ICT Calls It a “Two-Stage Delivery.” We Tested It on NFP, CPI, PPI and FOMC. It Went 0 for 4.

ICT’s “two-stage delivery,” in his own words

In a lecture built around the October 2026 jobs report, ICT lays out a two-step playbook for trading the release. He says to let the 8:30 a.m. ET number print and run, without touching it; wait 15 to 20 minutes to see what the initial move leaves behind; then look to trade back into the range that move just carved out.

He gives the two steps a name of their own:

In the same breath, he says this is how he treats CPI and PPI mornings too, and — as the next sections show — he separately frames FOMC afternoons with a similar move-first-then-retrace shape. That one idea is why this article checks it four different ways: as a trade on jobs-report and CPI/PPI mornings, and as a price pattern — no trade attached — on the open retracement and on FOMC afternoons.

To ICT’s credit, he hedges this setup twice. He says it doesn’t fire on every release:

And separately, he warns against jumping in before the number prints:

He repeats that same caution by name for CPI and PPI — not a claim that the wait-then-fade mechanism is shaky on those releases, but that guessing which way any of the three will print ahead of time is a loser’s game:

Elsewhere in the same lecture he mentions personally making “just less than $25,000” on the session he is narrating. That is one trade’s result in dollars, not a win rate or an average, so it plays no part in any number below.

What ICT says, and what we tested

Before any numbers, here is the translation from his words to a rule a computer can run, and the comparison each rule is checked against.

That comparison matters more than it might seem. A chart you’re already looking at after the fact can look like it delivered a clean move and a clean fade no matter what actually happened — hindsight makes almost any trade look like it worked. The only fair test is one built to risk the same amount either way: the same stop distance, the same reward distance, just a different direction. A looser comparison that happens to give the fade more room to win than the alternative gets isn’t a comparison at all, it’s a thumb on the scale.

That isn’t a hypothetical risk here. An earlier version of this study’s own comparison trade did exactly that — it gave the alternative trade more reward distance than the fade got, by construction. The NFP section below says what happened once that gap was fixed.

How to read the numbers

Wait, then trade back into the range — the NFP fade

The settings, once: a target at the near edge of the range the news delivered; a 20-minute wait before looking for the trade; when the stop and the target are both touched inside the same one-minute bar, the stop counts; fills happen on touch; no trading costs. The rule only fires if price has actually broken out of its own pre-release range first — 38 of 124 NFP mornings since 2016 never did, so they produce no trade at all.

Across the 86 mornings that fired a trade, the fade averaged a loss of about a fifth of the amount risked per trade (−0.222R), with a plausible range running from about a half-R loss to a small 0.02R profit. That’s the closest any of the four claims tested here came to a real effect — but “closest” still isn’t “confirmed”: the range still touches a profit, and the result doesn’t clear the stricter bar.

Compared against the fairly-built comparison trade — same amount risked, same reward distance, aimed at continuation instead of reversal — the two came out with the exact same win rate. An earlier, less careful version of that comparison trade had given the continuation leg more reward distance than the fade got, which made the fade look like a clear winner by comparison; fixing that gap erased the win-rate edge entirely. The lesson: that old result was about how the comparison trade was built, not about the fade’s own bracket.

A different way of scoring the same trades — by rank instead of average size — does look stronger, but that’s a known side-effect of how many stop-outs tie at exactly the same loss, not a second confirmation of an edge. Dropping the smallest trades, or testing 24 other combinations of entry choice and how long to wait, doesn’t change the story either — if anything, the two worst combinations skip the wait entirely, the opposite of what ICT recommends.

Only 3 of the 11 years since 2016 came out positive on average (2017, 2022 and 2026). In 2025, every trade that fired stopped out for the maximum loss.

A flat, simple trading cost would push the average loss further, to roughly two-fifths of the amount risked per trade at one NQ point round turn. And the same rule, run on an independent second data feed, shows the same loss — if anything, a bit larger.

