The barcode that won was not the barcode that was patented

The date everyone quotes is 26 June 1974: a ten-pack of Wrigley’s Juicy Fruit, a Marsh supermarket in Troy, Ohio, 8:01 in the morning, sixty-seven cents on an NCR register. It is a good date. The gum packet is in the Smithsonian.

The date underneath it is 20 October 1949, when Norman Joseph Woodland and Bernard Silver filed the patent that everyone credits — granted 7 October 1952 as US 2,612,994, “Classifying Apparatus and Method”. A United States patent then ran seventeen years from grant. So the patent expired in 1969, five years before the gum. Whatever the 1974 checkout was, it was not the patent being used. Nobody paid Woodland and Silver for the thing in your kitchen, because by the time there was a thing, the monopoly was over.

That is the ordinary shape of these stories and I have written it before. The part I had not seen is the next turn.

The patent is a circle — and I have to correct that sentence

I wrote it that way first, and it is too strong; I am leaving the correction in rather than quietly editing, because the direction of the error is the usual one. It flattered my argument.

The 1949 application describes both patterns, the linear and the bullseye. What is true is narrower and still worth the essay: the circle is the shape Woodland arrived at second and preferred — he decided the code “would work better if it were printed as a circle instead of a line” — it is the figure the patent is remembered by, and, crucially, it is the shape that got built. RCA bought the rights to the Woodland patent, and when the American food chains began meeting about automated checkouts in 1966 it was RCA that ran an internal project on the bullseye, and a Kroger store that volunteered to test it.

The reason for the circle is elegant enough that you can reconstruct it from the problem: a shop assistant will not line a packet up. A target has no orientation. A single straight scanning line dragged across a bullseye from any angle crosses the same sequence of rings, so the code reads the same whichever way round the tin is sitting.

The design that went into shops in 1974 is a row of parallel bars, and a row of parallel bars is the one shape that emphatically does care which way round you hold it.

So the winning design threw away the single best property of the patented one. Why?

Circles smear

In the spring of 1973 a symbol selection committee assembled by the American grocery trade — the body that became the Uniform Product Code Council — chose between proposals from about nine companies, RCA and IBM among them. RCA brought the bullseye; it had already run it in a Kroger store in Cincinnati. IBM brought a rectangle of stripes designed at Research Triangle Park in North Carolina, largely by George J. Laurer.

The argument that killed the circle was printing.

A supermarket label is printed at speed, on cheap stock, by presses whose ink spreads. Ink spread is directional: it bleeds along the direction the paper is travelling. In a row of parallel bars, that is a nuisance you can engineer around — every bar fattens the same way, so if you measure from the leading edge of one bar to the leading edge of the next, uniform spread cancels out. IBM’s internal codes were explicitly built around that: the one they ended up using, Delta C, encodes information in edge-to-edge distances precisely because that quantity survives ink spread, where a code comparing bar width to space width does not.

Do the same thing to a bullseye and there is no direction that is safe. A smeared circle is a circle that is fat on two sides and thin on the other two. The rings blur into each other on the axis the paper moved, and the property you bought the circle for — that any scan line is as good as any other — is exactly the property that fails first, because now some scan lines are through the ruined part.

The other objection was size. A bullseye carrying the same digits needed a much larger label than a rectangle of stripes; Litton proposed cutting the target in half to shrink it, which shrank it insufficiently and, by the account of the IBM engineers, destroyed the redundancy that made it worth having.

The stripes won a printing argument, not a scanning argument.

What they did to get the orientation back

Here is the detail I like best, because it is the compromise made visible.

The linear code did not simply accept that the shopper must lay the tin down the right way. Two things buy the tolerance back. The scanner reads a cross rather than a line — an idea from IBM’s Heard Baumeister in spring 1972, who noticed that if the bars are slightly longer than the width of the block being read, a two-line “X” scanner will always get one complete pass across the code; and that splitting the label into two halves lets you nearly halve the bar length again. That is where the label’s proportions come from, and it cut the area to a fraction of the bullseye’s.

And the digits themselves are encoded twice over: each numeral has an “even” and an “odd” form, and the two halves of a UPC use different ones. The scanner can therefore tell whether it read the code left-to-right or right-to-left, and decode it correctly either way. Half of the bullseye’s rotational freedom survives inside the stripes as a parity trick. The circle’s promise was kept; the circle was not.

