The 53° collapse that the same numbers draw at 4°

Dark navy board slide, masthead “Alderwick Distribution · Network planning · S&OP review, FY27”, headlined “Destocking has gone vertical — we are draining the network faster than we can refill it”. A subhead reads: the total business inventories-to-sales ratio published by the U.S. Census Bureau, every month from January 2023 to June 2026 — one publisher, one definition, one series, forty-two consecutive monthly observations, every one of them printed on this slide; nothing is smoothed, averaged, indexed or interpolated, and the vertical axis runs 1.20 to 1.50 from top to bottom of the chart, one scale, no break, no change of rate, no second axis; the ratio has not left the 1.2–1.8 band once in thirty-four years. Down the left of the slide, under the label “Inventories / sales · total business · seas. adj. · monthly”, stands a single line chart in a tall, narrow panel roughly two and a half times taller than it is wide. Seven gridlines run across it at 1.20, 1.25, 1.30, 1.35, 1.40, 1.45 and 1.50, each labelled; four tick labels run along the bottom at Jan ’23, Jan ’24, Jan ’25 and Jan ’26. A gold line enters at 1.40, rises to a white dot annotated 1.43 near the top, jitters along between 1.38 and 1.42 for about two years, then turns down and plunges through the lower half of the frame very steeply, at better than fifty degrees, bottoming out below 1.30 and ending on a gold dot annotated 1.30 to its right. A note beneath the chart reads: vertical axis 1.20 to 1.50, seven gridlines, 0.05 apart; horizontal axis 42 consecutive months, evenly spaced; both scales are printed on the chart itself. In the middle column, a table headed “Every observation on the chart · ratio, end of month” lists all forty-two months in three pairs of columns, from Jan ’23 at 1.40 through Jun ’26 at 1.30, with March 2023’s 1.43, May 2026’s 1.28 and June 2026’s 1.30 picked out in gold; a caption notes what is highlighted. Below it a second table, “Quarterly means · the same forty-two observations”, gives 2023 Q1 1.413, Q2 1.420, Q3 1.397, Q4 1.400; 2024 Q1 1.403, Q2 1.407, Q3 1.403, Q4 1.400; 2025 Q1 1.390, Q2 1.387, Q3 1.370, Q4 1.370; and, in gold, 2026 Q1 1.333 and Q2 1.293, under the line “Twelve quarters between 1.37 and 1.42, then two that are not. The FY27 planning year opens on the steep part.” Down the right-hand side, a gold-bordered card carries −0.15 over the line “Fall in the ratio, March 2023 (1.43) to May 2026 (1.28) — and the line is steepest at the end”. A card headed “The slide down · ratio at end of month” lists June 2025 1.39, December 2025 1.36, March 2026 1.32, May 2026 1.28 and June 2026 1.30. A gold card headed “Read-out for the board” says the chart speaks for itself, that for two years the ratio was flat and then it fell off a cliff, that the descent is close to vertical and has not finished, that on this trajectory the network runs below any cover ever planned for, and recommends raising safety stock by eight days across all distribution centres, pulling FY27 Q3 replenishment forward one cycle and re-opening the buffer clause with the three tier-one suppliers. A card headed “Year on year” gives June 2025 1.39 against June 2026 1.30, twelve months and −0.09, six of those nine hundredths in the last two quarters. A card headed “Why one panel” says one series, one page, and that the panel is drawn to fill the space this slide had left for it, which is how every chart in the pack was sized. A final card, “Basis of measurement”, gives source U.S. Census Bureau, Manufacturing and Trade Inventories and Sales; series total business inventories/sales ratio, seasonally adjusted; frequency monthly, end of month; span January 2023 – June 2026, 42 observations, none omitted. Three ticked boxes run along the bottom: “One axis, end to end. The vertical scale runs 1.20 to 1.50 for the whole chart. No axis break, no change of rate partway, no log scale and no second axis anywhere on this slide”, “Nothing left out. All forty-two consecutive monthly observations are plotted and tabulated. Nothing smoothed, nothing averaged, nothing set aside”, and “One publisher, one definition. Census Bureau MTIS, total business, seasonally adjusted — the same series, defined the same way, for every month shown”. A footnote credits the Census Bureau MTIS programme, retrieved via FRED series ISRATIO on 27 August 2026, notes that the June 2026 figure of 1.30 is as published in Census release CB26-132 of 14 August 2026, and states that the slide is a demonstration drawn by us in the manner of an S&OP review pack, has never been published anywhere, and that the ratios on it are real. — Showing the misleading chart

