Four first sweeps from five again, and both sides kept landing on ratings of 13 to 29%

The week at a glance

Every number in this table is transcribed from the close capture of that morning. Nothing is inferred and nothing is computed after the fact.

Four-way reads high only / low only / both / inside and always sums to 100. No cell carried the ~ era flag this week. High and Low carry the marginal probability that side is taken at all, which is that side's exclusive share plus the both-sides share, not the number printed on the level itself. The only red folder USD release of the week, the FOMC minutes, printed at 14:00 ET on Wednesday, two and a half hours after the session closed; it is marked against the day for the record.

Morning by morning

Open capture on the left, close capture on the right. Click any chart to enlarge.

Monday 17 August

The low first, as called, and then the session went and took the high as well

First sweep ✓ · modal branch ✗ · both sides, rated 29%

Eurasia printed 21.1 points, below average and inside its own predecessor, and handed US AM an open a few points above its low. The tables put 63% on the low going first and it went at 08:25, on a dip to 4435 that was the session low.

From there gold climbed for ninety minutes, back through the whole Eurasia range, and took the high at 09:55. Both sides at 29% was the second-rated branch; low only at 41% was the favourite and did not land, so the morning scores on the first axis and not the second.

Axis 3 gives it two from two: the low at 70% went and the high at 53% went. A pair of levels twenty-one points apart is a pair that is hard to avoid in a session that travels forty-five, and the marginals said as much.

• Eurasia range21.1 pt, below average
• Prior sweepInside
• Open vs Eurasia EQBelow EQ
• Four-way24 / 41 / 29 / 6
• First sweep▲ 37% · ▼ 63%
• P(high taken)53% → taken 09:55
• P(low taken)70% → taken 08:25
• Samplen 718, High

Tuesday 18 August

An 18-point Eurasia, and the 09:30 candle took both levels ten minutes apart

First sweep ✗ · modal branch ✗ · both sides, rated 27%

The narrowest prior session either gold week has produced, 18.7 points, which put both levels within ten points of the open. The tables leaned to the high anyway: 65% on it going first and 45% on high only. For ninety minutes nothing happened at all.

Then the 09:30 bar, the US equity open, dropped straight through the low and came back; the high went ten minutes later at 09:40 on the rebound, and the session then fell fifty points to 4407 by the close. First sweep wrong, branch wrong, and both marginals resolved as taken, 72% on the high and 51% on the low.

This is the week's double miss and it is worth stating plainly: a 65% call went the other way, on the tightest setup the record has seen. It is also a fair illustration of why a narrow prior session is a poor place to read direction from. When both levels are a few points away, which one a spike takes first is closer to a coin than the 65 suggests, and the per-side view, which only asks whether they go, is the one that held.

• Eurasia range18.7 pt, below average
• Prior sweepSwept low
• Open vs Eurasia EQAbove EQ
• Four-way45 / 24 / 27 / 4
• First sweep▲ 65% · ▼ 35%
• P(high taken)72% → taken 09:40
• P(low taken)51% → taken 09:30
• Samplen 729, High

Wednesday 19 August

The same cell as Tuesday 11 August, the same call, and a 125-point run behind it

First sweep ✓ · modal branch ✓ · high only, as rated🔴 FOMC minutes · 14:00 ET

71 / 11 / 14 / 4, first sweep 87 / 13, out of 603 sessions: the table Tuesday 11 August drew, to the percentage point. That morning the high went at 08:05 and nothing else did. This morning the high went at 08:30 and nothing else did, and gold then ran about 125 points to 4555, the largest US AM move either gold week has shown.

The FOMC minutes printed at 14:00 ET, two and a half hours after the session closed, and had nothing to do with any of it. The day is marked because the calendar is marked on every page, not because the release touched the session.

Both marginals resolved as stated: the high at 85% went, the low at 25% did not. Two draws from one cell, two identical outcomes, which tells you exactly as little as the two-draws-two-outcomes cases elsewhere on these pages. What it does tell you is that the forecast is a lookup. Nothing about the 19th was used to produce the 19th's numbers.

