On June 16, 1887, Frank Primrose, a wool dealer in Philadelphia, wired his buyer in Kansas, William Toland. Telegraph companies charged by the word, so merchants wrote in commercial codes in which one word stood for a whole phrase. In Primrose's codebook, bay meant "I have bought." The message reached Toland as buy.
Toland bought, about 300,000 pounds by Primrose's account, and the loss ran past $20,000. Primrose sued Western Union. In 1894 the Supreme Court held that the most he could recover was $1.15, the price of the telegram.
The reason was printed on the form he had filled out. Western Union disclaimed liability for mistakes in any "UNREPEATED message, whether happening by negligence of its servants or otherwise, beyond the amount received for sending the same." For half the normal rate on top, under sixty cents in Primrose's case, a sender could have the message telegraphed back and compared. Fuller protection meant insuring the message as well, at one or two percent of the amount covered.
So the form offered two products. Repetition bought detection, a chance to catch an error before anyone acted on it. Insurance bought compensation if an error got through anyway. Primrose bought neither. Justice Horace Gray held that ordinary-rate messages traveled "at the risk of the sender," and that the company couldn't have foreseen the stakes, since knowing Primrose dealt in wool "had no tendency to show what the message was about."
Consider what that arrangement asked of the carrier. On ordinary messages, its exposure to its own mistakes was the fee it had already collected, and anxious customers paid it a surcharge for the check. I see no reason to think Western Union garbled telegrams to sell repeats. The terms just gave it very little reason to garble fewer.
The code made the problem worse. Plain English has slack in it: "I have bought" survives a slipped letter. Compressing a phrase into one short word squeezes the slack out, so a single changed vowel becomes a different instruction that still looks valid. Codebook makers knew this. Steven Bellovin's history of telegraph codes shows some of them choosing words that differed by at least two letters, so a single error produced nonsense. Merchants who economized on that margin ended up paying for it again through the repeat service. And because the code hid the meaning from the carrier, it also helped Gray call the loss unforeseeable.
Nearly a century later, Jerome Saltzer, David Reed and David Clark gave the idea an engineering form in "End-to-End Arguments in System Design." Tracing a file copied between two computers, they observed that checks inside the network couldn't catch errors made while the file was read, held in memory or saved. Their safeguard has the receiver read back what it stored, compute a checksum (a short number that changes whenever the data does) and send it back for the sender to compare. Only the party who knows what the message was meant to say can confirm it arrived intact.
That is a design principle rather than a ruling on damages, but in practice it tends to leave the cost where Gray left it. A program placing an order on someone else's website usually gets back a page saying the order was received. That page establishes receipt. Whether the quantity on record matches the quantity intended is a separate fact, and confirming it takes a read-back: a second request, more engineering, more time. The sender pays for that check or goes without, the same choice the printed form put to Primrose for under sixty cents.