The same series on a cream background, kicked “The same forty-two numbers · the same axis · measured off the drawing · Census MTIS · 2023–2026” and headlined “The line falls at 53.2° on the slide and 4.2° here.” A subhead explains that it is the same series, the same forty-two monthly values and the same vertical axis running 1.20 to 1.50, with not one number moved between the two drawings: all that changed is the shape of the box, which on the slide is a plot 420 × 1040 px and here is 2180 × 300 px — and that a slope is not a property of a series but the vertical exchange rate divided by the horizontal one, which the two drawings run at rates 18.0× apart, a figure neither chart prints because no chart ever does. Across the top, under the label “The same series in a wide box · inventories/sales ratio, total business, seasonally adjusted · axis 1.20–1.50, exactly as on the slide”, runs one very wide line chart with gridlines at 1.20, 1.30, 1.40 and 1.50 and tick labels Jan ’23, Jan ’24, Jan ’25, Jan ’26 and Jun ’26. The gold line is nearly horizontal throughout: from 1.40, up to a dot marked 1.43, almost flat for two years, then a gentle sag to a low marked 1.28 and a finish on a dot marked 1.30. A caption reads that peak to trough is the same −0.15 over the same 38 months, that drawn here it is a 4.2° descent while in the slide’s tall panel the identical pair of points is 53.2°, and that this box is a choice too and not the recommended one — at 7.27 : 1 it flattens the series further than either banking rule below would, which is the argument rather than an exception to it. Below sits a panel headed “One series, four boxes · the same forty-two values and the same 1.20–1.50 axis in every one · plot height fixed at 200 px, width varied”, holding four framed line charts of increasing width, each labelled with its aspect ratio and, in gold beneath, the angle from the March 2023 peak to the May 2026 low: 0.60 : 1 at 42.0°, 1.20 : 1 at 24.2°, 2.35 : 1 at 12.9°, 4.70 : 1 at 6.5°. Its caption notes that nothing in the data distinguishes the four pictures, and that the slide’s box was narrower than any of them. Beneath that, a panel headed “Zero-basing does not fix it · axis 0–1.50” shows two more boxes, one tall and narrow marked 15.0° and one wide and short marked 0.85°, captioned as scaled copies of the two boxes above — 0.40 : 1 and 7.27 : 1 — both now starting at zero: the fall draws 15.0° in one and 0.85° in the other, still 18.0× apart, because zero-basing changes the range and leaves the shape alone, and where an axis begins and how far the two axes are pulled apart are separate decisions of which only the first has a rule. Beside it, a panel headed “The level, read against the whole record · monthly, Jan 1992 – Jun 2026 · axis 1.20–1.80” plots the whole thirty-four-year series: down from about 1.56 in 1992 to a low near 1.24 in 2011, through the 1.3s and 1.4s, a sharp spike to 1.74 in April 2020, then back. A dashed gold rule marks 1.30 and the slide’s window is a shaded strip at the far right. Its caption records 1.30 undercut in 101 of the 414 months since January 1992, with a record low of 1.24 in March 2011. Five cards run down the right. “The two exchange rates” tabulates the slide against this drawing: plot width × height 420×1040 against 2180×300; aspect ratio 0.40 : 1 against 7.27 : 1; pixels per month across 10.24 against 53.17; pixels per 0.01 of ratio up 34.67 against 10.00; peak-to-trough angle 53.2° against 4.2° — noting that dividing pixels-per-hundredth by pixels-per-month in each column shows the slide running its vertical-to-horizontal exchange rate 18.0× higher, which is the whole difference between a cliff and a drift and is printed on neither drawing. “So which shape is correct?” answers that there is none in the abstract: banking to 45° puts the segments a reader should compare near a diagonal, but you must say which segments — this series’ month-to-month wiggle asks for 1.37 : 1 and its 38-month descent for 0.54 : 1, taller than wide and 2.5× from the first answer. “Read the level, not the angle” notes that 1.30 is ordinary — the 2010s ran 1.24 to 1.44 and averaged 1.34, the five pre-pandemic years averaged 1.40 — and that the level is the part of a chart that survives being resized. “Read the level and it says the opposite” gives June 2026 inventories of $2,740.2 bn, up 3.0% on June 2025 — about $80 bn more stock, not less — against sales of $2,111.3 bn, up 10.0%: the ratio fell because the denominator outran the numerator, and nothing was draining. “What the slide got right” lists one publisher, one definition, one continuous series, all forty-two months plotted and every one tabulated, no smoothing, one scale end to end, no axis break, no log axis and no second axis: ten things right, and not one of them says anything about the shape of the box. A gold-ruled callout at the foot reads “Every chart has a shape, and almost nobody chose it,” and observes that a dashboard tile, a column width or a slide template settles every angle on a chart while no caption mentions it; that Cleveland, McGill and McGill gave the one rule there is, banking to 45°, and Heer and Agrawala showed why it cannot be automatic; that the shape is therefore an argument about which comparison matters and belongs in the caption beside the units; and that one habit does the work meanwhile — redraw it at twice the width in your head, since what survives was in the data and what evaporates was in the box. A footnote gives the Census MTIS source and FRED retrieval, the release CB26-132 figures, all forty-two plotted values in order, a note that angles and pixel figures are exact for the published 3,200 × 1,800 images, the two banking calculations with citations to Cleveland, McGill & McGill (1988) and Heer & Agrawala (2006), the dimensions of the zero-based pair, and the statement that both drawings are our own. — Showing the misleading chart

