Distillate fuel oil is not diesel, and the tables are written in the former

Distillate fuel oil and diesel are not the same thing, and EIA's weekly tables are written in the former while every news report is written in the latter. Distillate fuel oil is a refinery output category that includes on-highway diesel, off-road diesel, heating oil and some industrial fuels. Diesel sold at a pump is one sulphur grade inside it. So a headline about a distillate stock draw is not necessarily a headline about the diesel supply, and the sulphur split below is the part that answers the question a reader is actually asking.

This is the single most common category error in reporting on this market. A weekly distillate stock draw is reported as a diesel shortage; a distillate crack is reported as a diesel margin. In the present mix those readings happen to be close to right, and the next section shows why — but they are right by coincidence of composition, not by definition, and the coincidence is not stable.

The sulphur split — how much of the distillate headline is diesel

Stocks as of the week ended October 16, 2026. The share column is the Institute's computation.

Stock category	Million barrels	Share of total distillate	What it is
Total distillate fuel oil stocks	104.3	100.0%	The number that appears in headlines, and it reversed. It BUILT 1.7 million barrels from 102.6 million — unrounded, 104,286 thousand against 102,611, a build of 1,675 thousand barrels, or 1.6 per cent of the stock in a single week. Last week this row carried the largest draw it had ever printed, 2,527 thousand barrels; this week gives back about two thirds of it. 🔴 AND THE COMPOSITION PATTERN THE PRIOR ENTRY DESCRIBED HAS INVERTED EXACTLY, WHICH IS WHY THAT ENTRY IS REFERENCED RATHER THAN REWRITTEN. Last week the two large grades fell together and the small off-road grade was the only line that rose. This week the two large grades rose together and the small off-road grade is the only line that fell: ULSD built 1,300 thousand barrels, the heating-oil grade built 652, and the off-road grade drew 278. Every one of the three grades reversed its own direction from the prior week. The reference records the symmetry and draws nothing from it. A composition reading that holds for one week is a reading about that week, and a composition reading that inverts in the following week is the evidence for saying so. 🔴 COMPOSITION CHECK, AND THIS WEEK IT IS EXACT IN BOTH DIRECTIONS, WHICH THE PRIOR WEEK WAS NOT. At the precision this row prints, 92.9 plus 4.2 plus 7.2 is 104.3, the total exactly. Unrounded, the three grades sum to 104,286 thousand barrels against a published total of 104,286 — no disagreement at all. Last week the same identity carried a one-thousand-barrel gap, which was disclosed here as the publisher's own independent rounding. Both weeks are correct, and the pair is the useful part: an independent-rounding gap is a property of how a particular week's numbers happen to round, not a standing property of the table. A reader who saw only last week's gap would reasonably have concluded the table does not foot. It does, this week, to the thousand barrel. 🔴 THE YEAR-ON-YEAR COMPARISON IS RESTORED FOR THIS ROW, AND SO IS THE TWO-YEARS-AGO COMPARISON, BECAUSE THE PUBLISHER PRINTED BOTH BASES RATHER THAN BECAUSE THE INSTITUTE REMEMBERED THEM. The two prior entries withdrew these comparisons on the provenance rule — the bases had been read for the week ended October 2 and could not be carried to a different week. Table 4 of the release read on this date prints its own matched columns, headed 10/17/25 and 10/18/24: distillate fuel oil 115.6 million barrels a year ago and 113.8 million two years ago. Against them, 104.3 million is 9.7 per cent BELOW the year-ago level and 8.4 per cent below the two-years-ago level. 🔴 THOSE TWO PERCENTAGES ARE THE PUBLISHER'S OWN AND ARE NOT THE INSTITUTE'S ARITHMETIC, AND THE DIFFERENCE IS A TENTH OF A POINT THAT MATTERS FOR A REASON WORTH STATING. Table 4 carries its own note: “Differences and percent changes are calculated using unrounded numbers.” Computed off the rounded millions printed in the table, 11.3 on 115.6 gives 9.8 per cent; the publisher, working from the thousand-barrel figures behind them, prints 9.7. The publisher's figure is the better one because its inputs are finer, so it is the one carried here. The general practice this sets: where a table prints both a level and its own percent change, print the publisher's percentage and do not re-derive it off the rounded levels — a re-derivation is a measurement of the rounding, and it will disagree with the source by about the amount the rounding is worth. 🔴 SUPERSEDED AND LABELLED, AND THIS ONE IS AN AMENDMENT TO A CLAIM MADE EARLIER TODAY RATHER THAN TO AN EARLIER WEEK. The first draft of this entry, written on this date, read that “the publisher's matched year-ago column carries the distillate TOTAL and does not break it into sulphur grades,” and on that ground it restored this row's comparison while leaving the three grade rows below withdrawn. That sentence was a correct measurement of the stocks table on the highlights page, which is where it was read, and it is FALSE about the release. Table 4 of the same publication breaks the distillate total into all three sulphur grades and prints a matched year-ago and two-years-ago column for every one of them. The grade comparisons are therefore restored in the rows below, and the withdrawal recorded here is withdrawn in turn. 🔴 THE DISCRIMINATOR, BECAUSE AN AMENDMENT THAT ONLY SOFTENS A CLAIM HAS ADDED NOTHING. The defect was not a misreading of a page; every figure in the superseded sentence was read correctly off the page it was read from. The defect was SCOPE — a property of one table asserted as a property of the publication containing it. The release is a multi-part document whose highlights page is a summary and whose numbered tables are the detail, so a series absent from the summary is not thereby absent from the release. The rule that follows is the one this reference already applies to its own sources in the other direction: “not in the table we read” and “not published” are different claims, and a provenance withdrawal must name the table it was measured against, not the publisher. Every withdrawal in this ledger now names a table for exactly this reason. And the two prior entries remain the standing illustration of why a fresh base matters at all: both recorded a year-on-year deficit NARROWING, and in both cases the narrowing turned out to be the BASE moving rather than the level — which cannot be detected unless the base is read fresh. Here both bases were read fresh, by the publisher, for the right week.
