Nine of ten windows reached a rail, and the strongest lean of the week went the other way

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. They are the same three buckets named twice, compressed against accumulation, expanded against expansion, normal against normal, so a divergence would be one ruler putting the box in a different bucket from the other, and none of this week's ten did. Mult is an output of that regime, never a setting, and every window here reads auto. The three-way shares sum to 101 on one row, Tuesday's hour, because each is rounded to a whole percent; every other row sums to 100. 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 on four of the five days. Monday was Labor Day: COMEX gold closed at 14:30 ET, the last bar of the day is 14:25, and Monday's hour is scored to that bar because there was no later one. The tick and cross in the outcome column mark whether the favoured rail was the one that went, for calls and abstains alike; only the five Atlas rows count towards the Atlas score.

Morning by morning

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

Monday 7 September

The holiday half hour paid on the 10:45 bar; the narrowest hour on the record stopped 1.7 points short

30m Atlas mode ✓ · 60m abstained, no target hit🔴 US bank holiday · COMEX gold closed 14:30 ET

Labor Day. COMEX gold closed at 14:30 ET, so Monday's last 5-minute bar is 14:25 and not the 16:55 bar every other window on this record is scored to. The hour below could not reach 16:55 because the exchange was shut; that is a calendar fact, not an outcome, and it is the reason Monday's hour is scored short.

The half hour drew 48 / 25 / 27 from 306 sessions, the third time this cell has come up on the gold record. Friday 28 August it fired long and price went up; Monday 31 August it fired long and price went down; here it fired long and price went up. The box was 10 points against a 28.88 point five session average, tagged compressed, so the multiplier stayed at one and the target was the full 10 points. Price then held between 4446.2 and 4457.7 for an hour and a quarter before the 10:45 bar ran to 4465.0 through a rail at 4460.30.

The hour was 13.3 points wide against a 48.5 point average, the narrowest of the twenty-five hours on the gold record, and it read 25 / 17 / 58: a 58 percent unresolved share, the third largest of the week, and both gates failed, 59 percent of the decided share against a 62 gate and 42 percent decided against a 50 gate. It abstained, and nothing resolved. The post-box range was 4446.2 to 4465.2 against rails at 4466.90 and 4427.00, so the upper one was missed by 1.7 points on a day that ended two and a half hours early. There is no score for nearly, and there is no way to know what the missing hours would have done.

• 30m box10.0 pt · compressed
• 60m box13.3 pt · compressed
• 30m three-way48 / 25 / 27
• 60m three-way25 / 17 / 58
• 30m modeAtlas, long
• 60m modeabstain, both gates
• Outcome30m up rail 10:45 · 60m no target
• Durationsagreed

Tuesday 8 September

Both durations leaned down, both lower rails went, and the hour repeated an August cell to the point

30m Atlas mode ✓ · 60m abstained and its favoured rail landed

The half hour fell from about 4457 to 4443 inside the window, a 14.3 point compressed box, and read 23 / 49 / 28 from 215 sessions: 68 percent of the decided share and 72 percent of comparable sessions resolving, both over the gates, so Atlas mode fired short. Price then held between 4433.9 and 4454.4 for two and a half hours, reaching neither rail, before the 12:05 bar went through the lower one at 4429.30.

The hour drew 12 / 28 / 61 from 265 sessions, which is the cell Tuesday 11 August drew, to the point and out of the same sample. Both times the down side was the larger share at 28, both times the decided share was too small to call: 40 percent here against a 50 gate, so it abstained. And both times the lower rail went first. Two draws is two draws; it is recorded, not interpreted.

A 61 percent unresolved share is the largest of the week, level with Friday's hour, and the window resolved anyway, late, on the 14:45 bar, whose low of 4405.6 carried it through the lower rail at 4411.50. The bare tick on the chart is the honest mark for this: the indicator said the down side was more likely than the up side and declined to make a call about either, and the down side went. Nothing about that counts as an Atlas result, and the scorer does not count it as one.

