Six chart patterns, measured
M17 measured the bull flag end to end and it did not survive. This page does the same to five more: the rising wedge, the falling wedge, the bear flag, head and shoulders, and its inverse. Same method, same controls, same standard of proof — and the results are reported whatever they say.
First, the other half of this lane: readings in ranges
“Today's RSI 58.33 … It's been close to forty recently though, so the range matters … Everything cannot be very fixed thresholds … not only if it touches forty.”Alan, 23 September
Tap any RSI cell on the Hub and the panel now answers in ranges, not on one line. It says which band the reading sits in, how often that band happens, what followed a visit to it — and it pairs where the reading is now with the lowest it has been lately, which is the shape you described.
Your case, answered from SPY's own history
SPY's daily RSI stands at 58.31 (55 to 60). In the last 60 days the lowest it went was 38.63 (35 to 40, on 2026-07-29). Days that looked like that — in the 55 to 60 band now, having been down in the 35 to 40 band within 60 days — happened 213 times in the 5,780 days measured (3.7% of them). After those days SPY was higher 20 days later 61% of the time, middle move +0.90%, against 66% and +1.35% from any day at all. At 60 days: 65% and +2.38% against 71% and +3.28%.
Read that carefully: the shape you asked about has been followed by slightly worse than an ordinary day, not better. The 20-day version of the same question (lowest 41.76, 40 to 45) says much the same: 64% up, +0.85%. It is a small difference on a few hundred days and it is not a signal. What it does do is stop the reading being described as unusual when it is not.
Every band
| band | how often | visits a year | 20 days later | 60 days later |
|---|---|---|---|---|
| under 30 | 1.6% of days | 2.2 | 74% up · +3.31% | 78% up · +7.56% |
| 30 to 35 | 3.2% of days | 5.3 | 68% up · +2.85% | 78% up · +5.24% |
| 35 to 40 | 5.8% of days | 9.8 | 72% up · +2.47% | 75% up · +4.89% |
| 40 to 45 | 9.8% of days | 16.0 | 66% up · +1.98% | 70% up · +4.03% |
| 45 to 50 | 12.0% of days | 19.8 | 64% up · +1.47% | 70% up · +3.23% |
| 50 to 55 | 14.2% of days | 21.6 | 66% up · +1.39% | 67% up · +3.04% |
| 55 to 60 · where it is now | 16.8% of days | 22.6 | 64% up · +1.18% | 69% up · +3.28% |
| 60 to 65 | 17.0% of days | 20.9 | 63% up · +0.98% | 70% up · +3.08% |
| 65 to 70 | 12.5% of days | 14.7 | 67% up · +1.09% | 70% up · +2.39% |
| 70 and over | 7.1% of days | 5.6 | 70% up · +1.35% | 70% up · +2.49% |
Near misses, counted apart from touches
A turn that came within 3 points of a level and stopped there is counted separately from one that reached it. A turn is only known 5 days after it happens — the days either side are what make it a turn — so what followed is measured from that later day. Measuring from the turn itself would be reading the future, and it inflates the numbers by a lot: at 40, the same count measured from the turn said 97% up at five days, against 61% measured honestly.
| level | near misses | 20 days after | touches at a turn | 20 days after |
|---|---|---|---|---|
| within 3 of 30, from above | 23 | 61% up · +1.29% | 35 | 71% up · +2.57% |
| within 3 of 40, from above | 38 | 76% up · +2.06% | 132 | 68% up · +2.12% |
| within 3 of 50, from above | 42 | 59% up · +1.11% | 245 | 66% up · +1.47% |
| within 3 of 50, from below | 19 | 79% up · +4.20% | 346 | 63% up · +1.16% |
| within 3 of 60, from above | 17 | 59% up · +0.67% | 355 | 65% up · +1.48% |
| within 3 of 60, from below | 35 | 59% up · +0.37% | 250 | 63% up · +0.99% |
| within 3 of 70, from below | 57 | 61% up · +0.84% | 87 | 68% up · +1.38% |
All of it is counted by tools/how-unusual/build.mjs from finished daily bars and stored in data/how-unusual.json;
the reading itself is recomputed live when the panel opens, so the panel is never older than the last finished day.
