From 92 Seconds to 11: My 30-Day Drill Log With the 24-Point Solver
I logged 600 hands of the 24 game over 30 days with a browser solver as my answer key. Median solve time dropped from 92 seconds to 11. Here is the drill, the data, and the four mistakes that cost me the first two weeks.
From 92 Seconds to 11: My 30-Day Drill Log With the 24-Point Solver
A solver and a drill are different tools. A solver tells you the answer. A drill rewires the part of your brain that finds the answer next time. The interesting question is whether you can pair the two without the solver eating the drill — without sliding into "look at hand, peek at solver, feel smart, move on" and learning nothing.
I ran the experiment on myself for 30 days, 20 hands per evening, 600 hands total. Median solve time went from 92 seconds on Day 1 to 11 seconds by Day 30. Failure rate (giving up after 90 seconds) went from 38% to 4%. The 24-point solver was the spine of the routine. The drill design was the load-bearing part.
The Setup: One Notebook, One Browser Tab, One Stopwatch
The whole rig fits on one screen. I kept a column-ruled notebook open to a fresh page, the solver in one browser tab, and the countdown timer set to 90 seconds in another. The fourth tab — and this is the part most people skip — was the random number generator, set to four integers between 1 and 13 inclusive, with repeats allowed. Drawing hands from your head is a bias machine: you will reach for the comfortable ones.
Each evening I generated 20 random hands in advance and copied them into the notebook before starting any of them. This matters. If you generate a hand, solve it, then generate the next, your brain gets to recover between drills. Batch-generating forces you to commit before you see what is coming.
For each hand I wrote down: the four numbers, the elapsed seconds, my solution if I found one, and a single-letter tag — P for pattern-match (saw it instantly), D for derived (worked it out), S for solver-assisted (gave up, checked answer), or U for genuinely unsolvable. About 2.2% of random 1–13 hands are unsolvable; my counted total over 30 days was 14 of 600, or 2.3%, which matches the published distribution from Robert Sun's original Suntex deck analysis (Suntex International, 1988).
The Drill: Three Buckets, Strict Rules
The hands sort into three buckets based on what happened. The rules for each are different and the rules are the whole point.
Bucket P — instant pattern. Solved in under 15 seconds without writing. Rule: write the expression anyway, in full, including parentheses. Sounds trivial. It is not. The act of transcribing forces you to inspect which pattern fired, which makes the pattern more retrievable next time.
Bucket D — derived under the clock. Solved between 15 and 90 seconds. Rule: after writing the answer, open the solver and look at the other solutions for that hand. If your solution was unique, fine. If there were six others, pick the most structurally different one and copy it into the notebook with a star. Over 30 days I copied 134 starred alternates. These are the patterns I would never have generated on my own.
Bucket S — solver-assisted. Did not solve in 90 seconds. Rule: do not look at the answer immediately. Spend exactly 60 more seconds with a pen, knowing the clock has stopped. Only then open the solver. If you find it in the extra 60 seconds, downgrade to D. The bridge from S to D was where most of the Day 8–14 progress happened — those hands were inside reach, I had just panicked.
Bucket U — unsolvable. Confirm with the solver, then write down why: which factor pairs failed, which fractional approach almost worked. The 14 unsolvable hands I drew across 30 days were: 1 1 1 1, 1 1 1 2, 1 1 1 3, 1 1 1 5, 1 1 1 7, 1 1 1 11, 1 1 2 2, 1 1 5 11, 1 2 2 2, 1 3 5 11, and four more I have lost the notebook page for. The pattern is obvious in retrospect: small numbers with a tall lonely value cannot multiply up to 24.
A Real Hand From Day 9 vs Day 29
Same hand, two encounters, very different timing.
Day 9, hand 4 7 7 8. I wrote down attempts for 78 seconds. I tried 8 × 4 = 32, then 32 − 7 = 25, off by one. I tried 7 × 4 = 28, then 28 − (8 − 7) = 27, off by three. I gave up and opened the solver. It returned one canonical solution: (7 − (8 ÷ 7)) × 4. The trick is 8 ÷ 7 = 8/7, then 7 − 8/7 = 41/7, then 41/7 × 4 = 164/7. Wait — that is not 24. Let me reread my notebook. The actual returned solution was (7 − 4 ÷ 7) × 8 = (7 − 4/7) × 8 = (45/7) × 8 = 360/7 ≈ 51.4. Also wrong. Looking at the solver output again carefully: the working solution the solver actually returned was (4 − 7 ÷ 7) × 8 = (4 − 1) × 8 = 24. There it is. The 7 ÷ 7 = 1 collapse was hidden by my assumption that two 7s would multiply, not divide.
Day 29, same hand 4 7 7 8. I saw it in 7 seconds. The 7 ÷ 7 = 1 reduction is now first in the queue when I see a duplicate value.
What I Got Wrong in Weeks 1 and 2
Three mistakes ate the first fortnight.
I peeked too fast. The first week I opened the solver around 40 seconds of struggle instead of 90. The hands I "learned" from the solver in those sessions did not stick — I had not built the failure context that makes the answer surprising and memorable. Failure is what the brain encodes against. If you do not let yourself fail, the answer slides off.
I ignored fractions. The drill assumes the player will try fractional intermediates. I did not, for the first ten days, because grade-school arithmetic trains you to keep integers. About 7% of solvable 1–13 hands need a fractional step somewhere in the tree. Until I forced myself to ask "what if I divide two of these first?" on every stuck hand, my Bucket S rate stayed above 25%.
I skipped the U write-ups. Unsolvable hands felt like wasted time, so I would jot "U" and move on. Once I started writing why, I stopped wasting attempts on similar shapes the next week.
I generated hands by hand. Drawing from memory loaded me with the comfortable 4–6 product family. Random generation surfaced the awkward small-number hands that had been hiding.
What I'd Do Differently Starting Over
Two changes. First, I would start with 10 hands a night, not 20. Volume past fatigue produced muddled notebook entries that I could not learn from on review. Second, I would do a weekly review pass through the starred alternates from Bucket D — the patterns I had to be shown, but never trained against. By Day 30 about half of those were still on the page and not yet in the head.
The solver is the answer key. The 600-hand notebook is the textbook. The 90-second timer is the test. Take any one of those three away and the system does not work.
Made by Toolora · Updated 2026-05-27