Positioning Roundup: Week of July 28, 2026

Last week we asked whether the unanimous grain long was conviction or "the crowding that the next cool, wet run in the models unwinds." We got the answer in three sessions. In the COT week ending Tuesday July 28, managed money finished the job: soybeans +155,001 and soybean meal +88,655 — the highest net long ever recorded for this week of the year in the 14 years of the disaggregated file — while corn covered its way to +168,399, now the single largest position on the board. Then Wednesday through Friday, Midwest rain forecasts took 2.5–5.1% out of every leg of the complex and 6.6% out of the board crush. Two things complicate the neat version of that story, and both are in the data: corn's record add was mostly short-covering, not new buying, and soybean oil was already being sold into its own record.

Grains The record, and the three days after it

Start with what the snapshot actually says, because the seasonal framing matters more than the raw number. Soybeans at +155,001 is the largest week-31 net long in the 14 years the CFTC has published the disaggregated file (2013–2026); the prior high for the date was +106,338 in 2016. Soybean meal at +88,655 beat its own week-31 record of +80,018 from 2022. Soybean oil's +109,855 is likewise a week-31 record against +67,913 in 2017 — but it got there by falling 15,493 from a level so high the record survived the liquidation. Only 13 prior years sit behind each of these comparisons, which is a thin base for the word "record" and we would rather say so than lean on it.

Corn is the loudest number and the most misread. +75,490 in a single week to +168,399 — the 26th-largest weekly add in 707 weeks, and the biggest position on the board. But decompose it: gross shorts fell 57,167 while gross longs rose only 18,323. Roughly three-quarters of that move is shorts getting out, not bulls getting in. That is a squeeze, not an accumulation, and it explains why the level is only the 74th percentile despite the violence of the change — corn has been far longer than this, at +246,500 in 2021 and +242,801 in 2015. Over four weeks the swing totals +214,608, the 17th largest in 704 weeks. A position built that fast on covering is structurally different from one built on conviction: the fuel is spent, and the marginal buyer has already bought.

And it ran against the calendar, which is the detail that makes it worth writing about. Late July is normally when funds reduce corn: across 2013–2025 the median four-week change in managed money corn net over ISO weeks 29–33 is −25,436 contracts, a seasonal tilt of −22,479 against the all-year median. Only 2 of 67 prior late-July observations show a bigger four-week build than this one. Whatever this was, it was not the seasonal pattern reasserting itself — and the same goes for the records themselves. Week 31 is one of the weakest points of the soy complex's year: the 2013–2025 mean soybean net for this week ranks 47th of 52 weeks, meal and oil both 45th. Setting a record in week 31 is not a seasonal high that arrived on schedule; it is a record at a point in the calendar where positions are usually small.

Now the part that changes the reading. Funds were not buying into strength — they were buying into a tape that had already turned. Across the reporting week itself (July 21 to July 28), soybean oil fell 4.8%, meal 2.0% and beans 0.6%. Only corn rose, and only 1.3%. So the record week-31 long in beans and meal was assembled while those contracts were declining. That is averaging into weakness on a weather trade whose premium had already begun to leak — and it is a materially less flattering description of the crowd than "record conviction."

Then the forecast landed. From Tuesday's snapshot to Friday's close, soybean oil −5.1%, corn −3.9%, SRW −3.5%, soybeans −3.3%, HRW −2.6%, meal −2.5%. Measured Friday-to-Friday the week was worse still: oil −9.7%, a five-session decline exceeded in only 19 of 4,163 sessions since 2010 — a half-percent tail. That last figure needs a seasonal check, and the check splits the complex. Corn's −5.1% is much less remarkable than the all-sessions percentile implies: late July is routinely ugly for corn, with a median five-session return of −1.87% over Jul 20–31 across 2010–2025, and 23.5% of those windows were this bad or worse against 6.2% of all sessions. Soybeans at −6.1% sit in the worst 9.6% of late-July windows. Soybean oil is the one that survives the test outright: none of the 136 prior late-July five-session windows since 2010 were as bad as −9.7%. So the corn break is largely the season doing what the season does; the oil break is not. The board crush margin, the metric we said last week was the tell we had misread, fell from $2.98 to $2.53 a bushel, −15.1% Friday to Friday — and back below the $2.60 trough it bounced from on July 7. Measured from Tuesday's snapshot at $3.17 the drop is 20.2%. The entire round trip that built this position took four weeks; giving back its price took four days.

The other side Why this may mean less than it looks

Four reasons to hold this loosely, in descending order of how much they worry us.

One: the COT is a Tuesday photograph. Every number above is as of July 28. The break happened after. We do not know that funds held these positions through Friday — the honest possibility is that a good part of this long was already gone by the time the file was published, and the August 4 report will show a position that was never as exposed as the snapshot implies. Anyone telling you the crowd "got caught" is asserting something this data cannot show.

Two: the forward test is thinner than it first appears. Screen the record for weeks in the weather window (ISO 28–35) where the soybean net long sat at today's 84th percentile or above and you get exactly one prior instance: July 12, 2016, which liquidated 46,412 contracts and fell 7.7% over the next four weeks. One. Loosen the threshold to the 70th percentile and you get 11 cases, a median four-week price change of −7.2%, negative in 9 of 11, with the position liquidated in the same 9. That reads as a strong pattern until you notice those 11 cases fall in only four distinct years — 2013, 2016, 2020 and 2023 — and that consecutive weeks inside one summer are not independent observations. The effective sample is closer to four than eleven, and one of the four (2020) went the other way, +11.6% on Chinese buying.

