A Million Little Boxes

STAMFORD, Conn. — The contestants entered the ballroom to roaring applause, like it was a heavyweight prizefight. It was a crossword puzzle tournament.

They took their places on stage in front of giant whiteboards with empty crossword grids. Dry erase markers in hand, noise-canceling headphones on their heads, the three men stood in proper puzzling stance: right foot in front of left, marker arm crooked, clues in hand at convenient reading height.

“On your mark. Get set. Dan, begin,” said Will Shortz, New York Times crossword editor and tournament impresario. A tournament judge tapped Dan Feyer on the shoulder, and he began. Feyer, 37, is a professional musician and five-time defending champion.

Five seconds later: “Tyler, Howard, begin.” Tyler Hinman is the record-holder for youngest champion — he first won the tournament when he was 20, in 2005. He’s now 30, a grown-up prodigy. And while Howard Barkin has never won, this was his fourth trip to the finals.

The final puzzle was devilish. There’s wordplay aplenty. “Beat reporter?” is METRONOME. “Professional offers?” is ASSASSINS. “One who’s not committed?” is a SANEPERSON. And plenty of trivia. “Inventor of the spinning jenny” is HARGREAVES. “Four Holy Roman emperors” are OTTOS. “Rhododendron relative” is AZALEA. Whoever finishes first raises his hand and, assuming nothing is wrong in the puzzle, wins the American Crossword Puzzle Tournament.

After about seven minutes of solving, Feyer and Hinman were neck and neck, filling in their last boxes and checking that they hadn’t left anything blank. Hinman furiously erased a misentry with his left hand and rewrote with his right. Feyer, meanwhile, looked like Auguste Rodin, calmly — maybe too calmly! — surveying his handiwork.

Then one of them raised his hand.

“It would be a better story if someone else would win,” Feyer said. Feyer is the best crossword solver on the planet. We first met on a train platform in late March, preparing to go to Stamford for the tournament, which is the de facto national championship. Feyer has won every year for the last five years. He was heading north in pursuit of a record sixth consecutive title.

This was the 38th annual tournament, which was founded and is still run by Shortz, the elder statesman of crosswords. Jon Stewart has called him “the Errol Flynn of crossword puzzling.”

Five hundred sixty-seven competitors were en route to the ballroom of the Stamford Marriott. They’d be joined by dozens of tournament judges in what is part competition, part convention and part reunion.

While most of us mere mortals might measure our solving time in cups of coffee or brunch courses — if we even complete the puzzles we start — the top solvers at the tournament measure it in split seconds. Points are awarded for correctness, of course, but also for speed.

Over the next 72 hours, Feyer will be challenged by the puzzle-solving elite as he tries to defend his title. (Not all the challengers will be human.) And he’ll face a final puzzle that he suspects may have been designed to dethrone him.

Feyer came to crosswords late in life. If he ran across a New York Times Magazine in college, he’d take a crack at the Sunday puzzle. But as recently as 2006, he was oblivious to any of their finer points, he said.

“I didn’t know anything about them. I didn’t understand different constructors, I didn’t even really know the day of the week,” he said. Crosswords are well known to get tougher as the week goes on. Feyer had “a book of New York Times daily puzzles, and like every sixth one I wasn’t really finishing. It was like a year later I was like, ‘Ooh! Those were the Saturday puzzles.’”

The only way to get good at crosswords is to do lots of crosswords. And Dan Feyer has done lots of crosswords. By his back-of-the-envelope calculations, he’s spent “just” 4,000 hours of his life solving. Feyer started puzzling in earnest after seeing the crossword documentary “Wordplay” in the fall of 2007.¹ Now he routinely does 10 crosswords a day, and he used to regularly do 20. He’s done all of the nearly 8,000 puzzles in the New York Times’ online database, and routinely whips through puzzles from The Los Angeles Times, Newsday, The Chronicle of Higher Education, CrosSynergy, and others. There’s no guidebook to becoming a master crossworder, like there is for, say, Scrabble. The doing is the training.

By day, Feyer is a pianist and music director in San Francisco.² The tournament elite aren’t poets or Shakespeare scholars or copy editors; for the most part they’re musicians, programmers and the mathematically inclined. Jon Delfin, a seven-time champion, is also a pianist. These folks can quickly make sense of a lot of coded information.

