Nine windows reached a rail, and the one lean that missed was a 50% call

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 Monday's hour and Thursday'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 and Friday's half hour) because each is rounded to a whole percent. The only red folder USD release of the week, the FOMC minutes, printed at 14:00 ET on Wednesday, after every window on this page had closed; it is marked against the day for the record. Charts this week are 1-minute, so outcome times are read from the mark to the minute.

Morning by morning

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

Monday 17 August

The narrowest box of the week resolved in twenty minutes; the hour's lean stopped five points short

60m Atlas mode, no target hit · 30m abstained and its rail landed

The half hour produced a 7.9 point box, less than half its five-day average and the narrowest of the week. Its read was 43 up against 34, which is 56% of the decided share and short of the 62 the gate wants, so both labels said abstain. Price went up through the 7.9 point target anyway, about twenty minutes after the box closed.

The hour was the opposite shape: 34.8 points, an expansion against its own last five sessions, a normal box against the whole history. The ladder stepped the multiplier to a half and the read came back 46 / 22 / 32, which clears both gates, so Atlas mode fired long with a 17.4 point target.

It did not get there. The morning high sat about five points under the rail and the afternoon drifted away from it, so the window closed in the 32% branch. The abstain resolved and the lean did not, which is a fair reminder of what the gate is and is not: a statement about the split of comparable sessions, not a guarantee about this one.

• 30m box7.9 pt · compressed
• 60m box34.8 pt · expansion (box), normal (table)
• 30m three-way43 / 34 / 24
• 60m three-way46 / 22 / 32
• 30m modeabstain, direction gate
• 60m modeAtlas, long
• Outcome30m up rail ~09:51 · 60m no target
• Durationsagreed

Tuesday 18 August

One minute after the half-hour box closed, gold fell twenty points

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

The half hour closed a 10 point box at 09:30 and the very next minute, the US equity open, printed a single candle that fell roughly twenty points. The read had been 23 / 50 / 27 out of 258 sessions, 50 of the 73% that resolve being 68% of the decided share, Atlas mode fired short, and the target was reached before the label had finished drawing.

The hour contained that drop, so its box was 32.4 points, normal on both rulers, and its read was 41 up against 27 down. That is 60% of the decided share, two points short of the gate, so it abstained. The rail it leaned to was the upper one; the rail price reached, at about 11:14, was the lower one.

Worth being precise about what that is and is not. The gate did not predict the miss; it declined a lean it did not rate highly enough, and the lean happened to be wrong. On Monday the declined lean happened to be right. The gate does not know which week it is in, and neither outcome is evidence about it.

• 30m box10.0 pt · compressed
• 60m box32.4 pt · normal
• 30m three-way23 / 50 / 27
• 60m three-way41 / 27 / 32
• 30m modeAtlas, short
• 60m modeabstain, direction gate
• Outcomedown rail, both windows
• Durationsdisagreed

Wednesday 19 August

The ladder cut both expanded boxes to a quarter, and both long calls paid before 10:20

Both durations Atlas mode ✓🔴 FOMC minutes · 14:00 ET

Gold ran forty points in the half hour after 09:00 with no scheduled release behind it; the day's only red folder event, the FOMC minutes at 14:00 ET, printed long after both windows had resolved. The boxes came in at 40.9 and 43.1 points, both tagged expanded against the whole history, and the ladder cut the multiplier to a quarter on both.

That put the targets 10.2 and 10.8 points out, about the same distance as a compressed box at the full multiple, and the reads were strong: 70 / 27 / 3 on the half hour, where only 3% of comparable sessions fail to resolve, and 63 / 21 / 16 on the hour. Both fired long, and both reached their rail within a few minutes of each other shortly after 10:10.

One comparison is worth making. The half hour's sample, 306 sessions, matches the sample Friday 14 August's half hour drew at a half multiple, where it read 63 / 28 / 9. If that is the same bin, and the count says it is, then at a quarter multiple the same sessions read 70 / 27 / 3: nearer target, fewer unresolved, a slightly stronger lean. The multiplier changes the distance being asked about, not the history being asked.

