The 72-week cycle: what the data actually sayUS Wall Street 30 IndexPEPPERSTONE:US30KruegerAlgorithmsThe thesis Count 72 weekly bars from the April 2025 low and the box ends in late August 2026. The Dow is 2–4 % below its August high, bounced twice only to the 50 % retracement and was rejected both times, and is now, by the cycle count, in week 74. Week 73 wicked lower and was bought. The reading in the feed: a time window for a turn is open, and a weekly close below last week's low opens the downside. Enough traders watch the same count that the date itself could become the trigger. The two rejections at the 50 % retracement on the hourly chart, and the projection the feed draws from them. Daily: third test of the 50-day EMA, the 100-day below at 51,840. What speaks for it Turning points are not spread evenly in time. Volatility comes in regimes that persist for months, so lows and highs cluster, and any regular grid laid over a chart will catch some of them. If a large crowd trades the same date, the expectation can produce the move it predicts. And the count costs nothing to watch. Those are real mechanisms, not mysticism. What we measured The cycle mechanically, on daily cash-session bars of eight indices, 2015–2026, pivots defined as extremes within a ±20-day window. Does a pivot recur 72 weeks after a pivot more often than chance? Hit rate 20.1 % versus 21.7 % for a randomly chosen day. Rank 98 of 121 tested lags. A random anchor day performs the same as a real pivot (21.5 %). Across 12 definitions, 72 weeks beats the base rate in two cells with t ≈ 2.3. A permutation test on those cells: pretending only 72 was tested gives p = 0.04. Asking honestly whether any lag between 10 and 130 weeks would come out at least this strong gives p = 0.95. Noise produces a "72" somewhere 95 times out of 100. "But people make money with it": we scored our own validated intraday setups on cycle days against all other days. Cycle days +0.125 R, other days +0.097 R, difference +0.028 R, t = 1.7. A placebo with random anchor dates produces the same uplift (+0.020 R) and a larger maximum (+0.072 R) than the best real lag. The effect belongs to the construction "small subset versus rest", not to the pivot and not to 72. If a crowd trades the date, volatility should rise around it. Autocorrelation of the weekly range at lag 72: +0.002, rank 50 of 121. Volatility on cycle days versus the rest: t = 1.4 after clustering. The strongest volatility persistence sits at 103–111 weeks, a two-year regime effect, not a cycle. The anchor is soft. 72 weeks before 3 September 2026 is 17 April 2025, a holiday. The tariff low was 7–8 April (73.4 weeks), the retest low 21–22 April (71.4 weeks). At the Covid low, index and CFD data disagree by up to five trading days about which day was the bottom. A count that names a precise week already carries a week of uncertainty in its starting point, and a count that is still valid at 73 and 74 no longer names a week at all. The price condition, measured on the Dow since 2015: a weekly close below the previous week's low while 1–6 % below a high that is at most 12 weeks old. 28 cases. New all-time high first: 79 %. A drawdown of 8 % below the high first: 21 %. Without the condition, any such week: 75 % and 23 %. The three real tops of the period (February 2018, January 2020, January 2022) are in the 28, and so are 25 weeks after which the index went on to a new high. The close below last week's low did not move the odds. The 30-year yield at levels last seen in 2007 is the part of the thesis this test cannot touch, see the limit below. Two neighbouring claims, same data: the Dow has gone 42, 63 and 70 weeks without touching its weekly 50-EMA; after 21 weeks without a touch, the touch came 17 to 49 weeks later in all four cases, never within 13. Inside those long phases, pullback lows recur every 6–12 weeks; none of 21–33 measured gaps fell in an 18–22-week window. What is left Not nothing. A date makes a claim falsifiable, if it is written down before the fact. Ours, written now: if this week closes below last week's low, the cycle thesis predicts the Dow trades below roughly 50,000 (8 % under the August high) before it prints a new all-time high. Historically that resolved the other way in about three cases out of four. We will update this idea with the outcome either way. The structural base rate does not need the cycle: 2–4 % below a fresh high with no new high for two weeks, the Dow made a new high first in 73 % of 30 cases since 2015 and fell a further 5 % first in 23 %. The cycle does not move that number, because in the data it does not move anything. The limit Eleven years of data, eight correlated indices, pivots that need 20 days of hindsight to be identified, which means the test measures the upper bound of what a perfectly recognised cycle could deliver. Whether individual traders earn money with a cycle narrative was not tested and cannot be from price data; if they do, the source is sizing and exits, not the calendar. Rates, oil and the dollar are not in any of these counts; the yield chart above is context, not a measured input.