While he recently moved to the West Coast, Feyer spent 14 years in New York City — he still covets his 212 cell phone number. The city was central to fostering his crossword mastery: He did crosswords on the subway and in the orchestra pit when he was working an off-Broadway show. When he was starting out, he kept a stack of hard-copy puzzles he’d done in his apartment. It reached to roughly the height of a kitchen table. He even entered his first ACPT because of its convenient Brooklyn location. (The tournament moved back to its original Stamford home this year. Going into this year’s tournament, Feyer hadn’t won a title outside of New York.)

On the train, Feyer alerted me to some of his could-be challengers. They sound like the beginnings of an upstart think tank. There’s Joon Pahk, the Harvard physics preceptor and seven-time “Jeopardy!” champion; Andy Kravis, the Columbia Law grad who won $2.6 million on the now-defunct Million Second Quiz; and David Plotkin, the entomology — no, not etymology — Ph.D. student. Plotkin often matches Feyer on speed, and can sometimes outpace him. “He’s still on the rise,” Feyer said.

YEAR	FEYER	HINMAN	DR. FILL
2005	—	1st	—
2006	—	1st	—
2007	—	1st	—
2008	45th	1st	—
2009	4th	1st	—
2010	1st	4th	—
2011	1st	2nd	—
2012	1st	2nd	141st
2013	1st	3rd	92nd
2014	1st	2nd	67th

There’s also a computer. Dr. Fill is a software program whose creator proved to be a terrible crossword solver, but a brilliant computer scientist.

And then there’s Hinman, a game designer for Lumosity. Hinman plays the baby-faced, redheaded Apollo Creed to Feyer’s bearded, bald Rocky Balboa. I met him in the hotel lobby and he was — what else? — solving a crossword.

Before Feyer won five in a row, Hinman completed his own five-year run of victories. For the last decade, these are the only two people who have tasted crossword glory.

Hinman and Feyer are amiable enough — they’re in such a small circle of elite solvers that they almost have to be — but there’s a tension; a person can’t play second fiddle for too long without getting frustrated.

 

“He definitely enjoys his recent dominance, over me in particular,” Hinman said. “And there’s definitely a bloodthirsty streak that really wants to rip it away from him.”

Over the past few years, Feyer has been religious about tracking his solving times. On his blog, he posts a spreadsheet every day in which he records the time it takes him to solve various puzzles and invites other top solvers to add their own scores.

“Can I use the phrase ‘dick-measuring contest’?” Hinman asked. “I don’t feel like I need to do that. It’s for me, you know? If I want to prove it to people, this is where I go,” he said, gesturing to the tournament taking place around us.

And Hinman also lives in San Francisco, where Feyer recently moved. The town ain’t big enough for the both of ’em.

“That was really great, not to even be the top-ranked crossword puzzler in my own city,” Hinman joked. “We struck a bargain: I’m the best solver in the country east of Divisadero Street. Which is a pretty big swath, so I feel pretty good about that.”

Since late 2011, Feyer has finished over 95 percent of the New York Times puzzles in less than six minutes. He polishes off 88 percent of the Saturday puzzles — the hardest of the week — in less than five minutes.

Let the speed wash over you. Grab a puzzle, any puzzle. Solve it. Compare your time to Feyer’s. I’d be lucky to find a pen in my desk drawer before he finished. For New York Times puzzles, his average solve times, Monday to Sunday, are 1:40, 1:55, 2:10, 3:09, 3:17, 4:03 and 5:38.³

To the exasperation of many tournament solvers, puzzles at the ACPT are still solved on paper, so Feyer will have to use his trusty Bics for the next few days. As the tournament approaches, Feyer and many other solvers switch from the computer (which is a more efficient way to tear through a bunch of puzzles) to writing it all down. It’s physical training more than anything. They have to train their eyes to scan a physical list of clues and train their hands to write fast while they process what’s coming next. But even on paper, Feyer flies.

Over drinks the night before the action begins, I tried to get some assembled crossword constructors to handicap the field for me. One thing was clear: Feyer was a huge favorite — although the group couldn’t agree whether his odds of winning were 3-to-5 or 4-to-5. Six others, including Hinman, had a shot, they said. The entire rest of the field? 70-to-1.⁴

That same night, Matt Ginsberg walked me deep into the belly of the Stamford Marriott, to a room with a sign on the door that read “Do Not Enter.” We entered.

The room was filled with machines — printers, scanners, laptops. On one particular machine lived Dr. Fill, the artificial intelligence crossword solver.

