Two weeks ago I wrote here about the payrolls candle and why a breakout system has no business trading it. The argument was simple: a level only means something if someone was defending it, and during a scheduled release nobody is. Wednesday handed me a cleaner test case than I could have designed myself.
What actually happened
On 16 September the FOMC raised the federal funds target by 25 basis points to 3.75–4.00%, unanimously, the first increase since 2023. Spot gold did exactly what the textbook says it should: it sold off into the decision and traded down near $4,310 in the hours afterwards.
Then it did the opposite. Thursday gold added roughly 2.5%, closing near $4,369 after touching $4,380 intraday. Friday it pushed to about $4,383 — a one-week high, and the first weekly gain in four weeks. The rate hike sold gold for a few hours and bought it back for two days.
Why this is not a paradox
The reaction looks contradictory only if you believe the headline is the market. It isn't. Treasury yields spiked on the announcement and then corrected. Crude fell more than 4%, taking some of the inflation premium out with it. Gold ETFs had been taking inflows for eight straight sessions before the meeting — money that was positioned for the hike rather than surprised by it.
So the move down on Wednesday afternoon was not a breakdown. It was inventory clearing in a liquidity vacuum: spreads widened, resting orders were pulled, stops were filled against nobody in particular. The price went through a level, and the level was not taken — because there was nobody on the other side to take it from.
This is the same distinction I made about the payrolls candle on 4 September, when gold fell over 2% from ~$4,477 to ~$4,366 and then closed near $4,420. Different event, identical anatomy. Two of these in two weeks is not a coincidence; it is what scheduled catalysts do.
The rules this produces
A system that trades levels rather than headlines needs the distinction encoded, not admired:
- The level must exist before the catalyst. A high or low printed during the release is not structure. It is the residue of an absence of structure.
- The break has to survive a retest. Price returning to the level and being rejected is the only evidence I accept that ownership changed hands. A move that cannot come back and hold was never a break.
- The stop belongs to the level, not to the candle. If the stop is sized from event-day volatility, it is sized from noise.
- Time is part of the exit. My average holding time is a little over two hours. A position that hasn't resolved is telling me the level wasn't the driver.
The practical consequence of all four is the same: during event hours, the system does nothing. Not because of a news filter that has to fire at the right minute, but because the entry condition simply cannot be satisfied when no level is being defended. In August the account was active on 8 of 21 trading days. The other 13 are not a failure of the system — they are the condition doing its job.
The loss that makes the point better than the wins
On 3 August I shorted gold from around 4,020 and lost €66.81 — 1.67% of the account, and the largest single realised loss in two years of live trading. It was a fade against the prevailing direction at a level that had not been properly taken. The system's own rule was the thing I was thinnest on.
Two days later, on 5 August, the long from 4,120 worked: +3.16% in a little over three hours, into 4,168. Same instrument, same week, comparable position size. The difference between the two trades was not the setup's appearance. It was whether the level had been defended before price got there.
I publish the bad one because the good one proves nothing on its own. Anyone can show the 5 August chart.
Where the record stands
Two years live, verified: 799 trades, +198.84% gain, 80% win rate, profit factor 2.22, maximum drawdown 15.32%, average 4.58% per month, 20 of 25 closed months positive.
One note on numbers, because readers see two figures side by side and reasonably ask which is real: both are. My MQL5 signal page shows +153% because MQL5 measures growth from the day the signal was published (26 February 2026, equity €1,664.59). The full track record shows +198.84% because it measures gain on deposits across the entire history since August 2024. Same account, two starting points.
Looking into the rest of September
No forecast, only environment. Markets are currently pricing roughly a 60% chance of another hike next month, and 16 of 18 Fed policymakers see at least one more this year. The Bank of Japan has been tightening in parallel. That is a calendar dense with scheduled catalysts — which means more candles that look like breakouts and aren't, and more traders reading them as signals.
The levels I care about now are the ones that were built before Fed week, not during it: the low $4,300s that were tested and reclaimed, and the $4,380–4,400 band that capped Friday's move. Whether either gets taken is not something I can know. Whether it gets taken rather than merely broken is something I can measure — and that measurement is the whole system.
You can follow AlphaEdge Smart with the full verified statistics on my profile. Questions about the retest condition, the stop logic, or anything else in this post — ask in the comments, I answer everyone.
Past performance is not indicative of future results. Trading leveraged instruments carries a substantial risk of loss. Copying a signal introduces additional differences in spreads, slippage and execution that can materially affect results.
