Morning holds, afternoon pays.

Drops 01 and 02 were kills. This one is the strongest row in the desk’s own trading book — and it ships with the sentence most research never prints: what it has been validated by, and what it hasn’t.

The whole thing fits in one line. Morning holds, afternoon pays. If NQ has spent the morning without ever trading 0.30% below where it stood at 09:00 ET, being long from 11:00 to the close has paid. If the morning did trade through that line, the same 11:00 long pays nothing.

The rule

• When: 11:00 ET, buy at the open of the 11:00 bar.
• Only if: price has not traded 0.30% below the 09:00 open at any time since 09:00. The condition uses only the 09:00–10:59 lows, so it is fully known at 10:59:59.
• Stop: 0.30% below the 11:00 entry — its own stop, not the morning’s.
• Target: none. Time exit, flat at 16:59 ET.

The result

Win rate 0.47, stopped out on 45% of trades, max drawdown −10R. R is P&L divided by the stop distance, so NQ’s doubling since 2021 cannot flatter the recent years. By year:

Positive every full year, 2022 included — the one bear year in the window.

The baseline (the part that makes it a finding)

An 11:00 long on NQ has drift behind it whether or not you add a condition, so the honest opponent is not zero. It is the identical trade on the other days. The covariate here is the morning’s own path:

The lift is positive in both halves. That is the test the desk’s third judge asks for — a matched baseline, split in two, no sign flip.

The same 11:00 long on days the morning traded through: −0.039R. On days it held: +0.297R.

A plateau, not a cell

A real effect should survive nudging its parameters. Vary the survival width and the stop:

Inside the surviving mornings, where price sits in the 09:00–11:00 range at 11:00 does not decide it: lower half of the range +0.512R, middle +0.154R, top 15% +0.319R. The simplest cousin — price merely above the 09:00 open at 11:00, no path condition — pays +0.227R, t=2.84. The phenomenon is a morning that holds, and it is what actually makes the desk’s 09:00 drift trade work.

The mirror that fails

Every survivor here has to have a dead twin, or it is a drift artifact. Three of them:

• Short the same mornings. Shorting at 11:00 on survivor days: −0.224R, t=−3.95, negative in both halves.
• Re-enter after a morning stop-out. The 11:00 long on days that already traded through 0.30%: −0.039R. The complement of this play is not a second chance at it. Never re-enter.
• Down-mornings do not continue. Price at the bottom of the range at 11:00 → short: −0.193R, pre −0.381 / post +0.007. Same finding the Three Opens map showed at 10:00: downside continuation in this index dies by lunch.

Same phenomenon, two other names

Two siblings in the same book measure the same thing from different angles. Take one, not both.

day_long is the plain 09:00 drift trade — the same rule started at every other clock hour averages +0.012R, so it beats its own baseline by +0.084R, but t=1.61 is below the desk’s significance gate; it runs as an alert and is promoted only by its forward log. trend_day_long looks like a second edge until you simulate the book: the trades it adds beyond day_long_add — mornings that did trade 0.30% below the 09:00 open but were back at the highs by 11:00 — come out at −0.045R (n=71). All of its standalone +0.197R lives inside day_long_add days. Its short mirror fails too (−0.195R, t=−2.12).

Exit management, tested so you don’t have to

Winners take their heat early and pay late. Winning trades’ worst adverse excursion sits at 0.15 / 0.28 / 0.47% (p50 / p75 / p90) and lands about 30 minutes in; their best excursion (p50 0.54%) lands about 289 minutes in, on a 359-minute hold, and gives back 0.13% at the clock. What that implies, and what happened when we tested it:

• Stop = the winners’ p75 heat. 0.25% +0.273R, 0.30% +0.297R, 0.40% +0.241R — a plateau. Tighter than 0.25% raises R per trade but kills 40–60% of the eventual winners; wider than 0.50% halves the R.
• Target = the clock. Fixed targets lose in both halves: 1R +0.106R, 2R +0.187R, the clock +0.297R. 4R is roughly neutral, which is why the bracket carries 4R as a disaster cap and nothing tighter.
• Trailing stops do not beat the clock out of sample. The best in-sample trail (2.0R) posts +0.115R post-2023 against +0.242R for the plain clock exit. Break-even-after-1R: +0.258R vs +0.297R. The tightened stop catches the normal mid-session dip of the winners that go on to pay at the close.

Does it travel?

The same battery on ES (2023-02 → 2026-08, split 2025-01-01, 0.75-pt cost): +0.095R, t=1.18, n=533; lift over through-days +0.170R (pre +0.146 / post +0.196), both halves; the 0.2% survival variant lifts +0.248R, both halves. ES is the weaker cousin — less drift, less range — but the morning-holds → afternoon-pays structure replicates in sign. It is a replication witness, not a second book.

What it has and hasn’t been through

Two more sentences of honesty. First, the sizing reality: a block-bootstrap Monte Carlo of the desk’s whole five-play book (this play is one of them) says even one MNQ at the 0.30% stop — roughly $180 of risk — hits a $3,000 trailing drawdown in about 85% of simulated years; the book needs a drawdown budget of 50–70R of its own risk unit. This is a low-Sharpe drift trade, not a scalping machine. Second, the window is a bull half-decade with one bear year in it. 2022 paid, which is more than most long-only rules can say, but a five-year sample of an index long is still a five-year sample of an index long.

How the desk uses it

Not as a signal, and not sized from this page. The 09:00 open minus 0.30% is drawn on the chart as an awareness line: while it holds, the afternoon has historically belonged to the buyers, and fighting that with shorts is fighting a t of −3.95. Once it breaks, the number says the second half of the day is a coin flip with costs, and the desk stops looking for a long. Context for how the day is shaped, decided by 11:00, from one price you already have.