— of which 15 parts per million sulphur and under	92.9	89.1%	This is on-highway diesel, and it carried the week again — in the other direction. It BUILT 1,300 thousand barrels from 91,580 to 92,880, a 1.4 per cent build. That is 77.6 per cent of the category's entire 1,675 thousand barrel build, out of a grade that opened the week at 89.2 per cent of the category. So the build was SUB-PROPORTIONAL, by about eleven and a half points. 🔴 AND THE BEHAVIOUR IS THE SAME AS LAST WEEK WHILE THE SHARE CONSEQUENCE IS OPPOSITE, WHICH IS A SIGN TRAP WORTH MORE THAN EITHER WEEK'S NUMBER. Last week this grade supplied 85.2 per cent of a category DRAW out of 89.2 per cent of the stock — sub-proportional by four points — and its share consequently ROSE, from 89.15 to 89.25 per cent. This week it supplied 77.6 per cent of a category BUILD out of the same 89.2 per cent — sub-proportional by eleven and a half points — and its share consequently FELL, to 89.06 per cent. The grade moved less than its weight on both weeks. Because the category reversed direction between them, the identical behaviour pushed the share up once and down once. A reader tracking the share alone would read a reversal in the trucking grade's behaviour where the measurement says there was none. 🔴 THE PRIOR ENTRY'S COMMITMENT IS KEPT, AND THIS WEEK'S MOVE CLEARS THE FLOOR IT SET WITHOUT CLEARING THE BAR. That entry declined to read a direction into tenths of a point, on the reasoning that a share computed from two rounded levels is not a fine enough measurement to carry a weekly finding. Two corrections to that, both in the Institute's own disfavour. The shares printed here are computed from the publisher's unrounded thousand-barrel figures, not from the rounded millions, so they are not rounding artifacts and the reasoning as stated was too generous to itself. And this week's move is 0.19 points, nearly double the tenth the prior entry dismissed, and the largest share move this row has recorded since the 89.7-to-89.2 shift. It is still one observation, the grade still sits within a fifth of a point of where it has sat for three weeks, and no finding is declared on it. 🔴 THE YEAR-ON-YEAR AND TWO-YEARS-AGO COMPARISONS ARE RESTORED, AND THE FINDING THIS ROW HAD CALLED UNMEASURED IS NOW MEASURED AND HOLDS. Table 4 prints a matched year-ago and two-years-ago column for this grade, not only for the total: 105.0 million barrels at 10/17/25 and 105.1 million at 10/18/24. Against them, 92.9 million is 11.5 per cent below the year-ago level and 11.7 per cent below the two-years-ago level, both percentages the publisher's own rather than the Institute's arithmetic. The claim this row has carried for three weeks and declined to assert — that the grade going into trucks is scarcer than the category containing it — is the comparison of those figures against the total row's 9.7 and 8.4 per cent. It is true on both bases, by about two points against last year and better than three against the year before. 🔴 AND THE RESTORATION BUYS SOMETHING THE WEEKLY SHARE ARITHMETIC ABOVE CANNOT, WHICH IS THE ONLY REASON IT IS WORTH MORE THAN A NUMBER. The paragraphs above spend their length on a share that moved nineteen hundredths of a point and decline to read anything into it. The publisher's two bases let the same share be computed on a two-year reach: this grade was 92.4 per cent of the category two years ago, 90.8 per cent a year ago and 89.1 per cent now. That is a fall of about three and a quarter points over two years, in one direction, measured at three points on the publisher's own clock. A weekly share move of two tenths is noise and is treated as noise here; a three-point move across two annual observations is not the same object, and the reference prints it as the one thing in this row that a single week cannot produce. It is still a level comparison and not a cause: nothing here says whether the shift is demand, refinery configuration, or a change in what gets classified where. SUPERSEDED AND LABELLED: the first draft of this entry, written on this date, read that the comparison stayed “WITHDRAWN AND LABELLED” because “the publisher's matched year-ago column prints the distillate total and does not break it into sulphur grades.” That was a correct reading of the highlights page and a wrong claim about the release. The total row above carries the amendment and the reason.
— of which greater than 15 to 500 ppm sulphur	4.2	4.0%	Off-road and marine grades, and this week the only line in the category that FELL. It DREW 278 thousand barrels from 4,517 to 4,239 — a 6.2 per cent draw against a 1.6 per cent build in the total, and the sole line running against a week in which both larger grades rose. 🔴 THAT IS THE EXACT INVERSE OF LAST WEEK IN THIS ROW, AND THE INVERSION IS THE REASON THE PRIOR ENTRY'S RESTRAINT WAS RIGHT. Last week this grade was the only line that rose, against a category that drew; this week it is the only line that falls, against a category that builds. The prior entry declined to read a 75-thousand-barrel countermove as a signal about off-road or marine demand and said so in terms. Had it read one, the reading would have reversed within seven days — and it would have reversed on a grade whose own direction flipped while the thing it was supposedly telling the reader about cannot plausibly have flipped with it. A line that moves opposite to its category in two consecutive weeks, in two opposite directions, is a small series inside a larger one, not a leading indicator of anything. 🔴 AND THE MOVE IS LARGER THAN LAST WEEK'S IN BOTH ABSOLUTE AND PROPORTIONAL TERMS, WHICH DOES NOT MAKE IT MORE MEANINGFUL. Two hundred and seventy-eight thousand barrels is 3.7 times last week's 75, and 6.2 per cent is 3.6 times last week's 1.7 per cent. The Institute records the magnitude because the magnitude is measured. It declines to treat a bigger move in a four-million-barrel series as better evidence than a smaller one, because this row's own history is a series of moves that each looked larger than the last and none of which resolved into a direction. 