Three: the crop survey never backed the rally, and does not back the break either. Corn is at 63% Good+Excellent against a trailing five-year normal of 64.6% for the same week — an anomaly of −1.6 points. Soybeans are at 63% against 61.6%, so above normal. Yes, corn has fallen 5 points in three weeks (68 → 67 → 63) and that direction is real. But the level is ordinary. Managed money added 214,608 corn contracts over four weeks on a crop the survey describes as average, and the survey has not moved enough in either direction to justify the 5% break either. Conditions are a state-weighted subjective judgment, not a yield, and this is a week where they simply were not the operative variable.

Four: the belt split is real, and once you weight it by production it nets to almost exactly nothing — because one state is holding it up. Last week we flagged the 29.8-point spread between Minnesota at 77% Good+Excellent (+19.6 vs normal) and South Dakota at 50% (−10.2) as an open question, because a split only matters if it maps to where the corn actually is. It does not. Weighting each of the ten belt states by its 2025 corn production, the production-weighted condition anomaly for the belt is +0.08 points — dead normal — against a simple ten-state average of −0.32 and a published national figure of −1.6. Below-normal states carry 49.8% of belt production and above-normal states 50.2%. The two states in the headline spread are only 20.5% of belt production combined.

But there is a catch inside the catch, and it is the more useful number. Minnesota is carrying the entire result. At 12.5% of belt production with a +19.6 anomaly, it contributes +2.45 points on its own; drop Minnesota and the production-weighted belt anomaly falls from +0.08 to −2.71 — from normal to meaningfully below it. Soybeans behave the same way: +1.79 weighted, but −0.69 without Minnesota. So the honest reading is not "the belt is fine" and not "the belt is splitting." It is that a single exceptional state is masking a mildly below-normal crop everywhere else, and Minnesota is 12.5% of the corn while Iowa, Illinois and Nebraska together are 52.7%. That is a fragile thing for a market to be relaxed about, and it is the number we would watch degrade first.

Livestock The hog short is still short for the date

Both pen trades did what we said they would, and one of them deserves a correction of emphasis. Live cattle cut another 8,839 to +66,523, pushing the spring's record long to the 46th percentile — below median for the first time in this unwind, and five consecutive weeks of liquidation. The position that dominated the board in May is now unremarkable.

Hogs kept covering — +7,273 to −10,884, the second straight week of buying since the squeeze began. Last week we framed that as an extreme coming home. The seasonal comparison says something sharper: −10,884 is the lowest week-31 net position in the entire 14-year record, and one of only two negative readings for the date (2018 was −7,841). In twelve of the last thirteen years, funds were net long hogs at the start of August, by as much as +107,586 a year ago. So the covering is real and the position is still, by the standard of the calendar, historically short. The squeeze fuel is not spent — it is merely less full than it was.

Softs Coffee's inventory is doing something the position isn't

Coffee positioning barely moved again — +270 to +25,136, the 60th percentile — and arabica fell 7.6% after the snapshot, from 339.40 on Tuesday to 313.55 Friday. That is the frost premium continuing to drain, as it has for three weeks, and our frost model shows zero belt risk through August 7. But underneath the quiet position, the physical market moved hard: ICE certified arabica stocks fell 30.0% in July, from 377,465 to 264,179 bags — the third-steepest monthly draw since 2010 and a level in the 2nd percentile of the last 199 months. Four consecutive months of decline, −47% since April. July is mildly negative for certified stocks as a rule — the 2010–2025 July median is −1.45% — but none of the sixteen prior Julys came within a country mile of −30%, so this is not the calendar doing its usual work.

We are deliberately not building a thesis on that here. The two prior comparable draws since 2010 point in opposite directions — after September 2022 arabica fell 23% in a month; after November 2023 it rose 2%, then 18% over six. An n of two that disagrees with itself is not a signal, and it deserves its own piece rather than a paragraph. What is worth saying now is the shape: visible inventory at a 2nd-percentile level, and managed money holding a middling 60th-percentile long that expresses no view about it at all.

Elsewhere in the corner: sugar's short deepened another 15,817 to −112,413, its 14th percentile, on a 2.2% price decline — the one soft where funds are actively pressing. Cocoa re-shorted 2,523 to −14,127 after last week's bounce, and cotton was flat at +52,410 while price firmed 1.6% after the snapshot — the only agricultural market that rose while the grain complex broke.

Energy & Metals The crude divergence closed

For three weeks we have flagged crude rallying without the specs — a 6th-percentile managed-money long against a rising tape. This week the specs showed up: WTI +21,402 to +108,307, the largest add on the energy board, with the nearby at $85.84. The gap we kept pointing at closed from the positioning side rather than the price side, which is the resolution that favors the people who were already long. It is still only the 12th percentile of the 2010-to-date record, so "showed up" means arrived, not crowded.

Natural gas went the other way and simply got heavier: −2,911 to −105,605, a 15th-percentile short at $2.78 with a 1.68 million-lot open interest. Products were mixed (RBOB flat at +73,967, its 82nd percentile; ULSD −2,317 to +11,374). In metals, gold eased 3,258 to +120,328 and gave up the top of the board it held through the spring; copper came off its breakout by 7,195 to +66,490, though at the 93rd percentile of its shorter history it remains the most crowded metal; silver slipped to +8,387.

Ahead What to watch this week

The whole question now is whether the record was a photograph of a crowd that has already left. Two reports settle it. If the crowding read is right, the next two COTs should show net liquidation in beans from +155,001 — the historical median four-week drawdown from this zone is about 36,000 contracts. If instead the bean long holds above +140,000 through the August 12 WASDE, the crowding story is wrong and what we are looking at is real demand that took a weather-driven price hit and stayed. That is the falsifiable version, and we will grade it here on August 8 and August 15.