• 30m box40.9 pt · expanded
• 60m box43.1 pt · expanded
• 30m three-way70 / 27 / 3
• 60m three-way63 / 21 / 16
• 30m target10.2 pt at ×0.25
• 60m target10.8 pt at ×0.25
• Outcomeup rail, both windows
• Durationsagreed

Thursday 20 August

Tuesday's half-hour cell again, and this time the short call was wrong

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

The half hour returned 23 / 50 / 27 out of 258 sessions, which is Tuesday's cell exactly. Tuesday it fired short and paid inside a minute. Thursday it fired short and price went the other way, clearing the upper rail at about 10:36. That is the week's only Atlas mode miss, and it is what a 50% call looks like half the time.

The hour abstained on the other gate. Its direction was 23 against 16, which is 59% of the decided share and short anyway, but the reason the labels were grey is that only 39% of comparable sessions reached a rail at all. This one did, up, at about 10:57, and kept going.

Put the two mornings side by side and the pair is the week's most useful exhibit. The same cell, the same call, a hit and a miss three days apart. Neither day says anything about the cell on its own. Together they say what 50 of 73 means, which is a lean, not a result.

• 30m box14.7 pt · compressed
• 60m box20.8 pt · compressed
• 30m three-way23 / 50 / 27
• 60m three-way23 / 16 / 61
• 30m modeAtlas, short
• 60m modeabstain, resolution gate
• Outcomeup rail, both windows
• Durationsdisagreed

Friday 21 August

Monday's half-hour cell repeated and resolved the same way; the hour leaned wrong and had abstained

Both windows abstained · favoured rail 1 of 2

The half hour drew 43 / 34 / 24 from 301 sessions, Monday's cell. Monday it abstained and resolved up; Friday it abstained and resolved up, at about 10:26. Three repeat cells this week, then: one pair that agreed and one that split, with Monday's hour sitting between them as the one window that reached neither rail.

The hour is the more instructive abstain. Its lean was down, 28 against 13, which is 68% of the decided share and comfortably over the direction gate. It stayed grey because only 41% of comparable sessions reached a rail at all, and the second gate wants 50. Price reached a rail, the upper one, at about 11:52.

So of the five abstained windows this week, all five reached a rail: three to the side the table leaned to, two against. That is roughly what abstained windows should look like, because they are by definition the ones where the table's lean was not strong enough to call.

• 30m box12.3 pt · compressed
• 60m box20.1 pt · compressed
• 30m three-way43 / 34 / 24
• 60m three-way13 / 28 / 59
• 30m modeabstain, direction gate
• 60m modeabstain, resolution gate
• Outcomeup rail, both windows
• Durationsdisagreed

What the week actually showed

• Three repeat cells again, and again they did not all agree. Tuesday and Thursday's half hours were the same 23 / 50 / 27 out of the same 258 sessions; the short call paid on Tuesday inside a minute and was wrong on Thursday. Monday and Friday's half hours were the same 43 / 34 / 24 out of 301, both abstained, both resolved up. Two weeks in, the record now holds six repeat pairs, and three of them split. That is the clearest thing a record this short can show about what a 50% call is.
• The week's only Atlas mode miss was a 50% call, and its only unreached lean was a 46% one. Neither is a surprise and neither is a verdict. The gate fires on the split of the decided sessions, which is why a 46% rail can carry a magnet; the number to read beside it is the no-target-hit share, and on Monday's hour that was 32. The magnet says which way the decided sessions went, not how many sessions decide.
• The ladder did its job on both expanded boxes, and the contrast with Monday is the point. Wednesday's boxes were 41 and 43 points, the multiplier stepped to a quarter, the targets sat 10 and 11 points out, and both long calls paid inside fifteen minutes of each other. Monday's hour was a 35 point box at a half, a 17 point target, and it stopped about five points short. Same indicator, two distances, two answers.
• All five abstains reached a rail, three of them the favoured one. An abstain is not a forecast of a quiet session. It is the indicator declining to promote a lean it does not rate, and the outcomes of the five looked like what they were: a weak lean, sometimes right, sometimes not. Tuesday's hour declined a lean two points short of the gate that turned out wrong; Monday's half hour declined one that turned out right.
• Twenty races over two weeks: seventeen reached a rail, the favoured rail won twelve of those, Atlas mode fired ten times and was right on seven of the nine that resolved. These are published because the record has to accumulate somewhere, and they are far too few to mean anything yet. The place to judge calibration is still the indicator's own realised versus expected rows, which sat between 57% and 82% against a 66% to 77% expectation across this week's captures, with every z between -1.1 and +2.0.