🔴 THE COMPARISONS ARE RESTORED, AND THIS IS THE ROW WHERE THE PUBLISHER'S OWN PERCENTAGE IS WORTH THE MOST, BY A WIDE MARGIN. Table 4 prints 3.6 million barrels at 10/17/25 and 2.4 million at 10/18/24 for this grade. Against them, 4.2 million is 17.8 per cent ABOVE the year-ago level and 78.5 per cent above the two-years-ago level — both figures the publisher's, computed from the thousand-barrel numbers behind the printed millions. Re-derived off the rounded millions the way a reader would, 4.2 on 3.6 gives 16.7 per cent and 4.2 on 2.4 gives 75.0, so a re-derivation would be wrong by 1.1 and 3.5 points. The total row above sets the general practice and this row is the demonstration of why it is not pedantry: a tenth of a million barrels is one fortieth of this series and one twenty-fourth of its two-year base, so rounding to a tenth is worth two to four points of percentage here against a tenth of a point on the category total. On a small line, the publisher's unrounded percentage is not a refinement of the reader's arithmetic. It is a different answer. 🔴 AND THE YEAR-ON-YEAR DEFICIT NARROWED FROM BOTH ENDS AT ONCE, WHICH IS THE THING THIS ROW EXISTS TO WARN ABOUT. The prior entry carried 25.8 per cent above a year-ago 3.5 million; this entry carries 17.8 per cent above a year-ago 3.6 million. Eight points came off the comparison while the level fell three tenths and the base rose one tenth, so the level and the base both pushed the same way and neither alone accounts for it. That is a third distinct pattern in three weeks for this row, which has now printed a swing driven entirely by the base, one driven half by the base, and one driven by both ends together. The reference prints the percentage because the publisher measured both ends of it, and repeats that a year-on-year percentage on a four-million-barrel series is a statement about two weeks a year apart and not about a trend. Note also that the two bases are different week ordinals from the prior entry's, so 25.8 and 17.8 are not two readings of one quantity and the eight-point gap is not a change in anything. SUPERSEDED AND LABELLED: the first draft of this entry, written on this date, withdrew these comparisons on the ground that the publisher's year-ago column “does not break it into sulphur grades.” Table 4 does. The total row above carries the amendment and the reason, and the sentence this draft was proudest of — that the Institute would rather print a 278-thousand-barrel draw with no base beside it than a percentage against a base it did not read — is retained as a statement of principle and did not apply, because the base was there to be read.
— of which greater than 500 ppm sulphur	7.2	6.9%	Heating oil and industrial grades. It BUILT 652 thousand barrels from 6,515 to 7,167 — a 10.0 per cent build, and the fifth reversal this row has recorded in five weeks. 🔴 THE SEQUENCE IS NOW BUILD, BUILD, DRAW, BUILD, DRAW, BUILD, AND THE COMMITMENT TO PRINT IT WITHOUT INTERPRETING IT IS KEPT FOR A SECOND WEEK. Two entries back the reference read a second consecutive build as the category “building into the season it exists for”; the next entry withdrew that and warned that “a direction repeated twice is not yet a season”; the entry after that concluded that a six-to-seven million barrel series carries weekly moves of a tenth of itself and that no seasonal reading is available from weekly prints at all. This week's move is a tenth of the series almost exactly — 10.0 per cent — which is the clearest possible demonstration of that conclusion rather than an exception to it. The move is a 652 thousand barrel build, it is the second largest contribution to the category's build, and the reference proposes nothing about what it means. 🔴 AND THE SEQUENCE ITSELF IS WORTH ONE SENTENCE OF ARITHMETIC, BECAUSE THE ALTERNATION IS WHAT A READER WILL WANT TO READ SOMETHING INTO. Six prints alternating in direction four times is not rare: a series with no memory at all produces a reversal in any given week about half the time, so four reversals in five chances is close to what chance delivers and well inside it. The row is noting that the pattern a reader sees is the pattern an absence of pattern looks like. 🔴 AND IT IS THE LINE THAT RESTORED AN OFFSET THIS WEEK, IN THE OPPOSITE SENSE TO THE WEEK BEFORE. Two weeks ago this grade's 0.7 million build was the exact counterpart to the trucking grade's 0.7 million draw, which let a flat total be described as barrels moving between grades rather than leaving. Last week it fell alongside the trucking grade and the offset broke. This week it rose alongside the trucking grade. There is no offset again, in the other direction: the movement entered the category rather than shifting inside it. 🔴 THE YEAR-ON-YEAR COMPARISONS ARE RESTORED, AND THIS IS THE ROW WHERE THE RESTORATION CHANGES THE SIGN OF THE ANSWER. Table 4 carries matched columns for this grade: 7.0 million at October 17, 2025 and 6.3 million at October 18, 2024, with the publisher's own percentages of 2.4 and 13.4 per cent ABOVE. The prior entry read 6.3 per cent BELOW a year-ago 7.4 million. Both ends moved the same way — the base fell four tenths of a million while the level rose 652 thousand barrels — so the comparison did not narrow, it crossed zero. A reader holding the old number and not the new one would have the direction wrong rather than the magnitude, which is the clearest argument in this ledger for re-measuring both ends of a year-on-year figure every week instead of carrying it. Re-derived off the rounded millions the way a reader would, 7.2 on 7.0 gives 2.9 per cent and 7.2 on 6.3 gives 14.3, so a re-derivation would be wrong by half a point and by nine tenths of a point; the practice stated in the total row governs here and the publisher's unrounded percentages are the ones printed. SUPERSEDED AND LABELLED: the first draft of this entry, written on this date, withdrew these comparisons on the ground that the publisher's matched year-ago column “carries the distillate total and no sulphur-grade breakdown.” That was a correct measurement of the stocks table on the highlights page and it is false about the release. The total row above carries the amendment, the discriminator, and the standing rule it produced — that a provenance withdrawal names the table it was measured against and not the publisher.