All six at a glance
| pattern | claims | signals | middle move, 20 days | an ordinary day | drift removed | verdict |
|---|---|---|---|---|---|---|
| The rising wedge | a fall | 1,739 | +0.74% | +0.95% | -0.26% | does not survive |
| The falling wedge | a rise | 962 | +1.01% | +0.95% | +0.36% | does not survive |
| The bull flag | a rise | 100 | +0.01% | +0.95% | -0.62% | too few |
| The bear flag | a fall | 51 | +1.58% | +0.95% | +0.68% | too few |
| Head and shoulders | a fall | 247 | +1.57% | +0.95% | +0.60% | does not survive |
| Inverse head and shoulders | a rise | 393 | +1.57% | +0.95% | +0.15% | survives |
One of the six passes every test. Three fail. Two have too few signals to judge. That is the honest state of it.
How each one was measured
Every detector can only see the past. A turning point needs 5 days on either side to be a turning point at all,
so no detector is allowed to use one until 5 days after it happened. A test in tests/patterns.test.mjs cuts the history
short at three places and checks that the signals before the cut do not change — if a detector were reading the future, that test would fail.
The history is 38 symbols of finished daily bars from the chart API: index and sector funds, the large names, and laggards (INTC, F, T, VZ, PFE, GE) on purpose, so the set is not only the winners. Each signal is measured to the close 5, 20 and 60 finished days later, and compared with four things: an ordinary day in the same history; the same symbol's own drift in the year around the signal; 300 random draws of the same number of days; and the same rule with its dials moved. Then the whole thing is cut in half by date and both halves are read separately.
A pattern is only called a survivor if it beats an ordinary day at 20 and 60 days, lands outside the random band at least once, still points the right way after its own drift is removed, and tells the same story in both halves of history. Those five tests were written before the numbers were looked at.
What survivor bias still does to these numbers
The chart API holds the symbols that are listed today. Every company that failed and was delisted is missing, and no detector here can see them. That flatters every upward number on this page, including the ordinary-day baselines. The “drift removed” column is the part that survives this objection — it compares a signal with what that same symbol was doing anyway — and it is always the smaller number. Where a pattern only looks good in the raw column and not in the drift-removed one, the honest reading is that the pattern found a rising market, not an edge.
The rising wedge
the pattern claims a fall · 1,739 signals found · does not survive
What the books say
A wedge slopes against itself: both boundaries run the same way and converge, and prices usually fall away from a rising wedge rather than break out of its top.Edwards & Magee, Technical Analysis of Stock Trends — the chapters on reversal phenomena — paraphrase, not a quotation
The rising wedge is read as a pause that leans the wrong way: rising, but with the buying getting weaker, so it is treated as bearish.Murphy, Technical Analysis of the Financial Markets — the chapter on continuation patterns — paraphrase, not a quotation
Bulkowski's own counts put wedges among the weaker patterns, with a high rate of what he calls throwbacks.Bulkowski, Encyclopedia of Chart Patterns — the rising-wedge chapter — paraphrase, not a quotation
The rule, in words
Higher highs and higher lows, but the floor rises faster than the ceiling, so the two lines close on each other. The reading is that buyers are tiring. The signal is the close that breaks the rising floor.
One real example
What followed, counted
| horizon | after the signal | from any day | drift removed | random draws, same count | signals |
|---|---|---|---|---|---|
| 5 days later | 53% up · +0.21% | 54% up · +0.27% | -0.11% | +0.15% to +0.42% | 1,736 |
| 20 days later | 56% up · +0.74% | 57% up · +0.95% | -0.26% | +0.72% to +1.24% | 1,723 |
| 60 days later | 60% up · +2.18% | 60% up · +2.46% | -0.68% | +2.15% to +3.01% | 1,710 |
“From any day” is every day of the same 38 symbols over the same history — the thing a signal has to beat. “Drift removed” subtracts what that same symbol was doing anyway in the year around each signal, which is the part of the answer survivor bias cannot reach. “Random draws” is 300 draws of the same number of days, picked at random from the same histories: the middle 90% of what luck looks like.
