NY AM, week commencing Monday 7 September 2026

At a glance

Three scores, because the indicator makes three separable claims, and one event that sits outside all three.

The four share columns are the indicator's own figures, read off the status table at the session open, and they sum to 100 on every row. Gold marks the branch the tables rated highest. London printed below its own average range on four of the five days, Wednesday's 250 points being the exception. Two red folder USD releases fell inside the session this week, both an hour into it: PPI at 08:30 ET on Thursday, forty-five minutes after the low had gone, and CPI at 08:30 ET on Friday, on the bar that took the low. Monday was Labor Day: index futures closed at 13:00 ET, but the 07:30 to 11:30 window ran in full. Times are the 5-minute bar in which the level was crossed.

A branch printed at 0% happened

This is the first NY AM session in the record where price touched neither London level, and it arrived on a table that rated the inside branch at zero. Four things are worth saying about that, and none of them is that the number was right or that the number was wrong.

0% on the status table is a rounded whole number. The four shares are printed as integers and sum to 100, so 0% is what the table prints for any share below half a percent. In a bin of 840 measured sessions that is at most four of them, and possibly none of them. It is not a claim that the outcome cannot occur, and the indicator has no way to say "rare" more precisely than this.

Across the whole published record, roughly one inside session was due. Add up the inside share the table printed on each of the thirty sessions published so far and the total is 0.82 of a session. One has now happened. The count is close to what the tables implied; the day is not. Four of the thirty sessions were rated 10% inside, the highest any of them was rated, and every one of those four took a level. This one was rated 0%. The hub page puts the same thing at the product level: NY AM takes at least one London level in about 49 sessions out of 50, which is a statement that the fiftieth exists. Monday was a fiftieth.

One draw does not move a table measured on hundreds of sessions, in either direction. Six weeks and thirty sessions cannot confirm or refute a share of half a percent, and a page that treated a single inside morning as evidence about the 0% would be making exactly the mistake it warns about everywhere else. What it does show is the shape of the failure when it comes: not a violent move in the unrated direction, but a session that ran 110 points between its own extremes and finished two points and 37.75 points short of the two levels on either side of it.

The word inside does two jobs in this indicator, and only one of them is this week's story. In the Prior sweep column it describes what the previous session did to the one before it, and no row this week carries that value: they read swept high, swept low, swept low, swept low, swept high. In the four-way it means the NY AM window itself stayed between both London levels, which is the branch Monday landed. Monday's prior sweep was Swept high; it is the outcome that was inside. For what it is worth, Atlas Sessions on gold recorded an inside session on the same morning, which is on that product's own week page.

The five NY AM sessions

Left, the session opening at 07:30. Right, the same chart at 11:30 with the outcome latched.

Monday 7 September

Neither London level was touched, and the table rated that branch at 0%

Inside session, the 0% branch · no side taken, so the first sweep scores ✗US bank holiday · index futures closed 13:00 ET

The fifth draw of 10 / 52 / 38 / 0 from 840 sessions, the largest bin in the record. Price opened 13.25 points above the London low with 85% on that side going first, and 52 + 38 = 90% on the low being taken at some point in the morning.

It was not taken. The session low printed 29536 on the 08:20 bar, 2.00 points above the London low at 29534.00. Price then rallied to 29646 on the 09:30 bar, which stopped 37.75 points under the London high, and faded to close at 29577.75. Both levels finished the session unlatched, which is the branch the table printed at 0%.

Because neither side went, there is no first sweep to call, and the scorer has no bucket for that: it records a first-sweep miss by construction rather than a call that went the other way. The modal branch, low only at 52%, did not land either, and neither marginal did, the 90% low nor the 48% high. On the day it read 0%, that is four separate ways of being wrong on one morning, and it is one morning.

It was Labor Day. CME index futures shut at 13:00 ET, so the day was short, but the NY AM window of 07:30 to 11:30 ran in full and nothing about this session was truncated. The tape was thin, which is worth knowing and is not an explanation.

• London range150 pt, below average
• Prior sweepSwept high
• Open vs London EQBelow EQ
• Four-way10 / 52 / 38 / 0
• First sweep▲ 15% · ▼ 85%
• P(high taken)48% → not taken
• P(low taken)90% → not taken
• Session high29646, 37.75 pt short
• Session low29536, 2.00 pt short
• Samplen 840, High

Tuesday 8 September

The high at 08:55, a 260-point fall, and the low at 09:40

First sweep ✓ · both sides, the 18% branch

The fourth draw of 37 / 35 / 18 / 10 from 215 sessions, the only Medium sample of the week. Price opened at 29563.75, which is the midpoint of the London range to the tick, and the study classified the open below EQ. First sweep 53% to the high.