Source: EIA Weekly Petroleum Status Report, Table 4, published October 21, 2026. Shares are of the total distillate figure in the first row, so the three sub-categories sum to the total and their shares sum to 100 per cent. Sub-15-parts-per-million material is on-highway diesel.

On-highway diesel is 89.1% of total distillate stocks — 92.9 million barrels of 104.3. So in current practice a distillate headline is mostly a diesel headline, and a reader who treats the two as interchangeable will usually be near enough right.

The reason to know the split anyway is that the fraction which is genuinely not diesel — the 7.2 million barrels above 500 parts per million, heating oil and industrial grades — is the fraction that behaves differently in winter. Heating demand competes with freight demand for the same refinery output, which is why the New York Harbor heating-oil crack and the diesel crack move together and why a cold northern winter is a freight-cost event as well as a heating event.

Where the official forecast expects that stock to go

The observed stock above is 102.6 million barrels, for the week ended October 9, 2026. 🔴 A TESTABLE FORECAST HAS NOW BEEN TESTED, AND IT DID NOT HAPPEN. The EIA's September 9, 2026 outlook forecast that distillate inventories would drop below 100 million barrels in September 2026 and stay beneath the 2021–2025 five-year low through much of 2027. September 2026 is over. The weekly series printed 107.4, 105.2 and 105.1 million across its closing weeks and never approached 100 million. That is recorded here because the Institute printed the forecast when it was a forecast and owes the reader the outcome — a forecast quoted only while it is unfalsifiable is not being used honestly. 🔴 AND THE FIRST OCTOBER PRINTS MOVE TOWARD THE CALL WITHOUT REDEEMING IT. The week ended October 9 printed 102.6 million barrels, the lowest level this reference has carried and 2.5 million below the week before it, so the direction the September forecast described has resumed. That does not retroactively make the forecast right. The call named September 2026 and a sub-100-million level, and every weekly print inside September closed above 105 million. A dated monthly forecast is tested on its own month, and a later move in the forecast's direction is a separate fact the Institute reports as one rather than folding it back into the original call. Two caveats that cut in the forecast's favour and are stated rather than omitted: the weekly series and the monthly series EIA forecasts are not the same series, and a monthly average can differ from the weekly prints inside it; and the five-year-low half of the call is a different claim from the 100 million half and is not addressed by these observations. 🔴 THE AGENCY HAS MOVED IN THE SAME DIRECTION THE OBSERVATION DID. The October 6, 2026 edition raises the forecast for U.S. distillate closing stocks to 113.3 million barrels for 2026 from 109.1 in the September edition, and to 122.1 million for 2027 from 116.0. Those are revisions of 4.2 and 6.1 million barrels in one month, both upward, which is the agency conceding that the tightness it forecast in September has not arrived on the schedule it expected. The Institute has read those figures in the release's own comparison of the current and previous editions; it has not read whether the October narrative restates the below-100-million call, and does not assert either way. The forecast and the observation run on different clocks, and the gap has REOPENED to eight days: the stock is measured on the week ended October 9, 2026 and the forecast was still completed on October 1, 2026. 🔴 SUPERSEDED AND LABELLED, because this sentence stated a CONSEQUENCE and the consequence expired without anybody touching it. Until this week it read that the two clocks were “within a day of each other, for the first time in this reference's history”. That was true on the week ended October 2 and was the closest alignment this ledger has recorded. Nothing about the forecast changed: the observation advanced one week, the gap widened by seven days, and it will widen by seven more every week until the November edition is published. A two-clock gap is not a property of a pair of documents. It is a measurement that moves whenever either end moves, and the end that moves every week is the observation. The October forecast accounts for the 40 million barrels of Strategic Petroleum Reserve exchanges announced September 29, 2026 and does not account for the G7 announcement of October 2, 2026, which post-dates its completion by one day. The agency attributes the tightness to global distillate production running below last year's levels and to export incentives pulling barrels out of the domestic market, which is the same yield mechanism described below rather than a separate cause. It also cites East Coast distillate inventories running 32 per cent below the five-year seasonal average — a regional reading far tighter than the national one, and the figure that keeps the tightness case alive even as the national forecast is revised up. Read the 2027 price forecast accordingly: it does not assume inventories recover, only that they stop falling, and the stock forecast being raised while the price forecast is also raised is the agency saying it expects more barrels at higher prices rather than relief.