The same rule with the dials moved
| setting | signals | 20 days up | middle move | drift removed |
|---|---|---|---|---|
| pivot 4 days, closing to 60% | 1,362 | 57% | +0.80% | -0.25% |
| pivot 4 days, closing to 75% | 1,628 | 56% | +0.73% | -0.22% |
| pivot 4 days, closing to 90% | 1,842 | 56% | +0.73% | -0.27% |
| pivot 5 days, closing to 60% | 1,457 | 56% | +0.74% | -0.27% |
| pivot 5 days, closing to 75% | 1,739 | 56% | +0.74% | -0.26% |
| pivot 5 days, closing to 90% | 1,972 | 57% | +0.77% | -0.23% |
| pivot 7 days, closing to 60% | 1,227 | 57% | +0.86% | -0.18% |
| pivot 7 days, closing to 75% | 1,541 | 57% | +0.89% | -0.13% |
| pivot 7 days, closing to 90% | 1,778 | 57% | +0.96% | -0.08% |
| three touches per line, pivot 4 days | 998 | 58% | +0.85% | -0.24% |
| three touches per line, pivot 5 days | 789 | 57% | +0.87% | -0.15% |
| three touches per line, pivot 7 days | 462 | 55% | +0.66% | -0.21% |
If one setting shines and its neighbours do not, that setting is luck, not a rule.
The two halves of history
2003–2014: 833 signals, middle move +0.61%, against +0.93% for an ordinary day in that half.
2015–2026: 890 signals, middle move +0.83%, against +0.97% for an ordinary day in that half.
Verdict
- passes — 20 days: middle move +0.74% against +0.95% from any day — the pattern's way.
- fails — 20 days against chance: drawing the same number of days at random gives a middle move between 0.72% and 1.24%; this is inside that.
- passes — 20 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, -0.26% is left — still the pattern's way.
- passes — 60 days: middle move +2.18% against +2.46% from any day — the pattern's way.
- fails — 60 days against chance: drawing the same number of days at random gives a middle move between 2.15% and 3.01%; this is inside that.
- passes — 60 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, -0.68% is left — still the pattern's way.
- passes — both halves of history: 2003-2014: 0.61% against its own ordinary 0.93% (the pattern's way, 833 signals). 2015-2026: 0.83% against 0.97% (the pattern's way, 890 signals).
The falling wedge
the pattern claims a rise · 962 signals found · does not survive
What the books say
The mirror of the rising wedge: two down-sloping lines closing on each other, from which prices usually escape upward.Edwards & Magee — the chapters on reversal phenomena — paraphrase, not a quotation
A falling wedge is read as selling that is running out, and is treated as bullish.Murphy — the chapter on continuation patterns — paraphrase, not a quotation
Measured, it is another of the modest ones.Bulkowski — the falling-wedge chapter — paraphrase, not a quotation
The rule, in words
Lower highs and lower lows, but the ceiling falls faster than the floor, so the lines close on each other. The signal is the close that breaks out through the falling ceiling.
One real example
What followed, counted
| horizon | after the signal | from any day | drift removed | random draws, same count | signals |
|---|---|---|---|---|---|
| 5 days later | 56% up · +0.45% | 54% up · +0.27% | +0.21% | +0.06% to +0.42% | 962 |
| 20 days later | 56% up · +1.01% | 57% up · +0.95% | +0.36% | +0.42% to +1.17% | 961 |
| 60 days later | 59% up · +2.70% | 60% up · +2.46% | +1.03% | +1.53% to +2.87% | 955 |
“From any day” is every day of the same 38 symbols over the same history — the thing a signal has to beat. “Drift removed” subtracts what that same symbol was doing anyway in the year around each signal, which is the part of the answer survivor bias cannot reach. “Random draws” is 300 draws of the same number of days, picked at random from the same histories: the middle 90% of what luck looks like.