The high went at 08:55 and the session topped at 29686.75 on the 09:30 bar, 37.75 points through it. From there price fell about 260 points in forty minutes, latching the London low at 09:40 and bottoming at 29424.50, 54 points through. The recovery took it back to 29640 by 11:25 to close at 29608.50, inside the London range it had just been outside on both sides.

First sweep correct at 53%, and both marginals landed, the 55% high and the 53% low. The branch was both sides at 18%, the lowest-rated of the three live branches, against a 37% high-only favourite. Three earlier draws of this table went low only, low only and high only; this is its first both sides, and its 10% inside branch has still not appeared.

• London range171 pt, below average
• Prior sweepSwept low
• Open vs London EQBelow EQ
• Four-way37 / 35 / 18 / 10
• First sweep▲ 53% · ▼ 47%
• P(high taken)55% → taken 08:55
• P(low taken)53% → taken 09:40
• Session high29686.75, 37.75 pt through
• Session low29424.50, 54 pt through
• Samplen 215, Medium

Wednesday 9 September

Low only at 45%, and the rally back stopped 20.50 points under the high

First sweep ✓ · low only, the 45% favourite ✓

The fourth draw of 24 / 45 / 25 / 6 from 406 sessions. London had run 250 points, the widest prior session of the week, and price opened 53.25 points above its low with 62% on that side going first.

The low went at 08:05 and the session bottomed at 29332.50, 32.5 points through it. Price then turned and ran about 260 points to 29594 by 10:05, which is 20.50 points under the London high. That is how close the 25% both-sides branch came to landing. The last ninety minutes gave it all back and then some: the 11:20 bar traded at 29358.25, back under the London low, and the session closed at 29390.75.

First sweep correct at 62%, the modal branch landed, and both marginals resolved as stated, the 70% low taken and the 49% high not. This bin has now gone low only twice and both sides twice in four draws, against a table that says 45 and 25.

• London range250 pt, above average
• Prior sweepSwept low
• Open vs London EQBelow EQ
• Four-way24 / 45 / 25 / 6
• First sweep▲ 38% · ▼ 62%
• P(high taken)49% → not taken, by 20.50 pt
• P(low taken)70% → taken 08:05
• Session high29594
• Session low29332.50, 32.5 pt through
• Samplen 406, High

Thursday 10 September

Monday's table four days later, and this time the low went in the fourth bar

First sweep ✓ · low only, the 52% favourite ✓🔴 PPI · 08:30 ET

The sixth draw of 10 / 52 / 38 / 0 from 840 sessions, which is Monday's table on a 166-point London instead of a 150-point one. Price opened 11.25 points above the London low, two points closer than Monday, with the same 85% on it going first.

It went at 07:45, in the fourth bar. The fall carried to 29042.50 by 08:55, 283 points through the level, against a median penetration of 79 points in this bin. The steepest bar of the fall is the 08:30 one, the bar of the PPI print, by which time the level had been taken for forty-five minutes. Price recovered to 29275.25 by 10:35 and closed at 29236.50.

First sweep correct at 85%, low only at 52% landed, the 90% low taken and the 48% high not. Six draws of this bin have now produced four low onlys, one both sides and one inside, against a table that says 52 / 38 / 0, and the 10% high-only branch has never appeared. Monday and Thursday are the same forecast four days apart, and the page above and the page here are what happened to it.

• London range166 pt, below average
• Prior sweepSwept low
• Open vs London EQBelow EQ
• Four-way10 / 52 / 38 / 0
• First sweep▲ 15% · ▼ 85%
• P(high taken)48% → not taken
• P(low taken)90% → taken 07:45
• Session low29042.50, 283 pt through
• Median down penetration79 pt
• Samplen 840, High

Friday 11 September

The high at 08:15, the low on the 08:30 bar, and then 160 points through the high

First sweep ✓ · both sides, the 24% branch🔴 CPI · 08:30 ET

The second draw of 53 / 20 / 24 / 3 from 479 sessions, the table that first appeared a week earlier on the payrolls Friday. Price opened 26.75 points under the London high, above EQ, with 74% on the high going first.