Be the first to star this exhibit

Forty-two months of the U.S. inventories-to-sales ratio, one series, one axis running 1.20 to 1.50, every value printed on the slide — and a line that falls off a cliff. Redraw the identical numbers on the identical axis in a wider box and the same peak-to-trough descent arrives at 4.2° instead of 53.2°. Nothing moved except the shape of the plot, which is the one thing no chart ever labels.

U.S. Census Bureau, Manufacturing and Trade Inventories and Sales — total business inventories/sales ratio, seasonally adjusted, monthly; release CB26-132, 14 August 2026, and the full series via FRED (ISRATIO), 2026Added August 27, 2026

01The claim

Here is the total business inventories-to-sales ratio published by the U.S. Census Bureau, every month from January 2023 to June 2026 — one publisher, one definition, one series, forty-two consecutive monthly observations, and every one of them printed on the slide in a table underneath. Nothing is smoothed, averaged, indexed or interpolated. The vertical axis runs 1.20 to 1.50 from top to bottom of the chart: one scale, no break, no change of rate partway, no log axis and no second axis anywhere. The horizontal axis is forty-two consecutive months at equal spacing, and both ranges are printed on the chart itself. This is a series that has not left the 1.2–1.8 band once in thirty-four years. And the shape is unmistakable. For two years the ratio sat flat between 1.38 and 1.43. Then it fell off a cliff. The quarterly means say the same thing without any help from the drawing: twelve quarters between 1.37 and 1.42, then 2026 Q1 at 1.333 and 2026 Q2 at 1.293. Year on year, June 2025’s 1.39 against June 2026’s 1.30 — twelve months, minus nine hundredths, and six of those nine arrived in the last two quarters alone. The descent is close to vertical and it has not finished. On this trajectory the network runs below any cover we have ever planned to. Recommendation: raise safety stock by eight days across all distribution centres, pull FY27 Q3 replenishment forward one cycle, and re-open the buffer clause with the three tier-one suppliers at the next review.