The same rule with the dials moved
| setting | signals | 20 days up | middle move | drift removed |
|---|---|---|---|---|
| pivot 4 days, closing to 60% | 705 | 56% | +0.96% | +0.43% |
| pivot 4 days, closing to 75% | 891 | 57% | +1.04% | +0.42% |
| pivot 4 days, closing to 90% | 1,021 | 57% | +1.04% | +0.38% |
| pivot 5 days, closing to 60% | 745 | 56% | +1.01% | +0.33% |
| pivot 5 days, closing to 75% | 962 | 56% | +1.01% | +0.36% |
| pivot 5 days, closing to 90% | 1,138 | 56% | +0.97% | +0.23% |
| pivot 7 days, closing to 60% | 679 | 58% | +1.30% | +0.63% |
| pivot 7 days, closing to 75% | 906 | 57% | +1.08% | +0.42% |
| pivot 7 days, closing to 90% | 1,095 | 57% | +1.09% | +0.36% |
| three touches per line, pivot 4 days | 418 | 55% | +0.58% | +0.02% |
| three touches per line, pivot 5 days | 339 | 56% | +0.80% | +0.45% |
| three touches per line, pivot 7 days | 215 | 59% | +1.39% | +0.90% |
If one setting shines and its neighbours do not, that setting is luck, not a rule.
The two halves of history
2003–2014: 442 signals, middle move +0.89%, against +0.93% for an ordinary day in that half.
2015–2026: 519 signals, middle move +1.18%, against +0.97% for an ordinary day in that half.
Verdict
- passes — 20 days: middle move +1.01% against +0.95% from any day — the pattern's way.
- fails — 20 days against chance: drawing the same number of days at random gives a middle move between 0.42% and 1.17%; this is inside that.
- passes — 20 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, +0.36% is left — still the pattern's way.
- passes — 60 days: middle move +2.7% against +2.46% from any day — the pattern's way.
- fails — 60 days against chance: drawing the same number of days at random gives a middle move between 1.53% and 2.87%; this is inside that.
- passes — 60 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, +1.03% is left — still the pattern's way.
- fails — both halves of history: 2003-2014: 0.89% against its own ordinary 0.93% (the wrong way, 442 signals). 2015-2026: 1.18% against 0.97% (the pattern's way, 519 signals).
The bull flag
the pattern claims a rise · 100 signals found · too few
What the books say
A flag is a short, compact drift against a steep prior move — the pause that “flies at half-mast” in the middle of a run.Edwards & Magee — the chapter on consolidation formations — paraphrase, not a quotation
Flags and pennants are brief pauses in a fast move, and the move is expected to continue afterwards.Murphy — the chapter on continuation patterns — paraphrase, not a quotation
Flags are numerous and short-lived; their measured edge is small.Bulkowski — the flag chapter — paraphrase, not a quotation
The rule, in words
A sharp run up (the pole), then a short tight pause that gives back only part of it (the flag), then a close above the pause's high. This is M17's detector, unchanged.
One real example
What followed, counted
| horizon | after the signal | from any day | drift removed | random draws, same count | signals |
|---|---|---|---|---|---|
| 5 days later | 55% up · +0.32% | 54% up · +0.27% | -0.16% | -0.41% to +0.97% | 100 |
| 20 days later | 51% up · +0.01% | 57% up · +0.95% | -0.62% | -0.17% to +2.69% | 99 |
| 60 days later | 61% up · +2.06% | 60% up · +2.46% | -1.68% | +0.82% to +5.98% | 98 |
“From any day” is every day of the same 38 symbols over the same history — the thing a signal has to beat. “Drift removed” subtracts what that same symbol was doing anyway in the year around each signal, which is the part of the answer survivor bias cannot reach. “Random draws” is 300 draws of the same number of days, picked at random from the same histories: the middle 90% of what luck looks like.
