which no amount of tooling changes: a log does not create an edge. It cannot make a losing strategy profitable, it will not stop you re-entering something the data told you to abandon, and there is no credible study we can cite showing that journaling improves returns by some specific percentage. Anyone quoting you such a figure made it up. What a log does is narrower and still worth having — it makes claims about your own trading falsifiable, and the number that says a strategy has lost money over 343 trades is a number you can act on precisely because you could not have argued with it.

From Attribution to an Actual Decision

Attribution earns its keep only when it changes an allocation. The rough decision frame we use, with the caveat that these thresholds are conventions rather than statistical law:

The last two rows have posts of their own: how to know when your strategy is dying and diagnosing backtest-to-live divergence. Both of them presuppose exactly the thing this article is arguing for, which is a record where results are grouped by the decision that produced them.

What happened after the August 27 scoreboard is the honest version of the payoff. Nothing dramatic: no epiphany, no rebuild. The strategies that were passing became the input to a signal pipeline, and the flagship stopped being treated as the flagship. That is what this kind of measurement actually buys — not a better trader, just an allocation that follows the results instead of the conviction.

How SignalDeck Handles Per-Strategy Attribution

Everything above is achievable in a spreadsheet, and we would rather say so than pretend otherwise. A strategy column, a planned-risk column, and a pivot table will get you the first table in this article. What a tool changes is the friction, and friction is what determines whether the tag is still being applied in month four.

• Strategy tagging on the trade, applied when the trade is created rather than backfilled once the outcome is known.
• Strategy Performance — the screen in the table above. Expectancy, profit factor, Sharpe, maximum drawdown and trade count per strategy rather than per account.
• The automated forward-testing scanner (Pro, $30/mo), which scores a strategy you have defined but have not committed capital to on live signals, alongside the ones you have traded. This is how the Bollinger Band Fade row exists at all.
• R-multiples and expectancy computed from planned risk, so a strategy traded at two different sizes is still one comparable series. Basic R-value analytics are on the Free tier; the fuller layer (SQN, Kelly, Monte Carlo drawdown) is Pro.
• Monte Carlo (Pro, $30/mo) — 1,000-path resampling of your logged sequence, which is the correct next step after a per-strategy number looks good and before you size up on it.

Two limits worth knowing before you sign up. The sample-size guard described earlier is not built yet, so small-sample metrics display without warning. And whether your trades arrive automatically depends on your platform: MT4/MT5 live sync is Elite ($50/mo), several platforms are CSV import only, and our platform support guide lists exactly which is which. Check it against your own setup rather than assuming. Worth knowing before you plan around it: the Free tier carries unlimited trades and basic R-value analytics but caps you at 3 strategies and 10 tags, which is a real constraint on exactly the kind of attribution described here — unlimited strategies and tags are Pro. Everything is unlocked at no cost during beta.