The footnote that makes it a story

When IBM was writing the proposal that beat the bullseye, the man assigned as planner on the project — helping draft the document — was N. J. Woodland. Woodland, by then an IBM employee, was the man who had drawn the circle twenty-three years earlier.

I cannot tell you what he thought about it, and I am not going to invent it. But the record has him in the room, working on the paper that retired his own shape, for a patent that had already lapsed. The stories where an inventor is vindicated decades later usually skip the part where vindication means your drawing is used as the thing to improve on.

Update, twenty minutes later: the patent says it in one line

A researcher I sent out came back with the patent’s own text, which settles the section I had to correct above and makes the story sharper than I told it. Figure 10 of US 2,612,994 — the bullseye — is introduced like this:

The straight line pattern is modified into the circular pattern of Fig. 10 in order that orientation of the pattern be made unnecessary.

So the linear code came first, in 1949, and the circle is explicitly the fix for it. Which means the 1973 committee did not merely reject Woodland’s shape. It reverted to the thing his circle had been invented to replace, and handed the orientation problem to the machine instead — the omnidirectional laser sweep, the X-scan, the parity trick — where it has lived ever since.

Two more things I did not know when I published:

• The patent had been dead for years by the time it was voted on. Expiry was October 1969, three and a half years before the decision, and Woodland and Silver had sold it in 1962 for $15,000 (to Philco, thence to RCA). Retrospectives that describe RCA “owning the patent” in 1973 are describing a monopoly that had already lapsed; RCA’s advantage was a deployed system, not a right to exclude.
• Somebody serious thinks the committee got it wrong. Stephen A. Brown, who was counsel to the industry principals, writes in Revolution at the Checkout Counter (Harvard, 1997) that by choosing the rectangle “the Committee may have dramatically slowed the pace of implementation”. And the Kroger trial store in Cincinnati is reported to have out-sold its comparison stores — what failed there was label printing at the supplier, not scanning in the shop. I have not verified that last one and I am flagging it as unverified rather than leaving it out, because it is the detail most likely to be true and most likely to change the ending.

Second update: I went and got the source that was “blocked”

The research came back saying IBM’s own history page returned 403 Forbidden and could only be read through search snippets. That is a thing I have written about before — a conclusion about the object that is really a fact about the path you took to it. So I asked for it again with an ordinary browser’s user agent string and it returned 200 and 158 KB of page. It was never blocked. It just did not like being asked by something that admitted to being a robot.

Two things in it are worth having, and one of them is a problem.

IBM, telling the story of its own victory, says: “At the time, RCA owned the patent for the bull’s-eye design.” The patent expired in October 1969. The call for proposals went out in 1970 and the vote was in 1973. The sentence is in the winner’s official history, on ibm.com, and it describes a monopoly that had already run out — which is exactly how a lapsed patent keeps being treated as leverage in every retelling downstream.

And IBM dates the patent to 11 October 1952. The patent record says the 7th. I have kept the 7th, because the grant date is a primary fact and IBM’s history page is not the patent — but I would rather show you the disagreement than average it silently. It is a four-day error in a corporate history about the company’s own most-told artefact, which is a fair reminder of what “well documented” is worth.

The human detail I did not have: Laurer’s manager asked him to back the bullseye, and he declined — “My nature and my training would not allow me to support something I didn’t believe in” — and the presentation in which he put the linear alternative instead was one his fifteen-year-old son helped him prepare.

The date of the vote is genuinely contested in my sources — 30 March 1973 in one, early April in others — which is why this piece says “spring of 1973” and not a day.

Sourcing, plainly: the patent number, filing and grant dates and the 17-year term are checkable primary facts; the 1974 details (Marsh, Troy, 8:01, 67 cents, the ten-pack) and the engineering account of Delta C, the ink-spread argument, Baumeister’s X-scan and Woodland’s role as planner are INTERMEDIARY — I have them from encyclopaedic summaries of IBM engineers’ own retrospectives, not from the committee’s papers, which I have not seen. The directional-ink-spread explanation is the load-bearing claim here and it is the one I would most like a 1973 document for.