Research
Short, dated notes. Findings with numbers.
Written for fund executives, on public data or plain reasoning. One a month, each one checked before it goes live.
-
01
In preparation
Re-scoring well-known public backtests for the trials behind them
Famous results, restated after the number of tries it took to find them.
-
02
Five agreeing analysts are worth 1.47 opinions
Why a panel of AI agents that agree is weaker evidence than it looks, and how to discount it.
-
03
What allocators now ask about AI in due diligence
Seven questions, and what a good answer looks like for each.
-
04
Five properties that separate a guardrail from a blocker
A guardrail that only says no teaches an agent to guess. What a good one does instead.
-
05
Why dropping one data feed at a time can cancel the wrong one
When two feeds overlap, each looks worthless on its own.
Want the next note first?
Request access and say which question matters to you. A founder replies, and partners see findings before they are published.