• 30m box14.3 pt · compressed
• 60m box23.2 pt · compressed
• 30m three-way23 / 49 / 28
• 60m three-way12 / 28 / 61
• 30m modeAtlas, short
• 60m modeabstain, decided gate
• Outcome30m down rail 12:05 · 60m down rail 14:45
• Durationsagreed

Wednesday 9 September

The strongest lean of the week, 81 percent up, and the lower rail went first

30m Atlas mode ✗ · 60m abstained and its favoured rail landed

This ties the strongest lean the gold tables have produced. The half hour read 81 / 16 / 3 from 226 sessions: 83 percent of the decided share, 97 percent of comparable sessions resolving, an unresolved branch of three in a hundred. It is the same cell that fired long on Wednesday 2 September and Friday 4 September and paid both times. Atlas mode fired long with an 8.2 point target, the box having expanded to 32.6 points against a 24.26 point average and the ladder having cut the multiplier to a quarter. The 11:00 bar then fell about thirty-seven points, from 4476.8 to 4439.3, stopping just above the rail, and the 11:05 bar took the lower rail at 4434.55 on a low of 4425.0. About fifty points across the two.

An 81 percent cell is not a 100 percent cell. Sixteen sessions in a hundred out of that sample took the lower rail first, and this was one of them. The number on the label is the historical null the indicator carries, not a prediction about Wednesday, and a reader who treats 81 as a promise has misread it. The same 81 / 16 / 3 cell stands at two hits and one cross from three draws inside eight days, which is exactly the kind of run a 16 percent branch produces and exactly the kind of run that means nothing on its own.

The hour disagreed with the half hour, the only date this week where the two durations favoured opposite rails. It read 30 / 40 / 30 from 245 sessions, 57 percent of the decided share against the 62 gate, so it abstained with the down side as the larger share. Ten minutes after the half hour's cross, the 11:15 bar took the hour's lower rail at 4424.60. Same move, two windows, one scored against the indicator and one scored for its lean.

• 30m box32.6 pt · expanded
• 60m box36.2 pt · normal
• 30m three-way81 / 16 / 3
• 60m three-way30 / 40 / 30
• 30m target8.2 pt at ×0.25
• 60m target18.1 pt at ×0.5
• Outcome30m down rail 11:05 · 60m down rail 11:15
• Durationsdisagreed

Thursday 10 September

The half hour's long paid on the first bar after the box; the hour drew a repeat cell and went the other way

30m Atlas mode ✓ · 60m Atlas mode ✗🔴 Core PPI m/m and PPI m/m · 08:30 ET

Core PPI and PPI m/m printed at 08:30 ET, thirty minutes before the box opened. The 09:00 bar is not the release bar and the window contains no release bar at all; the timing is worth stating and nothing more than that.

The half hour read 65 / 26 / 9 from 226 sessions, 72 percent of the decided share and 91 percent resolving, and Atlas mode fired long with a 9.5 point target at a half multiple. The very next bar, 09:30, put its high of 4410.2 through the rail at 4409.25. The capture is timestamped 09:39:59 rather than 09:34:59 because the study still reads "IB building" on the bar the window shuts and only finalises the box on the one after, so the snip carries one extra candle. The mark sits over the 09:30 bar, which is where the hit is.

The hour drew 46 / 22 / 32 from 326 sessions, the largest sample of the week and the cell Monday 17 August drew. Then it fired long and reached neither rail by the close. Here it fired long again with an 18.7 point target, price fell through the whole afternoon, and the 15:00 bar took the lower rail at 4361.90. Two draws of a 46 percent up call, nothing and then the other way. Forty-six is the largest of three shares and is not a majority; the 22 and the 32 add to 54.

• 30m box19.1 pt · normal
• 60m box37.4 pt · normal
• 30m three-way65 / 26 / 9
• 60m three-way46 / 22 / 32
• 30m modeAtlas, long
• 60m modeAtlas, long
• Outcome30m up rail 09:30 · 60m down rail 15:00
• Durationsagreed

Friday 11 September

CPI morning, two abstains, two upward leans, and both lower rails went instead

30m abstained and the other rail landed · 60m abstained and the other rail landed🔴 Core CPI and CPI · 08:30 ET

Core CPI and CPI printed at 08:30 ET, thirty minutes before the box opened, and as on Thursday the 09:00 window contains no release bar. Both of Friday's windows leaned up, both abstained, and both lower rails went.