Production against consumption, and where the surplus went

Weekly flow	Figure	Measured	Source	What it carries
Distillate fuel oil production	5,232 thousand bbl/d	week ended October 16, 2026	EIA Weekly Petroleum Status Report, Table 3 and the weekly refiner and blender net production series, published October 21, 2026	What refineries made. Refiner and blender net production, UP 43 thousand barrels a day on the week from 5,189, recovering 43 of the 101 thousand barrels a day the series gave up last week and leaving it 58 below the 5,290 of two weeks ago. 🔴 THE FIGURE PRINTED IS THE PUBLISHER'S OWN TOTAL, AND THIS WEEK IT DISAGREES WITH BOTH COMPONENT SUMS BY ONE THOUSAND BARRELS A DAY. Three of the publisher's own figures bear on the same quantity. The weekly series total reads 5,232. Table 3's two legs — refiner net production 5,201 and blender net production 32 — sum to 5,233. Table 3's three sulphur grades — ULSD 5,010, the greater-than-15-to-500 ppm grade 153 and the greater-than-500 ppm grade 70 — also sum to 5,233. The published total is the one printed above, because it is a number the publisher prints rather than a number the Institute adds, and because the prior week's 5,189 is the same series, so both ends of the weekly change sit on one basis. The one-thousand gap is component rounding and nothing else: Table 3 carries the note that differences and percent changes are calculated using unrounded numbers, which is the same instruction read the other way. 🔴 AND THE PATTERN OF THAT GAP MOVED, WHICH IS A MEASUREMENT RATHER THAN AN IMPROVEMENT. Last week the grade sum of 4,945, 135 and 109 came to 5,189 and agreed with the published total exactly, while the two legs summed to 5,190 and sat one above it. This week BOTH component bases sit one above the published total. Last week's entry called the clean grade identity “the cleanest identity in this ledger this week”; it was clean and it was also a coincidence of rounding, and a reader who took it as a property of the release would be wrong this week. Three bases that agree to within one thousand barrels a day are three bases, not one. 🔴 THE GRADE STORY IS THE EXACT INVERSE OF LAST WEEK'S, AND THE SYMMETRY IS A REASON TO SAY LESS ABOUT IT RATHER THAN MORE. ULSD rose 65 thousand barrels a day from 4,945; the greater-than-15-to-500 ppm grade rose 18 from 135; the greater-than-500 ppm grade FELL 39 from 109. The two high-sulphur grades together fell 21. Last week the whole of the decline was ULSD, down 121, while the two high-sulphur grades together rose 21 — the same 21 thousand barrels a day with the sign reversed. A two-week reversal of that precision in a weekly estimate is more likely to be the series breathing than a refinery reconfiguring, and last week's entry is retained rather than rewritten on exactly that reading. 🔴 THE GENUINELY NEW READING IS AT GRADE LEVEL, AND IT IS THAT TWO OF THE THREE GRADES MOVED OPPOSITE WAYS IN STOCK AND IN PRODUCTION. Set the stock changes in the rows above beside the production changes here. ULSD stocks built 1,300 thousand barrels on production up 65 thousand barrels a day — same direction. The greater-than-500 ppm grade built 652 thousand barrels on production DOWN 39 — opposite. The greater-than-15-to-500 ppm grade drew 278 thousand barrels on production UP 18 — also opposite. Those are different units and are not being subtracted from one another. What the signs establish is narrower and is worth stating: the grade-level stock changes this week are not production-driven, because in two of three grades production moved against the tank. Last week ULSD moved the same way in both and this row read the two together; that reading is not available this week and is not being stretched to fit. 🔴 THE YIELD READING POINTS UP, WHICH IS THE THIRD SIGN FLIP IN THREE WEEKS AND IS NOW THE FINDING. Gross inputs to refineries FELL to 16,860 thousand barrels a day from 16,871, down 0.07 per cent, with weekly operable utilisation at 93.4 per cent against 93.5 — figures read off this same release and carried in the refinery row of the inventories ledger above, so both ends of this comparison sit on one clock. Distillate production rose 0.8 per cent against throughput that was flat. The implied distillate yield is 31.0 per cent of gross inputs against 30.8 per cent a week ago, up about three tenths of a point. Two weeks ago this row recorded a yield shift up, last week one down, and this week one up again. Three opposite readings in three weeks retires the yield reading as a weekly signal in this reference: it is reported because it is derivable from figures on one clock, and it should not be read as evidence about how refineries are configured. 🔴 THE YEAR-ON-YEAR WITHDRAWAL IS PARTLY LIFTED, AND THE PART THAT STAYS WITHDRAWN IS A PERCENTAGE THE PUBLISHER DECLINES TO PRINT. Last week's entry withdrew the comparison on the ground that the year-ago base sits in Table 9, which the Institute had not read. Table 3, which the Institute did read on this date, carries matched historical columns for each leg separately with the publisher's own percentages. Refiner net production: 5,201 against a year-ago 4,598, up 13.1 per cent, and against a two-years-ago 4,977, up 4.5 per cent, with the four-week average at 5,151 against 4,806, up 7.2 per cent. Blender net production: 32 against 34 a year ago, down 5.8 per cent, and against 34 two years ago, down 6.9 per cent. Table 3 prints NO combined total and NO combined percentage for the two legs together, so none is printed here. The combined figure could be added and the combined percentage could be derived, and both would be the Institute's arithmetic rather than the publisher's measurement; the same governing rule that keeps a release the Institute has not read out of this ledger keeps a percentage the publisher does not publish out of it as well. The withdrawal that remains is therefore narrower than last week's and rests on a different ground, and that ground is named: it is Table 3, not the release.