“Reasonable to anticipate a retracement going into 9:30”

This one is a measurement, not a trade: how much of the 8:30–8:49 ET rip gives back by 9:29, with no stop, target or cost involved. The comparison is every other Friday with no NFP, CPI or PPI release.

The comparison’s own number is the real finding here: an ordinary Friday morning with no NFP, CPI or PPI release already gives back roughly 40 to 50% of its own early rip by 9:30, on average.

Jobs-report mornings give back a bit more on average (50.8% vs. 46.4%), but that gap isn’t detectably different from zero — its plausible range comfortably includes no difference at all.

A simpler yes/no version of the same question — did the morning retrace at all, rather than by how much — does show a gap. But that cut wasn’t the measure decided on ahead of time, and wasn’t checked against the many other ways the same data could be sliced, so it’s a descriptive aside here, not a finding. Dropping a handful of calendar-shifted NFP dates changes nothing, and no single year, nor an up-vs-down split, stands out from the rest.

The same comparison, run on a second, independent data feed, lands on a similarly sized, still-not-detectable gap.

“It’s the same thing” — CPI and PPI

Later in the same lecture, ICT says the wait-then-fade shape he runs on jobs-report mornings is not NFP-specific. Talking about his usual approach to PPI and CPI releases, he says plainly, “it’s the same thing.” As above, he separately cautions that he doesn’t try to call the direction of any of these three releases ahead of time — that caveat is about pre-positioning, not about whether the after-the-fact fade works. This section tests whether treating CPI and PPI as one rule with NFP actually holds up.

Same settings, once: the identical NFP rule and settings above, run on CPI and PPI release mornings. The breakout requirement drops a lot of these mornings before a trade can even fire — over a quarter of CPI mornings (35 of 124) and exactly half of PPI mornings (63 of 126) never break out of their own pre-release range.

Pooled together, the result is essentially flat: an average of about −0.02R across 152 trades, with a plausible range that sits comfortably on both sides of zero.

But the pooled number hides a split. PPI alone is a genuine loss on average, with a plausible range that stays negative throughout. CPI alone leans positive on average, though not clearly so. The two releases disagree in sign — itself an argument against treating them as one rule, whatever each one does on its own.

The same quirk that makes the rank-based check look stronger than it is on NFP mornings (described above) shows up here too, so it isn’t treated as a second confirmation on CPI/PPI either. And just as on NFP mornings, there’s no win-rate edge over the comparison trade here, pooled or split by release.

“At 2:00 there’s a move… at 2:30, usually it’ll retrace”

Like the open-retracement check above, this is a measurement, not a trade: whether the 2:30-to-close stretch reverses back toward where price sat at 2:29 p.m. (within about 2 points counts as neither, a wash) or keeps running past it, against ordinary non-FOMC Wednesdays matched the same way.

It doesn’t show up. Across 84 FOMC afternoons and 465 comparison Wednesdays, the pooled gap in how often the day reverses runs slightly the other way — a few percentage points fewer reversals on FOMC days — with a plausible range that comfortably straddles zero and stretches far enough on the other side that FOMC afternoons could plausibly reverse somewhat more often, too. Either way, there’s no clear evidence that FOMC afternoons are more likely to reverse by the close than an ordinary Wednesday.

The same comparison, run on a second feed, again lands on no real difference, and shortening the window checked doesn’t change the picture either. Splitting out the handful of meetings with no press conference changes nothing meaningful, though that slice has only 12 meetings in the entire history since 2016 — too few to settle anything on its own.

There is one real, unresolved thread. When a FOMC afternoon is scored as running straight through rather than reversing, the move really is bigger than the matched comparison’s biggest moves. But FOMC afternoons are also roughly 2.4 times more volatile than an ordinary Wednesday regardless of how the day is scored — and that generic volatility gap is now about the same size as the “kept running” gap. Plain extra chop is at least as good an explanation as a 2:00-to-2:30 echo.