• Eurasia range40.8 pt, below average
• Prior sweepSwept high
• Open vs Eurasia EQAbove EQ
• Four-way71 / 11 / 14 / 4
• First sweep▲ 87% · ▼ 13%
• P(high taken)85% → taken 08:30
• P(low taken)25% → not taken
• Samplen 603, High

Thursday 20 August

Monday 10 August's cell, the low in the opening bar, and then the 24% high went too

First sweep ✓ · modal branch ✗ · both sides, rated 13%

11 / 69 / 13 / 7 out of 562 sessions is the cell Monday 10 August drew, when the low went at 08:20 and the high was never approached. This time the low went in the opening bar, which is the easy half of the forecast, and the session then turned and climbed seventy-five points, through the high at 10:35.

Both sides at 13% is the second-smallest share on the table, and it landed. Low only at 69% was the favourite and did not. The first-sweep call survives intact at 87%, because it only asks which side goes first, and the low did.

Per side, one from two: the low at 82% went, as it should, and the high at 24% went, which is the part the tables did not expect. Same cell as the 10th, opposite branch, and the pair together is worth more than either alone: it is what a 69 / 13 split looks like when you get to see it twice.

• Eurasia range34.6 pt, below average
• Prior sweepSwept low
• Open vs Eurasia EQBelow EQ
• Four-way11 / 69 / 13 / 7
• First sweep▲ 13% · ▼ 87%
• P(high taken)24% → taken 10:35
• P(low taken)82% → taken 08:00
• Samplen 562, High

Friday 21 August

A 97% call that took three and a half hours, and an inside session rated 26% that missed by five minutes

First sweep ✓ · modal branch ✓ · high only, as rated

The widest Eurasia range either gold week has produced, 76.8 points, well above average. A wide prior session pushes both levels away and makes containment likelier, and the table said so: inside at 26%, the highest inside share the gold record has shown. Against that, the high sat nine points above the open with 97% on it going first.

The session spent three hours below the open, dipping to 4622 in the first hour and grinding back, and took the high at 11:25, five minutes before the close. Five more minutes inside and the 26% branch would have landed instead of the 70% one.

The most confident call of the week, and the one that came closest to not happening. Both marginals resolved as stated, the high at 72% taken and the low at 4% not, and the nine-point distance at the open is, as usual, most of the explanation for the 97.

• Eurasia range76.8 pt, above average
• Prior sweepSwept high
• Open vs Eurasia EQAbove EQ
• Four-way70 / 2 / 2 / 26
• First sweep▲ 97% · ▼ 3%
• P(high taken)72% → taken 11:25
• P(low taken)4% → not taken
• Samplen 741, High

What the week actually showed

• Both sides landed three times in five, on ratings of 29, 27 and 13%. Across the ten gold sessions now published the both-sides branch has landed four times against a mean rating of about 13%. Ten sessions is nothing, and a 13% branch landing four times in ten is the kind of run small samples produce. It is also exactly the number the running record exists to keep an eye on, and we will keep saying so until it either closes or does not.
• Four first sweeps from five again, eight from ten over two weeks, and the miss was the narrowest setup on record. Tuesday's Eurasia was 18.7 points, both levels within ten points of the open, and a 65% call on the high went the other way when one bar at 09:30 spiked through the low. When both levels are that close, which one a spike takes first is a weaker question than the 65 makes it look.
• Levels went more often than stated for the second week running. Eight of ten against a mean stated 54%, after six of ten against 48% last week: fourteen of twenty against 51% in total. On twenty events that is about one and a half standard errors, which is not evidence of anything yet and is recorded because it belongs in the record.
• Two repeat cells, both from w/c 10 August, one agreed and one did not. Wednesday drew Tuesday 11 August's table and resolved the same way, high only. Thursday drew Monday 10 August's table and resolved the other way, both sides where the 10th had been low only. Same lookups, different mornings; the forecast is a table, not a fit.
• Nothing stayed inside, again, and Friday missed it by five minutes. Inside was rated 26% on Friday's 76.8 point Eurasia, the highest inside share the gold record has shown, and the high went at 11:25 with the session closing at 11:30. Across ten gold sessions the inside branch has averaged roughly 12% and landed zero times, which at this size is unremarkable in both directions.