The high went at 08:15. The 08:30 bar, the bar of the CPI print, spiked straight down through the London low to 29204.50 and closed back near where it had opened, latching the low 10.50 points through against a median of 81. Price then ran up to the session high of 29500.75 on the 09:00 bar, 160.75 points above the London high, and gave 166.75 of it back by 10:45, to 29334.00, six points under the London high it had taken two and a half hours earlier. The close at 29485 was back above it.

First sweep correct at 74%, both marginals taken, the 77% high and the 44% low, and the branch was both sides at 24% against a 53% high-only favourite. That is exactly the scoreline this table produced a week ago. Both of its two draws have landed on mornings with an 08:30 release, which is a coincidence of the calendar: the bin knows nothing about the diary.

• London range125 pt, below average
• Prior sweepSwept high
• Open vs London EQAbove EQ
• Four-way53 / 20 / 24 / 3
• First sweep▲ 74% · ▼ 26%
• P(high taken)77% → taken 08:15
• P(low taken)44% → taken 08:30
• Session high29500.75, 160.75 pt through
• Session low29204.50, 10.50 pt through
• Samplen 479, High

The same forecast, more than once

Across the thirty sessions published so far, eight context bins have each come up more than once, and twenty-nine of the thirty sessions sit in one of them. The only session still on its own is Monday 10 August. Same context bin, same lookup, same numbers to the percentage point, on days weeks apart with different London ranges and different price levels. This week every one of the five forecasts was a repeat, which has not happened before.

Two things follow, and they matter more than this week's score.

The forecast is a lookup, not a fit. Nothing about Thursday was used to produce Thursday's numbers. They are the numbers Monday got, and Monday 3 and Thursday 20 August and Monday 31 August and Tuesday 1 September got, because all six sessions landed in the same measured bin. A 150-point London and a 166-point one, four days apart, opened 13.25 and 11.25 points above their own London lows and drew the same table to the percentage point. One of them stayed inside and one of them ran 283 points through the low.

Identical forecasts resolving differently is the system working, not failing. The largest bin in the record, 10 / 52 / 38 / 0 from 840 sessions, has now drawn six times: four low onlys, one both sides and one inside, against a stated 52 / 38 / 0. The 10% high-only branch has not appeared in any of the six. Every bin that has repeated has produced more than one outcome. Six draws cannot tell you whether 52 is the right number, and neither can thirty sessions; what they can show is whether the table keeps returning the same answer for the same context bin, which it does.

What this week actually shows

Five things, from a week with one genuinely unusual session in it and four ordinary ones.

1. The first inside session of the record landed on the one day rated 0%. Monday's low stopped 2.00 points above the London low and its high 37.75 points below the London high, on a table reading 10 / 52 / 38 / 0 from 840 sessions. The honest summary is that the printed zero is a rounded integer, that the thirty published sessions implied 0.82 inside sessions in total, and that one draw at a rated 0% is a single observation and is treated here as one.
2. Monday's first-sweep miss is a miss by construction, not a call that went the other way. The indicator called the low at 85%, and no side was taken at all, so there was nothing to be first. The scorer has no bucket for a session that sweeps nothing and records it against the call, which is why the week reads 4 / 5 rather than 4 / 4. On the four days that swept anything the called side went first every time, at a mean stated 69%. Across the record the first-sweep call stands at twenty-one from thirty.
3. Both branch misses were the same shape: called high only, both sides went. Tuesday and Friday each took the high first as called, then took the low as well, at 18% and 24% respectively against high-only favourites of 37% and 53%. That is a scoring loss on the branch and a scoring gain on the first sweep and on the per-side view at the same time, from one piece of price action. It is worth knowing which of the three claims you are actually using.
4. Six of ten levels taken, against a mean stated 62%. That is the joint lowest stated mean of the six weeks, and Monday supplied two of the four not-takens, one of them a 90%. Across the record, forty-three of sixty levels have been taken against a mean stated 67%, a realised 72%. The per-side view is the least dramatic of the three scores and the one with the most events behind it.
5. Thirty sessions settle nothing, and the indicator's own row is behind for a sixth week. Sixty level events cannot confirm frequencies measured over hundreds. The indicator's realised versus expected row read 51 to 52% against 56% expected across the week, on 263 to 278 setups loaded, which is behind for the sixth week running. The record has a shape, not a verdict, and that row is still the number to keep watching.