The two abstains failed on different gates, which is worth spelling out because the chart labels read the same. The half hour was 51 / 42 / 7 from 287 sessions: 93 percent of comparable sessions resolved, comfortably over the 50 gate, but the up side was only 55 percent of the decided share against a 62 gate, so the direction gate stopped it. The hour was 29 / 10 / 61 from 211 sessions: the up side was 74 percent of the decided share, well over the direction gate, but only 39 percent of comparable sessions resolved at all, so the decided gate stopped it. One window was too undecided, the other too unresolved.

Both went down. The half hour's lower rail at 4412.15 on the 10:35 bar and the hour's at 4398.00 on the 12:20 bar, both against the larger share. Neither is an Atlas result and neither is scored as one; they are counted in the 5 of 9 for the favoured rail, where they are both on the wrong side. A 51 / 42 split is close to a coin and the table said so by declining to call it.

• 30m box18.9 pt · normal
• 60m box23.6 pt · compressed
• 30m three-way51 / 42 / 7
• 60m three-way29 / 10 / 61
• 30m modeabstain, direction gate
• 60m modeabstain, decided gate
• Outcome30m down rail 10:35 · 60m down rail 12:20
• Durationsagreed

What the week actually showed

• Nine of the ten windows reached a rail, which ties the best week on the record. The weeks of 17 and 24 August also came in at nine of ten; the week of 31 August managed five. The one window that reached nothing was Monday's hour, and it missed the upper rail by 1.7 points on a session that shut two and a half hours early for Labor Day. All five half hours reached a rail, the latest of them at 12:05. None of that says gold is more resolvable than it was; it says this week resolved.
• Atlas mode fired on five of the ten windows and landed three. Monday's half hour, Tuesday's half hour and Thursday's half hour paid; Wednesday's half hour and Thursday's hour crossed. Both crosses were long calls that met a fall, and the two of them are why the week's favoured rail count is 5 of 9 rather than 7 of 9. The five abstains split two on the direction gate, two on the decided gate and one on both.
• An 81 percent lean went the other way, and that is what an 81 percent lean does sixteen times in a hundred. Wednesday's half hour carried the joint strongest stated share the gold tables have shown, 81 up against 16 down with a 3 percent unresolved branch out of 226 sessions, level with the same cell on 2 and 4 September and with the 17 / 81 / 2 short lean of 25 August. The 11:00 and 11:05 bars then fell about fifty points together and the second of them took the lower rail. The share on the label is a historical frequency the indicator carries, not a forecast of Wednesday, and the whole reason it is printed as a percentage is so a reader can see how often the other side wins. Wednesday was the other side.
• Four cells repeated, and three of the four disagree with themselves. The 48 / 25 / 27 half hour cell has now fired long three times, on 28 August, 31 August and Monday, and gone up, down and up. The 81 / 16 / 3 half hour cell has fired long three times in eight days and gone up, up and down. The 46 / 22 / 32 hour cell fired long on 17 August and Thursday and returned nothing and then a fall. The 12 / 28 / 61 hour cell is the exception: it abstained down on 11 August and on Tuesday and the lower rail went both times. Two or three draws of a cell cannot separate a 46 percent call from a 60 percent one, and this page is not going to pretend otherwise.
• Gold and the index book did not move together this week. On the NQ side of the same five mornings, eight of the ten windows reached neither rail by the close; here, nine of ten reached one. That is two instruments, two window times and ten windows each. It is recorded because both pages are published from the same week, not because it supports any conclusion about either.
• Fifty windows over five weeks: forty reached a rail, the favoured rail won twenty-five of those, Atlas mode fired twenty-nine times and was right on fifteen of the twenty-three that resolved. These are published because the record has to accumulate somewhere, and fifty races remain far too few to mean anything. The place to judge calibration is still the indicator's own realised versus expected rows, which this week ran 68% to 80% against a 66% to 71% expectation on the general hit rate, every z between +0.4 and +1.4, and 69% to 73% against a 72% to 75% expectation on the Atlas rate, every z between -0.6 and +0.1.