The same rule with the dials moved
| setting | signals | 20 days up | middle move | drift removed |
|---|---|---|---|---|
| rise 10%, gives back 40% | 234 | 54% | +1.17% | -0.66% |
| rise 10%, gives back 50% | 310 | 55% | +0.47% | -0.82% |
| rise 10%, gives back 62% | 367 | 55% | +0.90% | -0.82% |
| rise 15%, gives back 40% | 78 | 52% | +0.34% | +0.42% |
| rise 15%, gives back 50% | 100 | 51% | +0.01% | -0.62% |
| rise 15%, gives back 62% | 121 | 49% | -0.26% | -0.85% |
| rise 20%, gives back 40% | 35 | 62% | +2.70% | +0.81% |
| rise 20%, gives back 50% | 41 | 60% | +1.49% | +0.72% |
| rise 20%, gives back 62% | 49 | 58% | +1.49% | +0.72% |
If one setting shines and its neighbours do not, that setting is luck, not a rule.
The two halves of history
2003–2014: 31 signals, middle move +0.34%, against +0.93% for an ordinary day in that half.
2015–2026: 68 signals, middle move -0.26%, against +0.97% for an ordinary day in that half.
Verdict
The bear flag
the pattern claims a fall · 51 signals found · too few
What the books say
The same shape upside down: a steep fall, then a small drift back up, then the fall resuming.Edwards & Magee — the chapter on consolidation formations — paraphrase, not a quotation
A pause in a decline, expected to resolve downward.Murphy — the chapter on continuation patterns — paraphrase, not a quotation
Counted with the flags.Bulkowski — the flag chapter — paraphrase, not a quotation
The rule, in words
A sharp fall (the pole), then a short tight drift that recovers only part of it, then a close below the drift's low. The same rule as the bull flag, mirrored, so the two can be compared without arguing about the definition.
One real example
What followed, counted
| horizon | after the signal | from any day | drift removed | random draws, same count | signals |
|---|---|---|---|---|---|
| 5 days later | 63% up · +1.35% | 54% up · +0.27% | +1.25% | -0.56% to +1.10% | 51 |
| 20 days later | 55% up · +1.58% | 57% up · +0.95% | +0.68% | -1.03% to +2.63% | 51 |
| 60 days later | 58% up · +4.37% | 60% up · +2.46% | +3.78% | -0.55% to +5.96% | 50 |
“From any day” is every day of the same 38 symbols over the same history — the thing a signal has to beat. “Drift removed” subtracts what that same symbol was doing anyway in the year around each signal, which is the part of the answer survivor bias cannot reach. “Random draws” is 300 draws of the same number of days, picked at random from the same histories: the middle 90% of what luck looks like.
The same rule with the dials moved
| setting | signals | 20 days up | middle move | drift removed |
|---|---|---|---|---|
| fall 10%, gives back 40% | 105 | 61% | +1.96% | +1.74% |
| fall 10%, gives back 50% | 162 | 60% | +2.07% | +1.80% |
| fall 10%, gives back 62% | 227 | 59% | +1.96% | +1.27% |
| fall 15%, gives back 40% | 30 | 50% | -0.42% | -1.23% |
| fall 15%, gives back 50% | 51 | 55% | +1.58% | +0.68% |
| fall 15%, gives back 62% | 68 | 56% | +1.73% | +0.82% |
| fall 20%, gives back 40% | 16 | 56% | +7.46% | +5.55% |
| fall 20%, gives back 50% | 22 | 50% | -1.24% | -3.18% |
| fall 20%, gives back 62% | 28 | 54% | +2.47% | +0.81% |
If one setting shines and its neighbours do not, that setting is luck, not a rule.
The two halves of history
2003–2014: 36 signals, middle move -3.54%, against +0.93% for an ordinary day in that half.
2015–2026: 15 signals, middle move +9.93%, against +0.97% for an ordinary day in that half.
