A second-year biology major at a state university in Ohio closed her lecture slides at 11:40 on Tuesday 26 August, the first week of the fall semester, and opened a puzzle trainer in the next tab. Eight minutes of tactics puzzles. One game at five minutes plus three seconds a move. Two minutes going through that game to find the move that lost it. By Tuesday 2 December she had done the same thing 98 days in a row, her puzzle rating had gone from 940 to 1,310, and the note on her phone that tracks study blocks said the first ten minutes of each one, the part where her attention used to leave, had stopped being the hard part somewhere around week five.

The twenty minutes, as logged on 26 August

The whole thing ran on a free Lichess account and a laptop in a library carrel on the second floor.

ClockWhat happenedWhere
11:40 to 11:48Ten rated puzzles, three failed, failures saved to a listLichess puzzle trainer
11:48 to 11:59One 5+3 game against a 1,020 opponent, lost on move 31Lichess, rated
11:59 to 12:01The game replayed with an engine to find one moveAnalysis tool
12:01Slides reopenedCarrel

Ten puzzles took eight minutes that day. By November, ten puzzles at the 1,300 level took closer to eleven, so the set dropped to seven and the clock stayed the same.

Why the puzzles went first every day

The first puzzle on 26 August was a knight fork at the 950 level, solved in 14 seconds. The seventh was a back-rank mate she missed, and it went on the list. The list is where the method lives. On days 3, 6, 10 and 13 of every two-week block, the failed puzzles came back before any new ones, and some came back three times before they stopped failing. Four shapes made up most of the list through September.

ShapeTimes on the failure list, SeptTimes on it, Nov
Fork113
Pin92
Back-rank mate80
Discovered attack64

Lichess tags every puzzle with its theme and lets a user filter on one tag for free, and Chess.com does the same, so days 9 and 11 of each block were one shape only. Every other day was mixed, since nobody in a real game announces which shape is on the board.

The 5+3 game at 11:48

Five minutes a side plus three seconds a move finished in under twelve minutes on 91 of the 98 days. The seven that ran long were endings with both clocks under thirty seconds, and the slides opened late by two or three minutes. One game a day, never two. On 14 October a second game started at 12:00 and ended at 12:14, and the 12:10 lecture got missed. That was the only time.

The two minutes at 11:59

The lost game from 26 August went into ChessHelper.ai at 11:59, and the tool stepped through it on the board and stopped at move 19, a knight retreat to e7 that walked into a pin on the king she had solved as a puzzle four times that same week. The dashboard kept the game and, by October, showed the same pin appearing in six games across five weeks, which a single post-game look would never have shown. The opening trainer took her first eight moves from the same game, a Queen's Gambit Declined, and showed where move 6 had left the book line she had been trying to learn.

On every one of the 98 days the engine came out after the game and never during it. Engine help in a rated game is cheating on Lichess and Chess.com alike, the detection catches most of it, and the ban is permanent. The two minutes decided what went on the failure list for tomorrow and nothing else.

What a 2016 study found, and what the log found

A meta-analysis by Giovanni Sala and Fernand Gobet published in 2016 pooled 24 studies of chess instruction and put the overall effect on cognitive and academic measures at about a third of a standard deviation, larger in programs that ran 25 hours or more and smaller in the better-designed studies. Ninety-eight days at twenty minutes comes to 33 hours. Her grades on the two midterms in October were a B+ and an A-, against a B and a B+ in the same courses' first exams the previous spring, which proves nothing about chess and which she wrote in the log anyway.

What the log does show, in her own entries, is the study block. On 2 September the entry reads "took 12 min to actually start reading". On 7 October it reads "started at once, 50 min clean". From week five onward the time-to-start column sits between one and three minutes on 81 of the remaining days.

Chess on that campus, which she had not noticed in year one

The student union has a chess club that meets on Wednesdays at 7:00 pm with between 14 and 30 players, and she went for the first time on 1 October with a 1,100 game rating, which put her in the middle of the room. The club sends a four-player team to the Pan-American Intercollegiate Team Championship, which has run since 1946 and whose top four teams play the President's Cup each spring. The University of Texas at Dallas, Webster University, the University of Maryland Baltimore County and Saint Louis University recruit titled players on chess scholarships for that event, and nobody at the Wednesday club is at that level or expects to be.

On this campusPlayersEntry
Wednesday club night14 to 30Turning up
Rated club events, four a yearAbout 20US Chess membership, $20 a year for students
The Pan-American team4 plus two reservesA rating around 1,800 and up

From 8 October her Tuesday 11:48 game got played on Wednesday evenings instead, across a board, against a civil engineering student who beats her two games out of three.

2 December

Puzzle rating up 370. Seconds per puzzle down from 52 to 34. Failed puzzles per ten down from four to two. The knight retreat into a pin has not appeared in a game since 29 October. Ninety-eight sessions in ninety-eight days, two of them at an airport gate and one on a dormitory stairwell during a fire drill. The slides still close at 11:40 and reopen at 12:00, now with a 1,310 next to her name and a failure list that is shorter than it was.