02The trick

Every ratio on that slide is right, and the axis is not the trick. Draw the identical forty-two numbers on the identical 1.20-to-1.50 axis in a wider box and the cliff becomes a drift. The descent from the March 2023 peak of 1.43 to the May 2026 low of 1.28 is the same −0.15 over the same 38 months in both pictures; on the slide it draws at 53.2°, and in a wide panel it draws at 4.2°. A slope is not a property of a series. It is one exchange rate divided by another — how many pixels the vertical axis gives a unit of data, over how many the horizontal axis gives a step of time — and both are set by the size of the plot rather than by anything in the numbers. Measure them. The slide’s plot is 420 × 1040 px, so one month is 10.24 px across and 0.01 of ratio is 34.67 px up. The wide drawing’s plot is 2180 × 300 px: 53.17 px per month, 10.00 px per hundredth. Divide one pair by the other and the slide runs its vertical-to-horizontal exchange rate 18.0 times higher. That single figure is the entire difference between a collapse and a shrug, and it is printed on neither drawing, because no chart prints it. The axis labels declare the two ranges and never the shape of the box those ranges are drawn in. It is not the truncated axis wearing a new coat, either, and this is the part worth keeping: put both drawings on a scale that starts at zero, holding each one’s shape exactly — 0.40 : 1 and 7.27 : 1 — and the same fall comes out at 15.0° and 0.85°. Still 18.0 times apart, the identical factor as before, because zero-basing changes the range and leaves the shape alone. Where an axis begins and how far the two axes are pulled apart are separate decisions, and only the first has a rule. Line the same numbers up in four boxes of the same height and increasing width and the descent reads 42.0°, 24.2°, 12.9° and 6.5° — four pictures, one dataset, nothing in the data to choose between them, and the slide’s panel narrower than any of the four. Read the level instead of the angle and the finding thins out considerably. A ratio of 1.30 has been undercut in 101 of the 414 months since January 1992; the 2010s ran between 1.24 and 1.44 and averaged 1.34, and the five years before the pandemic averaged 1.40. The record low is 1.24, in March 2011. There is one more thing hiding behind the word “destocking”. The Census release the slide’s series comes from puts June 2026 inventories at $2,740.2 billion, up 3.0% on June 2025 — roughly $80 billion more stock, not less — against sales of $2,111.3 billion, up 10.0%. The ratio fell because the denominator outran the numerator. Nothing was being drained. (The slide is our own demonstration, drawn in the manner of an S&OP review pack. The ratios on it are real, from the Census Bureau.)

03The fix

Choose the shape of the box on purpose, then say what you chose. The oldest rule is still the right place to start: banking to 45°, from Cleveland, McGill and McGill’s 1988 paper on the shape parameter of a two-variable graph, which sets the aspect ratio so that the segments the reader is meant to compare sit near a diagonal, because a slope is judged most accurately around 45° and gets harder to read as it flattens toward 0° or steepens toward 90°. Banking comes with a catch, and the catch is the most useful thing in this exhibit: you have to say which comparison you are banking. Bank this series’ month-to-month wiggle to 45° and the box wants to be 1.37 : 1. Bank the 38-month descent running through it and the box wants to be 0.54 : 1 — taller than it is wide, two and a half times from the first answer, and closer to the slide than to the honest redraw. That is not a defect in the rule; it is what Heer and Agrawala named in 2006, showing that a series carries several scales of variation at once and each one asks for a different box. It is also the plainest possible demonstration that no aspect ratio is correct in the abstract: the shape of the box is an argument about which comparison matters, and it deserves to be made out loud rather than inherited from a slide template. So pick the comparison the chart is for, bank that, and print the ratio in the caption the way you print units on an axis. Then take the weight off the angle altogether. Annotate the levels, put the first and last values on the chart, and write the headline about the amount rather than the steepness, because the amount is the part that survives being resized — “the ratio fell 0.15 over three years, to a level it has undercut in a quarter of the last thirty-four years” cannot be redrawn into something else. Where the same series appears more than once, in a deck or a report or a dashboard, hold its box fixed across all of them, for exactly the reason a small multiple holds its axis fixed: the moment the shape varies, the reader is comparing layouts. And when the container really is out of your hands — a phone screen, a tile in a grid, a figure resized to fill what the page had left — say so and give the numbers, because a chart whose shape is set by a CSS grid is a chart whose slope belongs to the CSS grid. The check costs ten seconds and catches this anywhere. Redraw it in your head at twice the width. If the finding survives, it was in the data; if it evaporates, it was in the box. Any headline resting on the words vertical, cliff, hockey stick or flatlining is a claim about the layout as much as about the world, and it should be read as one.