Verdict
Head and shoulders
the pattern claims a fall · 247 signals found · does not survive
What the books say
The best known reversal: a peak, a higher peak, a lower peak, and a neckline drawn through the two valleys between them. The pattern is not complete until the price closes through the neckline.Edwards & Magee — the chapter on important reversal patterns — paraphrase, not a quotation
The neckline break is the signal; everything before it is a shape that may still fail to form.Murphy — the chapter on reversal patterns — paraphrase, not a quotation
One of the better-performing classical patterns in his counts, with failure rates that rise sharply in strong bull markets.Bulkowski — the head-and-shoulders top chapter — paraphrase, not a quotation
The rule, in words
Three turning points with the middle one highest, shoulders within a little of each other, and a neckline through the two valleys. The signal is the close below that neckline — not the shape itself.
One real example
What followed, counted
| horizon | after the signal | from any day | drift removed | random draws, same count | signals |
|---|---|---|---|---|---|
| 5 days later | 50% up · +0.01% | 54% up · +0.27% | -0.11% | -0.08% to +0.59% | 247 |
| 20 days later | 61% up · +1.57% | 57% up · +0.95% | +0.60% | +0.10% to +1.60% | 246 |
| 60 days later | 61% up · +2.62% | 60% up · +2.46% | +1.16% | +0.90% to +3.66% | 246 |
“From any day” is every day of the same 38 symbols over the same history — the thing a signal has to beat. “Drift removed” subtracts what that same symbol was doing anyway in the year around each signal, which is the part of the answer survivor bias cannot reach. “Random draws” is 300 draws of the same number of days, picked at random from the same histories: the middle 90% of what luck looks like.
The same rule with the dials moved
| setting | signals | 20 days up | middle move | drift removed |
|---|---|---|---|---|
| pivot 4 days, shoulders within 8% | 195 | 64% | +2.19% | +1.38% |
| pivot 4 days, shoulders within 12% | 201 | 64% | +2.15% | +1.24% |
| pivot 4 days, shoulders within 18% | 204 | 64% | +2.16% | +1.26% |
| pivot 5 days, shoulders within 8% | 240 | 61% | +1.45% | +0.57% |
| pivot 5 days, shoulders within 12% | 247 | 61% | +1.57% | +0.60% |
| pivot 5 days, shoulders within 18% | 251 | 61% | +1.43% | +0.53% |
| pivot 7 days, shoulders within 8% | 206 | 59% | +0.82% | +0.29% |
| pivot 7 days, shoulders within 12% | 213 | 59% | +1.07% | +0.34% |
| pivot 7 days, shoulders within 18% | 219 | 58% | +0.85% | +0.26% |
If one setting shines and its neighbours do not, that setting is luck, not a rule.
The two halves of history
2003–2014: 108 signals, middle move +1.94%, against +0.93% for an ordinary day in that half.
2015–2026: 138 signals, middle move +1.30%, against +0.97% for an ordinary day in that half.
Verdict
- fails — 20 days: middle move +1.57% against +0.95% from any day — the wrong way.
- fails — 20 days against chance: drawing the same number of days at random gives a middle move between 0.1% and 1.6%; this is inside that.
- fails — 20 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, +0.6% is left — the wrong way.
- fails — 60 days: middle move +2.62% against +2.46% from any day — the wrong way.
- fails — 60 days against chance: drawing the same number of days at random gives a middle move between 0.9% and 3.66%; this is inside that.
- fails — 60 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, +1.16% is left — the wrong way.
- fails — both halves of history: 2003-2014: 1.94% against its own ordinary 0.93% (the wrong way, 108 signals). 2015-2026: 1.3% against 0.97% (the wrong way, 138 signals).
Inverse head and shoulders
the pattern claims a rise · 393 signals found · survives
What the books say
The same figure inverted at a bottom, where the neckline is broken upward.Edwards & Magee — the chapter on important reversal patterns — paraphrase, not a quotation
Read as a bottom that has been tested three times, the middle test the deepest.Murphy — the chapter on reversal patterns — paraphrase, not a quotation
Among his stronger classical patterns.Bulkowski — the head-and-shoulders bottom chapter — paraphrase, not a quotation
The rule, in words
Three turning points with the middle one lowest, shoulders alike, a neckline through the two peaks between them, and a close above it as the signal.
