Nine windows reached a rail again, and Atlas mode landed two of its six calls

The week at a glance

Every number here is transcribed from the capture of that window. Nothing is inferred and nothing is computed after the fact.

Regime is the table's tag, which measures this box against the whole history; the tag printed on the box itself measures it against the last five sessions, so the two legitimately differ, as they do on both of Monday's boxes and on Wednesday's half hour. Mult is an output of that regime, never a setting, and every window here reads auto. The three-way shares sum to 101 on two rows (Monday's half hour and Wednesday's hour) because each is rounded to a whole percent. Charts this week are 5-minute, so an outcome time is the bar in which the rail was crossed and is good to the bar. The session close is the 16:55 ET bar; Wednesday's hour, the one window that reached neither rail, is scored to that bar. Both red folder USD releases are marked against their day: Core PCE and preliminary GDP at 08:30 ET on Wednesday, thirty minutes before the box opened, and Fed Chair Warsh with the payrolls benchmark revision at 10:00 ET on Friday, the bar on which the hour's box closed and the bar after the half hour's rail was reached.

Morning by morning

The 30 minute window on the left, the 60 minute on the right. Click any chart to enlarge.

Monday 24 August

Both compressed boxes abstained and both resolved down; only the hour had leaned that way

Both windows abstained · favoured rail 1 of 2

The half hour drew 43 / 34 / 24 from 301 sessions, the cell that Monday 17 and Friday 21 August both drew and both resolved up. This time it went down. Price spent the late morning about sixteen points under the upper rail, turned at 11:00, and the 11:45 bar fell through the lower rail. The direction gate had declined the lean at 56% of the decided share; the third draw of this cell is the first to go against it.

The hour drew 13 / 28 / 59 from 258 sessions, which is Friday 21 August's cell. That one leaned down, abstained on the resolution floor, and resolved up. This one leaned down, abstained on the same floor, and resolved down at about 13:20 on a wick. Two draws, one each way, and the label was grey both times because only 41% of comparable sessions reach a rail at all.

Both boxes were tagged compressed by the table and normal against their own last five sessions, so the multiplier stayed at one and the targets were the box widths, 16.6 and 25.4 points. Neither window fired. Both reached a rail, one with the larger share and one against it, which is roughly what two abstains should look like.

• 30m box16.6 pt · compressed
• 60m box25.4 pt · compressed
• 30m three-way43 / 34 / 24
• 60m three-way13 / 28 / 59
• 30m modeabstain, direction gate
• 60m modeabstain, resolution gate
• Outcomedown rail, both windows
• Durationsdisagreed

Tuesday 25 August

An 81% short call and a 44% short call, and gold went up through both upper rails

Both durations Atlas mode ✗

Gold fell about thirty-three points in the first half hour with nothing on the calendar, and the table read the shape as 17 / 81 / 2 out of 215 sessions: 81 of the 98% that resolve, the strongest lean on the gold record at the time. The ladder cut the multiplier to a quarter, the target sat 8.3 points below the box, and Atlas mode fired short. Price bottomed within twenty minutes of the box closing, a few points above the target, and went up.

The hour contained the fall, so its box was 38 points, normal against the whole history and an expansion against its own last five days. It read 24 / 44 / 32 from 262 sessions, 65% of the decided share, and fired short with a 19 point target. The upper rail went first, at about 11:40, ten minutes after the half hour's.

Two magnet-and-cross marks on one morning. The 81% call is the one to sit with: it is the highest share this indicator has printed on gold in three weeks and it was wrong. Nothing about that is a surprise in the sense that matters, because 17% of 215 sessions is not a small number, but it is the kind of miss the record has to carry in public if the hits are going to mean anything.

• 30m box33.3 pt · expanded
• 60m box38.0 pt · normal
• 30m three-way17 / 81 / 2
• 60m three-way24 / 44 / 32
• 30m target8.3 pt at ×0.25
• 60m target19.0 pt at ×0.5
• Outcomeup rail, both windows
• Durationsagreed

Wednesday 26 August

PCE before the open: the half hour's abstain resolved in twenty minutes, the hour's long call reached nothing

60m Atlas mode, no target hit · 30m abstained and its rail landed🔴 Core PCE + Prelim GDP · 08:30 ET

Core PCE and the preliminary GDP print landed at 08:30 ET, thirty minutes before the box opened, and gold spent that half hour falling. The 09:00 half hour then formed a 24 point box at the bottom of it, expanded by the table's ruler, so the multiplier went to a quarter and the target sat six points out. The read was 55 / 42 / 3 from 293 sessions: 97% of comparable sessions resolve, but 55 is only 57% of the decided share, so both labels abstained. Price went up through the rail on the 09:50 bar.

The hour was a 34.3 point normal box and read 54 / 12 / 35 from 211 sessions. That is 82% of the decided share, comfortably over the gate, and Atlas mode fired long with a 17.2 point target. The number beside it is the 35: more than a third of comparable sessions reach neither rail, and this one joined them. Price fell to 4637.9 after the box, a little over two points above the lower rail, and ranged between 4637.9 and 4688.5 to the 16:55 bar.

So the window that abstained resolved and the window that fired did not, which is the same shape as Monday 17 August. The gate fires on the split of the decided sessions; it does not say how many sessions decide, and 35% is a large share to have on the other side of the call.

