A Swing-Trading Workflow
The four-step loop
Every step exists to answer one question, and every step can veto.
| Step | Question | Source |
|---|---|---|
| 1. Radar | Is anything structurally stretched right now? | Two-timeframe σ scan |
| 2. Positioning | Is there fuel behind the move? | Weekly CFTC COT |
| 3. Divergence | Is momentum already disagreeing with price? | RSI on the confirming timeframe |
| 4. Calendar | Is a scheduled event about to overwrite all of it? | Economic calendar + policy rates |
1. The radar fires
A signal means a pair reached |1.2σ| on the daily while 4h and 1h had both turned against it — the full rule is in the radar guide. Treat this as the pair earning fifteen minutes of your attention, nothing more.
2. Check the fuel
Open the COT panel and read the currency on both sides of the pair. You are looking for three things, in this order:
- Direction — the sign of the net position. This is the only column with direction in it.
- Crowding — Idx52. Below 10 or above 90 means a full tank of potential energy.
- Ignition — has Δ, the weekly change, just flipped sign after a long run the other way? This is the difference between stored fuel and a lit fuse.
And one veto: if the trade means selling a currency that speculators are already maximally short, the setup loses its edge regardless of how good the chart looks. Full reasoning in the COT guide.
3. Check divergence
The third layer asks whether momentum has already stopped confirming price. We use the traditional definition rather than a loosened one, because the loose version fires constantly and means nothing:
| Requirement | Setting |
|---|---|
| Pivot basis | Bar highs and lows, not closing prices |
| Pivot confirmation | 3 bars on each side |
| Minimum separation | ≥ 4 bars between the two pivots |
| RSI margin | > 1 point of genuine divergence |
| "Strict" qualifier | The older pivot's RSI below 35 or above 65 |
Which timeframe to check depends on the signal that fired: a daily signal is checked against 1h and 4h; a weekly signal against 4h and 1d. In our testing, a daily-preset signal accompanied by strict 4-hour divergence averaged +0.165% per week (t = 2.6), and a weekly-preset signal with strict daily divergence averaged +0.119% the next day (t = 2.4). Small effects, but consistently positive — which is why divergence adds to the score rather than driving it.
4. Check the calendar
The last check is the cheapest and the most often skipped. A technically immaculate exhaustion setup four hours before a rate decision is a statement about the last few hours, and the next few hours belong to the central bank. The fundamentals board carries the economic calendar, current policy rates and the resulting carry direction — worth a glance if only to notice that holding through Wednesday's close triples the swap charge.
The grade
The site combines these into a single letter so the answer is legible in one second on a phone:
| Grade | Meaning |
|---|---|
| A | Radar, positioning and divergence all agree, with no event risk and no veto. Rare by design. |
| B | The main factors agree; one is missing or mildly against. |
| C | The radar fired but the supporting layers are neutral or conflicting. Most signals are C, and that is the point. |
The distribution is meant to be bottom-heavy. If most signals graded A, the grade would carry no information.
One thing the grade deliberately does not answer is how much to risk. Size is its own decision, derived from stop distance and account risk rather than from conviction — the position size calculator does that arithmetic for forex, stocks and crypto in one place.
Limits worth knowing
- COT is three days stale by construction. The Friday report reflects Tuesday. It is context, never a trigger.
- The effect sizes are small. Tenths of a percent per week with t-statistics near 2 to 2.6 are real but modest, and they are averages across many observations — no individual setup is owed its average. To feel what a 55–61% hit rate actually implies over a long sequence, run it through the risk of ruin calculator: the answer depends far more on sizing than on the hit rate itself.
- Backtests flatter. Everything above was measured with hindsight over a fixed sample. Live results include spread, slippage, swap and the fact that you will not take every signal.
- The radar has no standalone edge. We keep saying this because it is the thing most easily forgotten once signals start arriving on a phone.
Used honestly, the workflow's real contribution is narrower than it looks: it stops you looking at 31 instruments, and it makes the reasons for and against a setup explicit before you act rather than after.