Ginsberg loaded up that week’s Saturday New York Times puzzle — the hardest day of the week, and one Dr. Fill had never seen. At the click of a button, the doctor got to work. The grid flickered and filled. Black, blue, purple and green letters came to life. In no more than five seconds, the grid was full. Ginsberg gave it a once over. It was correct.

Ginsberg, 59, is Dr. Fill’s creator and something like The Most Interesting Man in Crosswords. He got his Ph.D. in relativistic astrophysics from Oxford when he was 24, and he moved in the same circles as Stephen Hawking and Richard Feynman, he said. He runs three software companies. He’s a stunt pilot and amateur magician. He wrote a play that was produced. He created GIB, a former computer bridge world champion.

“Dr. Fill is my attempt to get revenge on the crossword constructors because I’m such an awful solver,” Ginsberg said.

Dr. Fill⁵ knows millions of crossword clues and answers.⁶ It knows the title of every Wikipedia article, as well the articles’ lengths and when they were last edited — trendier is better in crossword answers. It’s memorized the complete works of Shakespeare, the Bible and Bartlett’s quotations. It knows the number of Google hits for 15 million different queries. While Ginsberg has an agreement with Shortz that Dr. Fill won’t access the Internet during competition, its offline database, compressed, is just shy of a gigabyte.

Here it is solving the 2013 tournament puzzles:

Ginsberg closed his MacBook Pro and put Dr. Fill to sleep until tomorrow. Also in that Marriott basement room was a stack of boxes, ready to be opened the next morning. In all caps, in bold red Sharpie, they read: “TOURNAMENT PUZZLES.” Dr. Fill had no idea.

It was a few minutes after 11 a.m. on Saturday. “Ready. Set. Go,” declared tournament impresario Shortz. We were underway.

With pencils at the ready, the 500-plus solvers flipped over their face-down puzzles. It sounded like Shortz had frightened a flock of birds out of the ballroom. But for the next 15 minutes, the competitors worked in library-like silence.

They’d do this seven times over the next two days — and an eighth time for the three finalists. All told, nearly one million little boxes would be filled in. For the majority of the contestants, it’s a lark — a chance to solve puzzles and be among friends. But for a handful, it’s also a mad dash to qualify for the final.

Some of the puzzles (No. 1 and No. 6) are easy — Monday- or Tuesday-level. Some (No. 3) are tougher. Some (No. 7) have larger grids, like the one you’d find in Sunday’s New York Times Magazine. And some (No. 5) are infamous. We’ll get to that.

Feyer brought his A-game. He ripped into Puzzle 1 (see image at left) like a kid on Christmas morning. The final time was 1:55, which was the first sub-two-minute puzzle ever completed in the tournament. It was less than a commercial break’s worth of time into the competition, and I’d already witnessed crossword puzzles’ equivalent of the first four-minute mile. (“Hashtag Bannister,” Feyer joked.) To make sense of his time, I conducted a little experiment. I wrote my first name over and over again in the same grid, as fast as I could while maintaining legibility. It took me 1:40.

Shortz took the mic between rounds to introduce each puzzle and offer regular updates on Dr. Fill’s performance. When Shortz announced that Dr. Fill finished the first puzzle in under a minute, the humans groaned — some combination of resignation and annoyance. But, Shortz added, it finished with one mistake! An eruption of applause and laughter followed. Dr. Fill was unaware of the “Parks and Recreation” actress Retta (47 across). Humans 1, Computer 0.

 

This fraught relationship between humans and machine continued throughout the weekend. What makes crosswords difficult for Dr. Fill is precisely what makes them fun for humans — themes, gimmicks, neologisms and wordplay. And the humans — except for Ginsberg, perhaps — would like to keep it that way.

While crossword construction is unavoidably linguistic, constructors these days rely heavily on data and programming. Take, for instance, Puzzle 5, always the hardest of the tournament. This year, Jeff Chen was the sadist behind the curtain. An aspiring novelist from Seattle, he runs the crossword database XWord Info and has authored or co-authored 37 New York Times puzzles.

Chen’s Puzzle 5 was titled “Attention, Newbies!” The conceit was adding new B’s (get it?) to familiar phrases. “Vocalist” became “vocab list,” “alloy” became “ballboy,” and so on. To find workable and interesting phrases like this, Chen wrote a Python script and applied it to his master word and phrase list. Using this program, he realized “caroms” could become “car bombs,” for example.

