For years I brought a spreadsheet to my $400 fantasy football auction: one fair value per player.
This year with Claude Code I built a tool that reprices every player after every pick, then ran ~1,000 simulated drafts against it.
Finished with the #1 projected roster in a 16-team league. Here's what I learned. π§΅
@GeoffreyHuntley 90% of email writing is LLM generated-
1. Sending human gives AI bullet points to turn into beautiful, grammatically correct email
2. Receiving human uses AI to turn it back into bullet points
Inference providers win even when inputs and outputs are the same!
Full writeup, with the failures (a broken strategy bot, a rejected depth premium, CPUs that were too soft):
https://t.co/uJul6M2k9j
I'm building more things like this. December recap coming. Follow along.
For years I brought a spreadsheet to my $400 fantasy football auction: one fair value per player.
This year with Claude Code I built a tool that reprices every player after every pick, then ran ~1,000 simulated drafts against it.
Finished with the #1 projected roster in a 16-team league. Here's what I learned. π§΅
The general lesson isn't about football.
A static valuation is a prior. The yardstick is marginal value given your current state and the remaining alternatives.
Any time you carry a frozen number into a sequential decision, ask what the live version would be.