Nine modules of method. This one is the balance sheet: everything the method has actually produced, including the parts that are still question marks.
Forty-eight preset versions have been filed. Eighteen are rejected outright, one is retired, eight are candidates, and eight are running with real money. Of those eight, four carry the site's strongest label and four are explicitly flagged as something weaker. That is the whole estate.
One mechanism edge
The Tokyo fix is the only thing here that passes every window we can test it on.
At 09:55 Japan time, banks settle customer orders at a published rate. Japanese importers are structural buyers of foreign currency, so the flow into that fix has a direction that is set before the day starts, on dates the calendar fixes in advance. The strategy fades the move that flow produces, holds for twenty-five minutes, and goes home.
Five pairs, tuned on 2019–2026 and validated on 2010–2018 — a window that spans a different rate regime, a different volatility regime and the yen's entire round trip:
pair 2019-2026 2010-2018 (out of sample) per trade
USDJPY PF 2.10 538 trades PF 1.31 464 trades $0.23
EURJPY PF 2.16 538 trades PF 1.36 401 trades $0.23
GBPJPY PF 1.99 480 trades PF 1.34 157 trades $0.26
AUDJPY PF 1.56 503 trades PF 1.30 210 trades $0.13
NZDJPY PF 1.21 481 trades PF 1.04 164 trades $0.045The first four are proven and live. The fifth is not, and the reason it is not is module 6's rule 9: a profit factor above one earned at four and a half cents a trade is below the spread it has to pay, and 164 out-of-sample trades producing $2.44 in total is noise that happened to land on the right side.
Why we call it a mechanism rather than a pattern
The distinction module 6 drew between a mechanism edge and a regime edge is not a feeling about how much we like a strategy. It is a claim that has to be supported, and here the support is the last column of that table.
Look at the per-trade gradient: roughly 23 to 26 cents on USDJPY, EURJPY and GBPJPY, 13 cents on AUDJPY, four and a half on NZDJPY. That ordering is not something we tuned for. It follows the size of Japan's actual trade relationship with each currency — the flow is biggest where the importing is biggest, and it thins out to nothing on the pair with the least of it.
A pattern has no reason to be ordered by anything outside the price data. A mechanism does. The gradient is the closest thing we have to evidence that we are looking at a participant rather than a coincidence, because a curve-fit would have no reason to arrange itself by trade weight.
It also predicts its own boundary. NZDJPY is where the flow becomes too small to pay for the spread, and that is exactly where the strategy stops working — which is a far more comfortable place to find a failure than in the middle of the range.
Four instances that are explicitly weaker
The London range-revert strategy is live on four pairs, and its label says regime rather than proven for a specific published reason: it passed 2023–2026 and failed 2019–2022, mildly, with profit factors between 0.88 and 0.94.
Module 5 covered what that buys it — minimum lot only, a mandatory kill rule, no sizing up on backtest evidence. What it does not buy is an assumption that the edge is real in the way the fix appears to be. We hold four live instances of something that worked for four years and did not work for the four before that, and we are running them as a forward test rather than as a conclusion.
That is a defensible position. It is not the same as a good one, and the labels are there so nobody has to guess which is which.
What the kill switch cannot do
Module 5 presented the rolling-profit-factor switch as the layer that notices when a regime ends. It is worth closing the loop honestly, because we tested it and it does less than the description suggests.
Run over 2019–2022 — the window the range-revert presets failed — with the switch on versus off:
USDJPY 145 trades PF 0.93 -$6.7 (switch off: -$14)
EURJPY 173 trades PF 0.86 -$19 (switch off: -$22)
EURUSD 194 trades PF 0.90 -$12 (switch off: -$9)
GBPJPY 183 trades PF 0.84 -$32
GBPUSD 249 trades PF 0.78 -$52It fired between four and fifteen times per pair across those four years — roughly quarterly — and never once on 2025–2026, which is the correct behaviour in both directions. But look at the size of the help. It halved the loss on USDJPY, shaved a few dollars on EURJPY, and on EURUSD it made the result worse, because a twenty-day pause followed by a ten-trade probation puts you back into the same regime that caused the pause.
The conclusion we wrote down, and the one that matters if you build something similar: it trims rather than rescues — a fire alarm, not a fire door. It tells you the regime has turned. It does not save you from the regime turning, and it must not be sold as protection.
That is the kind of finding that is easy to leave out of a course, because module 5 reads better without it.
The denominator
Eighteen rejected versions, each with its numbers and its reason on the site. Behind them sit entire strategies that no longer exist: the trend entry across two timeframes and two sessions, the pullback variant, the breakout engine, four flavours of daily-timeframe gating, and a fade strategy shelved as a documented candidate.
And the research lines with no engine at all — liquidity sweeps, calendar effects, time-series momentum, scheduled US releases, the London month-end fix, the Tokyo option cut, the ECB reference rate — each rejected, each with a note saying what would have to change before it is worth testing again.
Against all of that: one mechanism edge on four pairs, and four regime instances running at a hundredth of a lot.
That ratio is the honest yield of the process, and it is the single most useful number in this course. If your own hit rate looks dramatically better, the most likely explanation is not that you are better at this. It is that you are not publishing your denominator.
What we do not know
The list that a results page usually omits.
Whether any of it works live. The current fleet has one closed trade. Everything above is backtest evidence, and module 7 was explicit that a backtest is the optimistic case — built on our data, our cost assumptions, and an account that never runs out of margin. The first thirty trades will answer whether fills, spreads and margin behave as modelled. They will not answer whether the edge is real.
Whether the drawdown figures are complete. They are computed on closed trades, so they do not count how far a position ran against us before it closed. The published number is a floor, not a maximum, and fixing that needs data the ingest side does not record.
Whether the trade-weight story is true. The per-trade gradient is consistent with importer flow, which is the best explanation we have — but consistency is not proof, and module 9 was about exactly this trap. We tested and rejected the sweep explanation for the London dip; we have not done equivalent work to eliminate alternatives here. It remains our best story rather than a finding.
Whether the regime is already over. The range-revert presets were validated to September 2026. Regimes do not announce themselves, and the layer built to detect this one trims rather than rescues.
Whether the account can carry the system. The first live fix day filled one of three charts because the others were refused for margin. That is a funding constraint, not a strategy question, but it determines how much of this runs at all.
What this course was actually about
Ten modules, and the MQL5 was never the hard part. Reading a closed bar, splitting strategy from engine, sizing a position, gating a trade — that is a few hundred lines, and the language is not difficult.
The hard part is everything built around the code to stop you fooling yourself: an out-of-sample flag the caller cannot set, a harness that refuses to file a report from the wrong EA, a gate that gained two rules the day results slipped past it, counters that say which of seventeen reasons stopped a trade, sizing that refuses rather than rounds up, and a record published from its first trade rather than its thirtieth.
None of that makes a strategy work. All of it makes the difference between knowing whether one does and believing that one does — and a system you merely believe in is indistinguishable from a system that is about to cost you money.
Everything in this course is checkable. The code is in the repository, every preset and every backtest window is in the catalogue or at /api/presets, and the live record — all one trade of it — is on the performance page, updating as trades close.
Check it. That is the entire point of publishing it.