One real example
What followed, counted
| horizon | after the signal | from any day | drift removed | random draws, same count | signals |
|---|---|---|---|---|---|
| 5 days later | 56% up · +0.26% | 54% up · +0.27% | -0.09% | +0.04% to +0.55% | 393 |
| 20 days later | 59% up · +1.57% | 57% up · +0.95% | +0.15% | +0.42% to +1.45% | 393 |
| 60 days later | 61% up · +3.15% | 60% up · +2.46% | +0.11% | +1.63% to +3.48% | 389 |
“From any day” is every day of the same 38 symbols over the same history — the thing a signal has to beat. “Drift removed” subtracts what that same symbol was doing anyway in the year around each signal, which is the part of the answer survivor bias cannot reach. “Random draws” is 300 draws of the same number of days, picked at random from the same histories: the middle 90% of what luck looks like.
The same rule with the dials moved
| setting | signals | 20 days up | middle move | drift removed |
|---|---|---|---|---|
| pivot 4 days, shoulders within 8% | 327 | 61% | +1.90% | +0.85% |
| pivot 4 days, shoulders within 12% | 348 | 61% | +1.68% | +0.80% |
| pivot 4 days, shoulders within 18% | 352 | 61% | +1.67% | +0.78% |
| pivot 5 days, shoulders within 8% | 372 | 59% | +1.63% | +0.17% |
| pivot 5 days, shoulders within 12% | 393 | 59% | +1.57% | +0.15% |
| pivot 5 days, shoulders within 18% | 395 | 59% | +1.55% | +0.14% |
| pivot 7 days, shoulders within 8% | 369 | 52% | +0.43% | -0.66% |
| pivot 7 days, shoulders within 12% | 386 | 53% | +0.49% | -0.65% |
| pivot 7 days, shoulders within 18% | 389 | 53% | +0.46% | -0.66% |
If one setting shines and its neighbours do not, that setting is luck, not a rule.
The two halves of history
2003–2014: 192 signals, middle move +1.22%, against +0.93% for an ordinary day in that half.
2015–2026: 201 signals, middle move +1.63%, against +0.97% for an ordinary day in that half.
Verdict
- passes — 20 days: middle move +1.57% against +0.95% from any day — the pattern's way.
- passes — 20 days against chance: drawing the same number of days at random gives a middle move between 0.42% and 1.45%; this is outside that.
- passes — 20 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, +0.15% is left — still the pattern's way.
- passes — 60 days: middle move +3.15% against +2.46% from any day — the pattern's way.
- fails — 60 days against chance: drawing the same number of days at random gives a middle move between 1.63% and 3.48%; this is inside that.
- passes — 60 days after removing drift: after subtracting what the same symbol was doing anyway in the year around each signal, +0.11% is left — still the pattern's way.
- passes — both halves of history: 2003-2014: 1.22% against its own ordinary 0.93% (the pattern's way, 192 signals). 2015-2026: 1.63% against 0.97% (the pattern's way, 201 signals).
What could be wrong
- The rules are one reasonable reading of each pattern, not the only one. The dials sweep shows how much the answer moves when the reading changes; for the wedges it moves a lot.
- The books are not on this Mac. The descriptions above are careful paraphrases, cited to chapter titles rather than numbers, and should be checked against the texts before any of this is printed.
- Delisted companies are missing entirely, as above.
- Nothing here accounts for costs, slippage, or the money you would have to risk to take a signal. A 0.3% edge before costs may be no edge at all.
- “Survives” means it passed five tests on this history. It does not mean it will work next month.
What was not done
- No pattern here is wired into the Hub, and nothing places a trade or suggests one.
- Volume is not part of any rule, though the books make it part of several. That is the next thing to add.
- Targets and stops (the “measured move” each book gives) are not tested — only what happened at 5, 20 and 60 days.
- Intraday and weekly bars are untouched; this is daily only.