The Holy Grail of Bug Bounties: Knuth's Reward Checks

A Typo Worth Chasing

In 1968, when the first volume of The Art of Computer Programming came off the press, Donald Knuth made an offer that sounds almost too small to matter: find a mistake in his book, and he will mail you a check¹. Not a big check. $2.56, to be exact, a number Knuth chose because it is 256 cents, or one "hexadecimal dollar" in the base-16 counting system computers use under the hood^1,2. A smaller reward, $0.32, went to readers who made a valuable suggestion that fell short of being a genuine error.

The joke was the point. Knuth was not trying to pay people a fair price for their time. He was signaling, in the native language of computer scientists, that he took errors in his notoriously rigorous books seriously enough to put real money behind them, however little money it was^1,5. The errors in question were not limited to typos. Knuth paid for technical mistakes, wrong historical attributions, and flawed algorithms too, and he still does, decades and multiple volumes later¹.

What happened next is the part people love to tell: the checks became collector's items. According to multiple accounts, recipients tended to frame them rather than deposit them, hanging a piece of paper worth less than a cup of coffee because of whose signature was on it^1,6. By October 2001, Knuth said he had written more than 2,000 checks, averaging over $8 apiece once you factor in multi-error finds and progressive software bug rewards^1,3. By March 2005, the lifetime total had crossed $20,000^1,3,7.

The system that made this possible, though, eventually became its own liability. Writing thousands of personal checks over four decades meant printing thousands of real bank account and routing numbers, and some of those checks ended up photographed and posted online as trophies⁸. In 2008, Knuth explained on his Stanford web page that modern check fraud made this impossible to sustain: a criminal only needs the numbers printed at the bottom of a check to drain an account, and he had already closed three checking accounts because of attacks tied to his bankers' exposure⁴. So in October 2008, he stopped writing negotiable checks for good^1,5.

In their place came something Knuth clearly enjoyed designing: certificates of deposit from the "Bank of San Serriffe," an offshore fictional bank he said has branches in Blefuscu and Elbonia, borrowing the name of a famous fictional nation invented for a 1977 April Fool's hoax in The Guardian^1,9,4. The reward is now denominated the same way, in hexadecimal dollars, but nobody can walk it into a teller window. Knuth himself noted that only 9 of the 275 checks he had sent since the start of 2006 had actually been cashed, which was less a scandal than a confirmation of what everyone already suspected: people wanted the honor, not the eight dollars⁴.

The Reward, By the Numbers

• $2.56: standard reward for a qualifying error, equal to 256 cents or one "hexadecimal dollar"^1,2.
• $0.32: typical reward for a valuable suggestion that isn't itself an error¹.
• $2.56 to $327.68: the progressive reward scale for bugs found in Knuth's TeX and METAFONT software, doubling every year¹.
• 2,000+: checks written as of October 2001, averaging more than $8 each^1,3.
• $20,000+: cumulative value of checks signed by March 2005^1,3,7.
• 9 of 275: checks actually cashed out of those Knuth sent between early 2006 and his 2008 policy change, by his own count⁴.

Who's Who

• Donald Knuth: Stanford computer scientist, author of The Art of Computer Programming, creator of the TeX typesetting system, and originator of the reward-check program in 1968^1,2.
• Readers, students, and programmers worldwide: the people who actually hunted down the errors; rewards went to the first person to report each one^1,2.
• The Bank of San Serriffe: a fictional offshore institution Knuth invented in 2008 to issue non-negotiable certificates after he stopped mailing real checks, borrowing its name from a famous 1977 newspaper hoax nation^1,9,4.
• Chris Thompson and Bogusław L. Jackowski, among others: cited as recipients of the doubling-value "progressive reward" for TeX and METAFONT bugs¹.

What Gets Exaggerated

The popular version of this story tends to flatten some real nuance. "Knuth pays $2.56 for every typo" undersells it: the reward covers technical, historical, and algorithmic errors too, not just typography^1,2.

"Every check was for $2.56" is also wrong. Reported averages exceeded $8 per check by 2001, since checks could cover multiple errors, bigger suggestions, or the doubling software-bug bounty that eventually capped at $327.68^1,3.

"Nobody ever cashed one" is close to true but not exact. Knuth's own 2008 count found 9 of 275 recent checks had been cashed, and he once mentioned a recipient near Frankfurt who could have deposited over $1,000 in accumulated checks had he chosen to^3,4. The honest version is that most people kept them as trophies, not that cashing was somehow forbidden.

Finally, "the program ended in 2008" is misleading. What ended was real, bank-drawn checks. The reward itself continues today through the Bank of San Serriffe certificates, which still track errors by book, page, and hexadecimal value¹.

Timeline

1. 1968Volume 1 of The Art of Computer Programming is published; Knuth begins rewarding readers who find errors in his books.
2. 1 July 1996Knuth sends out more than 250 letters, 125 containing checks, settling errors reported since 1981.
3. October 2001Knuth reports writing more than 2,000 checks, averaging over $8 each.
4. 8 March 1999Date on a publicly circulated example check for $2.56.
5. March 2005Cumulative value of checks signed by Knuth passes $20,000.
6. October 2008Knuth stops writing negotiable checks, citing check-fraud risk, and introduces Bank of San Serriffe certificates instead.

Pictures