Chen’s master list has about 230,000 entries, each with a quality score attached.⁷ The higher the score, the better it would be to include in a puzzle. Much of the scoring is automated, largely based on phrases’ previous puzzle appearances, but some take a human touch. Because three-, four- and five-letter words come up so often in puzzles, Chen decided to score them all by hand.

Constructors’ databases are their livelihoods. In his book “Gridlock,” Matt Gaffney described how constructor Frank Longo would mail a CD copy of his crossword entry database to an “undisclosed location” each month. It helped him sleep better at night.

Feyer wasn’t too bothered by Chen’s construction, though — the mind-melting puzzle took him a tidy 7:35. He remained in first. Dr. Fill finished fastest — but made one mistake. Oh, the humanity!

On Saturday evening, the day’s puzzling was complete and things were going according to script. Feyer was in first, with a three-minute lead over the field. He could take his time on the next morning’s Puzzle 7 and still coast into the final.

Hinman had struggled. He just missed the time cutoffs on Puzzles 4 and 5, finishing seconds too slow and not earning crucial bonus points. He’d have to tear through Puzzle 7 the next morning to qualify to face Feyer.

Dr. Fill, meanwhile, was faster than anyone — but error-prone. On Puzzle 3, constructed by Merl Reagle, it made 12 mistakes, to the utter delight and hearty applause of the human solvers. But while it was nowhere near Feyer, it was on pace for its best-ever finish.

Feyer wasn’t concerned with who his competition would be in the final, but he was a bit concerned about the puzzle itself. Gretchen Margaroli, Feyer’s fiancée, told me later that Feyer had never worried about a final more than this year’s.

Feyer had a hunch — later proved right — that the final puzzle’s constructor was Byron Walden. Walden is a professor of math and computer science at Santa Clara University and is prone to using science-related crossword fill — not Feyer’s strongest suit. He prefers the constructor Patrick Blindauer, for example, and his penchant for musical theater clues.

In the tournament packet, Feyer also noticed that the final puzzle, the one that would decide the champion the next afternoon, would have only 60 words — a minuscule number. According to XWord Info, only 29 New York Times crossword puzzles have had fewer words during Shortz’s reign as editor. That meant it would be a wide-open grid with few black squares. Long words weaving through long words.

“This makes it a lot harder to solve,” Feyer said. “It also increases variance. It makes it easier to get stuck. They may be trying to take me down.”⁸

Hinman had steeled himself by Sunday morning. His new motto: “#SolveAngry.” He flew through puzzles 6 and 7 faster than anyone in the field, meeting key time cutoffs. He earned enough points to make the final.

Right before his showdown with Feyer began, Hinman tweeted:

Feyer was still in first, and according to tournament rules, that earned him a five-second head start.

But surely that wouldn’t matter, right?

 

It was Sunday, lunchtime, and the three finalists — Feyer, Hinman and Barkin — were sequestered in an oversize storage closet. If they so much as used the restroom, they were required to have an escort. The tense atmosphere, strained small talk and unexplained Christmas tree in the corner turned this into a sort of surreal waiting room.

“We have an hour to shit our pants,” Feyer said.

Finally, there was a knock on the door.

The finalists entered the ballroom and climbed onstage. Grids were empty, markers were uncapped. “Is everybody ready?” Shortz asked.

After seven minutes of tense solving, we join the action in the video below.⁹ Hinman is solving on the left, Feyer on the right:

 

Feyer finished, gave his grid a once-over, and calmly raised his left hand. A second later, after filling in his last letter, Hinman shot his right hand up. He glanced at Feyer and realized he was too late.¹⁰ The five-second head start had made the difference.

After a million boxes had been filled in, the champion was decided by just one. It was a record sixth consecutive victory for Feyer. And another silver for Hinman — his fourth in the last five years.

“I was on the floor after Georgia State hit that shot,” said sportswriter Adesina Koiki. “And that was nothing like this.”

I spotted Ginsberg, Dr. Fill’s creator, outside the hotel a few minutes later. Best final ever? I asked. “Best final ever,” he said. Dr. Fill did solve the final puzzle, but not on stage. It finished in just over two minutes — five minutes quicker than Feyer — but it made two mistakes. Nevertheless, the auto-solver still placed 55th, its best-ever result.

Feyer said it’d be a better story if someone else won. For once that weekend, the unflappable Dan Feyer was wrong.