• 30m box24.0 pt · expanded
• 60m box34.3 pt · normal
• 30m three-way55 / 42 / 3
• 60m three-way54 / 12 / 35
• 30m modeabstain, direction gate
• 60m modeAtlas, long
• Outcome30m up rail ~09:50 · 60m no target
• Durationsagreed

Thursday 27 August

Tuesday's hour cell again, the same short call, the same wrong answer

60m Atlas mode ✗ · 30m abstained, favoured rail missed

The half hour drew 45 / 31 / 24 from 261 sessions, the cell that opened the gold record on Monday 10 and Tuesday 11 August. It abstained all three times, at 59% of the decided share, and has now resolved up once and down twice. Here the 09:30 bar fell about fifteen points and the 09:45 bar's wick touched the lower rail.

The hour drew 24 / 44 / 32 from 262 sessions, which is Tuesday's cell to the percentage point. Tuesday it fired short with a 19 point target and price went up through the upper rail at about 11:40. Thursday it fired short with a 15 point target and price went up through the upper rail at about 11:50. Same cell, same call, same answer, two days apart.

That makes three magnet-and-cross marks in a row on this page, Tuesday's pair and this one, and it is the first gold week in which Atlas mode has missed more than it has landed. A 44% call that is wrong twice is not evidence about the cell any more than a 50% call that was right on Tuesday 18 August and wrong on Thursday 20 August was. It is what the number means.

• 30m box13.6 pt · compressed
• 60m box30.0 pt · normal
• 30m three-way45 / 31 / 24
• 60m three-way24 / 44 / 32
• 30m modeabstain, direction gate
• 60m modeAtlas, short
• Outcome30m down rail ~09:45 · 60m up rail ~11:50
• Durationsdisagreed

Friday 28 August

Both Atlas calls paid, the hour's on the first bar after the box, the half hour's on a spike that reversed inside the bar

Both durations Atlas mode ✓🔴 Warsh at Jackson Hole + payrolls revision · 10:00 ET

The half hour read 48 / 25 / 27 from 306 sessions, 65% of the decided share, and Atlas mode fired long with an 11.9 point target. Price rose out of the box and the 09:55 bar spiked through the upper rail to about 4681. The same bar then fell roughly 36 points from that high, and the 10:00 release was Fed Chair Warsh at Jackson Hole with the preliminary payrolls benchmark revision. The rail was reached and the mark printed; what the bar did after that is on the chart too.

The hour's box closed at 10:00 at 35.8 points, normal by the table, and read 13 / 53 / 34 from 203 sessions. That is the cell Monday 10 and Thursday 13 August drew, when the short call was wrong once and right once. Here Atlas mode fired short with a 17.9 point target and the first bar after the box fell through the lower rail.

The snip for the hour is one bar later than the others on this page. The study only finalises the box on the bar after it closes, so a hit on the first post-box bar cannot be captured until the bar after that. It changes nothing about what is scored and it is stated here so the capture rule stays honest.

• 30m box11.9 pt · compressed
• 60m box35.8 pt · normal
• 30m three-way48 / 25 / 27
• 60m three-way13 / 53 / 34
• 30m modeAtlas, long
• 60m modeAtlas, short
• Outcome30m up rail ~09:55 · 60m down rail ~10:00
• Durationsdisagreed

What the week actually showed

• Atlas mode went two of its six calls, and three of the misses were magnet-and-cross marks in a row. Tuesday's half hour fired short at 81 / 17 out of 215 sessions with a quarter-multiple target 8.3 points away, and price went up. Tuesday's hour and Thursday's hour were the same 24 / 44 / 32 cell out of 262 sessions, both fired short, and both went up. Friday's two calls landed. It is the first gold week on which Atlas mode landed fewer than it missed, and it is published as such.
• Four repeat cells, and the record on repeats is now mixed on purpose. Monday's half hour was the 43 / 34 / 24 cell from Monday 17 and Friday 21 August, which had resolved up twice and this time went down. Monday's hour was Friday 21 August's 13 / 28 / 59, which went up then and down now. Thursday's half hour was the 45 / 31 / 24 cell from 10 and 11 August, up then down then down. Friday's hour was the 13 / 53 / 34 cell from 10 and 13 August, wrong then right then right. Across three weeks the repeat draws have gone the same way as their predecessor about half the time, which is what a lean in the forties and fifties is.
• The strongest share on the gold record at the time was wrong. 81% of 215 sessions is the highest directional share the indicator has printed on gold in three weeks, and the 17% branch is the one that happened. That is not a defect and it is not a defence; 17% of comparable sessions is about one in six, and one in six is what showed up. The number to hold beside it is that the week's other four favoured-rail misses were two 44% calls and two abstains at 43% and 45%.
• Both windows on Friday paid, and the half hour's hit needs its whole bar described. The 09:55 bar spiked through the upper rail and then fell about 36 points from that high before it closed, into the 10:00 Warsh appearance and the payrolls benchmark revision. The rail was reached and the tick printed, so it is scored as a hit. Anyone reading a hit as a fill should look at that bar first.
• Thirty races over three weeks: twenty-six reached a rail, the favoured rail won sixteen of those, Atlas mode fired sixteen times and was right on nine of the fourteen that resolved. These are published because the record has to accumulate somewhere, and they remain far too few to mean anything. The place to judge calibration is still the indicator's own realised versus expected rows, which sat between 67% and 87% against a 66% to 76% expectation across this week's captures, with every z between -0.2 and +1.4.