Distillate fuel oil product supplied	3,732 thousand bbl/d	week ended October 16, 2026	EIA Weekly Petroleum Status Report, U.S. weekly product supplied series, published October 21, 2026	EIA's proxy for domestic consumption, computed as a disappearance residual. It FELL 591 thousand barrels a day from 4,323, down 13.7 per cent in one week, giving back about seven eighths of last week's 673 thousand barrel rise. 🔴 THE SIX-PRINT SEQUENCE IS NOW THE WHOLE OF WHAT THIS ROW HAS TO SAY, AND THAT IS THE THIRD TIME IT HAS BEEN WORTH SAYING. The sequence reads 3,501, 3,975, 3,948, 3,650, 4,323 and 3,732. Two weeks ago this row read the first four prints as “a two-week episode and not a level shift”. Last week it recorded that the fifth print overtook that account, standing 822 thousand barrels above the fourth and 348 above the highest of the first four. The sixth print sits back inside the range the first four occupied. Every directional account this row has offered has been overtaken by the next print, and all of them are retained rather than rewritten, because the pattern of being overtaken is now better evidence than any one of them was. The statement that has survived all three episodes is the standing one and it is the only one carried forward: product supplied is a residual, not an observation, and it carries whatever error the other terms carry. 🔴 THE EXPORT-AND-STOCK-BUILD CHANNEL WIDENED, AND IT WENT INTO TANKS RATHER THAN ONTO SHIPS. Production less product supplied WIDENED 634 thousand barrels a day, from 866 to 1,500. Exports moved six thousand barrels a day, from 1,373 to 1,379 — which is to say they did not move — while imports fell 28, from 145 to 117. So the widening is not an export story, and the stock rows above say where it went: distillate inventories BUILT 1,675 thousand barrels, against a draw of 2,527 thousand last week. 🔴 AND THE BALANCE IDENTITY CLOSES TO ONE THOUSAND BARRELS A DAY IN BOTH WEEKS, WHICH RETIRES THIS ROW'S REFUSAL TO RECONCILE — BUT NOT FOR THE REASON IT FIRST APPEARS TO. Net exports of 1,262 thousand barrels a day plus a stock build running at 239 thousand barrels a day across seven days comes to 1,501, against a channel of 1,500. Run the same arithmetic on last week: net exports 1,228 against a stock draw running at 361, giving 867 against a channel of 866. Last week's entry said the two sides “are not required to match and are not being reconciled here — exports, the crude adjustment and the residual's own error all sit between them.” Measured, they sit between them to within one thousand barrels a day in each of two consecutive weeks. That hedge is therefore SUPERSEDED AND LABELLED. What replaces it is not a stronger claim but a narrower one: product supplied is DEFINED as the residual of production, imports, exports and the stock change, so the identity closing is very close to definitional and is not independent evidence about demand. What it does establish is worth having, and it is the only arithmetic gate this ledger has on a week's reads: five figures taken off the same release reconcile to one thousand barrels a day, so none of them was transcribed wrong and none came off a different week. Had any one of the five been a stale print, this identity would not have closed. 🔴 DAYS OF SUPPLY ROSE TO 26.7 FROM 25.8, AND IT IS NOT COMPUTED ON THE FIGURE PRINTED ABOVE. The publisher's days-of-supply figure divides the stock by the FOUR-WEEK AVERAGE of product supplied, not by the weekly print: 104,286 thousand barrels over a four-week average of about 3,913 gives 26.7, where the weekly 3,732 would give 27.9. That is the reason to quote the publisher's number rather than derive one, and it is also the reason days of supply is the steadier of the two readings — it is not whipsawed by the weekly variance this row has now been wrong-footed by three times. Both ends of the rise are real: stocks built and the four-week demand average fell. The year-on-year and four-week-average comparisons for product supplied remain WITHDRAWN AND LABELLED, and the ground is restated here rather than deferred, because the ground the row above gave has been lifted and a withdrawal that points at a lifted reason is no withdrawal at all. The series the Institute read on this date is the weekly supply estimates table, which carries six consecutive weeks and no matched year-ago or four-week-average column for this line. That is a property of the table read, not of the release: “not in the table we read” and “not published” are different claims, and only the first is the Institute's to make.