A related cut, looking only at how often the day keeps running rather than reverses, does show a lead — but only on meetings with a press conference, it wasn’t the rule ICT described, and it wasn’t checked against the other ways the same data could be sliced. That’s a lead for a future, dedicated study, not a result here. Widening or narrowing how “flat” is defined, by up to eight times, doesn’t change the verdict either.

Could our test have missed a real edge?

Before trusting a “no edge” result, it’s worth checking whether the test could even have found an edge if one were really there.

We ran the identical trade-timing machinery with an invented edge planted in one direction, then again with the same edge planted in the opposite direction. The two came out pointing opposite ways, which shows the test isn’t dead or flipped.

The honest limit: at this sample size, neither the planted-edge version nor its mirror image reached the usual bar for “real” on its own. So this check confirms the test is alive and pointed the right way — not that it’s strong enough to reliably catch an edge as small as the one reported here.

The real NFP result also lands comfortably inside the range of outcomes produced by swapping in a random Friday’s price action with the same bracket, under two different ways of doing that swap. A handful of the real trades used in this check (4 of 86) actually fall on shutdown-shifted dates that weren’t Fridays, but were still matched against a Friday-only pool of stand-in days, so this check isn’t a perfectly clean day-of-week match.

A much larger sweep — hundreds of combinations of wait time, target choice, tie rule, fill rule and cost — changes nothing for the jobs-report fade: not one combination clears the bar once you’re honest about how many were tried. The same sweep, run on CPI and PPI, does turn up a few dozen combinations (out of 720) that clear that bar — but every one of them is a different, losing variant of the rule than the one described above, and the main combination used throughout this piece still doesn’t clear it. A separate, much smaller check of how the FOMC measurement is defined — a handful of window and press-conference splits — doesn’t change its result either.

What holds up, and what’s noise

None of the four claims clears the stricter bar required when several are tested together. The NFP fade is the closest call, and even it comes up short.

None of this says that trading ICT’s full approach, judgment included, doesn’t work. It only says that these four literal, mechanical readings of his words don’t show up in the data tested.

What we could not test

• Data for the main market ends September 7, 2026, and the second data feed used for replication ends August 11, 2026 — a different data vendor’s version of the same stock-index family, not a different market, so it checks the data rather than whether the pattern shows up somewhere else entirely. The lecture itself was uploaded October 2, 2026, so nothing here speaks to anything that happened after the data ends.
• One-minute price bars can’t always tell whether the stop or the target was hit first inside the same minute. This already changes the answer on about 1 of the 86 NFP trades even with no trading costs added, and it wasn’t separately checked for the CPI/PPI and FOMC samples.
• Every number here assumes no trading costs or slippage, and the futures contract used is a standard, unadjusted roll from one month to the next (a different roll rule wasn’t tried). A simple flat cost is shown separately for the NFP fade and makes the loss bigger, but it’s an assumption, not a broker quote; the two measurement-only claims (the open retracement and FOMC) have no cost model at all, because they aren’t trades.
• None of this models ICT’s actual discretionary judgment — which prior high he’d pick, what liquidity he judges was left in a move’s wake, or his bias going into the session. These are literal translations of his words, not a claim to reproduce how he trades live.
• Some of the rules tested are researcher-built stand-ins where ICT gave no fully mechanical version: the entry and target construction, the anchor for his fib tool, and the stop placement. Two of his claims — an older daily high used as a fib extension, and his general advice not to pre-position ahead of these releases — have no rule specific enough to test at all, and weren’t attempted.
• The one FOMC comparison that would fully isolate meetings without a press conference has only 12 of them in the entire history since 2016 — that number can’t grow, so this specific comparison can’t be settled with the data this market has produced.
• The check for whether the test machinery itself is sensitive enough is limited at this sample size, as described above. It shows the test isn’t dead, not how reliably it would catch a real edge this small if one existed.

Common questions