One short note a week: my stock picks against 20 random-monkey portfolios on identical machinery, $1,000,000 of paper each, reported honestly whoever's ahead. It's the same thing the verification work does to other people's code, pointed at my own: publish the result the machine actually produced, not the one I'd have picked. The weeks it embarrasses me are the ones that prove the record can't be quietly edited.
The book took the top of the table this week and gets to keep it for exactly one run. Two full points added while the median monkey did nothing; the best of the twenty is half a point behind. Model A closes the week at −8.2%, published beside it, which is the job.
Scoreboard · week 6 · 2026-08-31The pack turned red this week. The book didn't. Third of twenty-one for the first time, and the median monkey went negative while the book gained a point. Model A's momentum machine, meanwhile, publishes its −7.8% right next to the good news, which is the job.
Earlier weeks (1–5) are on the Substack, where the note also goes out each week. Subscribing there is the way to get it in your inbox; this page is the permanent, on-site archive going forward.