Refineries made 5,232 thousand barrels a day of distillate and domestic consumption absorbed 3,732. The difference, 1,500 thousand barrels a day, went to exports and to the stock build. In a week of near-record margins that is where the barrels went, and it is the mechanism by which a tight European or Latin American market pulls American product out of the country while American stocks look adequate.

Total distillate stocks of 104.3 million barrels against domestic consumption of 3,732 thousand barrels a day is 27.9 days of supply on the Institute's computation. Days of supply is the figure to use rather than the raw stock level, because a stock number in isolation says nothing about adequacy: the same tank farm is comfortable against weak demand and alarming against strong demand.

One caution on the consumption figure. “Product supplied” is not a measurement of consumption. It is a disappearance residual — what came out of the system minus what went into it — and it carries every measurement error in the surrounding accounts, in the same way the “distribution and marketing” line on the barrel-to-gallon ledger absorbs every error in the pump-price split. Read it as a good weekly proxy and not as a meter reading.

Why diesel costs more than gasoline, and how little of it is tax

A gap of $1.992 a gallon in the same week. The received explanation is tax, and tax is 4.1% of it.

Component of the diesel-over-gasoline gap	Cents per gallon	Share of the gap	Measured
Total gap at the pump — diesel $6.276 less regular gasoline $4.284	199.2¢	100%	week ended October 19, 2026
of which tax — diesel 59.90¢ less gasoline 51.67¢	8.23¢	4.1%	January 2026
of which everything else — crude grade, refinery yield, refining margin, distribution	191.0¢	95.9%	two clocks — see note

Retail prices: EIA, Weekly Retail Gasoline and Diesel Prices, U.S. average, all formulations; released October 20, 2026. Tax figures: EIA, Gasoline and Diesel Fuel Update — federal and state motor fuel taxes, stated by EIA as of January 2026, stated as of January 2026, and the state lines are volume-weighted averages excluding gross-receipts taxes and local levies. The two clocks in this table are the September 2026 retail week and the January 2026 tax schedule; the tax figures are fixed in cents and are updated roughly annually, so the mismatch is small but it is real and it is stated.

No federal fuel tax holiday is in effect. Congressional Research Service report R48948, dated May 15, 2026, records proposals only. The federal rate has not moved, which is precisely why the tax percentage of the pump price has fallen for four consecutive months while no tax was cut.

Diesel does carry more tax than gasoline — 59.90¢ against 51.67¢, a difference of 8.23¢ a gallon, federal and state combined. That difference is real and it is almost the whole of the conventional explanation for why diesel is dearer. It accounts for 4.1% of the current gap.

The remaining 191.0¢ a gallon is the market, and it is what the rest of this page is about. In an ordinary year the tax differential is a large share of a small gap and the conventional explanation works. In this episode the gap is several times its normal size, the tax term has not moved at all, and therefore essentially the entire widening is refining and crude-grade economics.

The yield constraint, which is the actual mechanism

Diesel is structurally tighter than gasoline in this episode and the reason is a refinery yield constraint rather than a demand surge. A barrel of crude yields a fixed range of product cuts, and shifting yield toward distillate is limited by the configuration of the plant and the grade of the crude. The crude that has been removed from the market is disproportionately medium and heavy sour, which is the crude that makes distillate best. So the market lost the barrels most suited to the product in shortest supply, and no amount of refinery utilisation fixes a yield problem. That is the mechanism behind a distillate crack running roughly 74 cents a gallon above gasoline since March.

The price evidence is on the same week's spot legs. The New York Harbor distillate crack computes to $111.20 a barrel against a gasoline crack of $24.85 — distillate out-earning gasoline by 205.6¢ a gallon, or $86.35 a barrel, from the same crude in the same week. Every leg and the full computation are on the refining ledger.

A margin that size normally calls forth supply within weeks. It has not, and the reason is that the constraint is physical. A refinery's yield can be shifted toward distillate only within a range set by its units and by the crude it is fed, and the crude removed from this market is disproportionately medium and heavy sour — which is the crude that makes distillate best. The market lost the barrels most suited to the product in shortest supply. No amount of running the plants harder fixes a yield problem.

That is also why the seaborne product trade figures matter more than usual here. Diesel exports from Russia, the Middle East and Asia fell about 1.3 million barrels a day year on year in July 2026, roughly a fifth of global seaborne diesel trade. A shortfall in traded product is harder to arbitrage away than a shortfall in crude, because the substitute has to be manufactured rather than merely shipped.

The record and this week’s price — which are no longer the same number

Diesel printed $6.276 a gallon in the week ended October 19, 2026, up 9.7¢ from $6.179 the week before. The run of three consecutive nominal records ended in the week ended September 28, 2026 and has not resumed; this week’s price remains below the high. The record is $6.529, set in the week ended September 21, 2026, and it stands. It is the highest figure in the weekly all-types series, which begins in March 1994; this week’s print is not, and the two carry different measurement dates for that reason.

The nominal high on the weekly all-types diesel series (EMD_EPD2D_PTE_NUS_DPG, which begins March 1994) is $6.529, set in the week ended September 21, 2026, and that is the figure the real-versus-nominal comparison measures. The run that set it has ended: the series took a record at $5.967 in the week ended September 7, 2026, broke it at $6.285 a week later, broke it again at $6.529 on September 21 — three consecutive weeks, each record standing exactly one week — and then fell 14.7¢ to $6.382 in the week ended September 28. So the record and the current print are no longer the same number, and this note is about the record. 🔴 ADVANCED October 20, 2026: the record has now stood for FOUR consecutive weeks, against one week each for the three prints that set it — and the run of falls beneath it has ENDED. The current print is $6.276, which puts the record-to-print gap at 25.3¢. That gap widened in each of the three weeks after the run ended and NARROWED this week, by exactly the print's own weekly rise, because the record is a fixed number and the gap can move by nothing else. Both figures in this sentence are derived from DIESEL_RECORD_NOW and from the current RETAIL_WEEKLY row rather than typed, which is the reason those two constants were split in the first place, and it is why this sentence needed no arithmetic edit when the sign of the weekly change flipped — only its prose. 🔴 SUPERSEDED AND LABELLED RATHER THAN DELETED, per the Two-Clock rule — read as of October 15, 2026, when the tokens resolved to $6.179 and 35.0¢: the record had then stood for THREE consecutive weeks and the series had fallen in every week since the run ended, and the gap was 'the widest it has been since the run ended, and widening in the same direction each week.' The first clause was true and the second is now false. The last high set before that run was $5.810 in the week ended June 20, 2022, and that is what the deflated comparison starts from. EIA changed the methodology of the diesel series on June 13, 2022 — before the 2022 comparator week — so the two readings sit on the same side of the break and the comparison is clean. This is a nominal record only; see the real-versus-nominal ledger, where it is not one.

A nominal record and not a real one, and the difference is $0.078 a gallon

The prior nominal high of $5.810 was set in the week ended June 20, 2022. Deflated onto September 2026 dollars using the June 2022 CPI-U it is $6.607 — above the nominal record and further above this week’s price. From the $6.276 printed in the week ended October 19, 2026, diesel would have to rise a further 5.3% to set a record in real terms; from the $6.529 high of the week ended September 21, 2026 the distance was 1.2%. The percentage moves with the price and the record does not, which is why the two are stated separately.

So “record diesel prices” is accurate about the series and misleading about the burden, and this reference says both things in the same sentence wherever it says either. The full deflation, the CPI sources and the same treatment applied to the gasoline records are on the real-versus-nominal ledger.

Why diesel is the product to watch

The Institute view

Diesel is the product to watch and gasoline is the product being reported. The distillate crack has run above the gasoline crack since March 2026 and the same-week computation puts the gap at $86.35 a barrel. Diesel is also the fuel with the broader transmission into general prices, because it is a cost of moving goods rather than a cost of commuting. A commentary cycle organised around the gasoline pump price is watching the smaller variable.

This is a yield problem, and yield problems do not respond to price. A crack of that size would ordinarily clear itself within a quarter. The constraint here is the configuration of the plants and the grade of the available crude, and it has held since March. Anyone modelling a resolution should be modelling the return of medium and heavy sour crude rather than a refining response.

The official forecast implies a resolution that has not started. The current outlook (forecast completed October 1, 2026) carries retail diesel at $5.19 for calendar 2026 and $4.49 for 2027, against a weekly print of $6.276 in the week ended October 19, 2026. The 2026 average is arithmetically reachable because January and February were low. The 2027 figure is a statement that the disruption ends, and the Institute prints it as a forecast rather than as a finding.

The rest of the reference

Hub

The Global Oil Reference

Prices, the supply and demand balance, refining, the two-clock rule and the plain-English glossary that serves all 13 ledgers.

Back to the hub →Ledger

The Barrel-to-Gallon Bridge

Where every cent of a $4.48 gallon goes — crude, refining, distribution and tax — with each of the four components on its own clock and three of the four defined differently from how a reader assumes.

Open the ledger →Ledger

Real and Nominal Prices Since 1950

Seventy-six years of crude prices deflated to constant September 2026 dollars. The 2008 records are still records in real terms and today's are not close.

Open the ledger →Ledger

The Barrel in 29 Currencies

One central bank, one date, one fixing time. A dollar barrel converted into 29 currencies without mixing thirty sources and thirty clocks.

Open the ledger →Ledger

The Forward Curve: Contango and Backwardation

What the shape of the curve does to inventory and why it drives storage behaviour rather than describing it — then the full strip: 65 listed WTI months from the exchange's own Daily Bulletin, $97.97 front against $80.13 a year out, $17.84 of backwardation, open interest printed beside every settlement.

Open the ledger →Ledger

Diesel and Distillates, on Their Own

Distillate fuel oil is not diesel, and the tables are written in the former

The sulphur split — how much of the distillate headline is diesel

Production against consumption, and where the surplus went

Why diesel costs more than gasoline, and how little of it is tax

The yield constraint, which is the actual mechanism

The record and this week’s price — which are no longer the same number

Why diesel is the product to watch

The rest of the reference

The Global Oil Reference

The Barrel-to-Gallon Bridge

Real and Nominal Prices Since 1950

The Barrel in 29 Currencies

The Forward Curve: Contango and Backwardation

Breakevens: Wellhead and Fiscal

Proven Reserves by Country

Production & Spare Capacity

OPEC+ Quota Against Actual

Rig Counts

Refining Capacity, Utilisation & Cracks

Inventories & the Strategic Reserve

Chokepoints & Disruptions