Things I'm learning

the honest version

Each series is me learning something properly and writing it down as I go. Published parts are real; unpublished parts are promises.

Causal inference, from scratch
outlining · 0 of 7

From "why your dashboard is lying" to a defensible effect estimate. No prerequisites beyond curiosity and a little algebra.

1. why correlation lies 2. DAGs 3. confounders 4. do-calculus 5. estimation 6. sensitivity 7. case study
Agentic harnesses, in practice
outlining · 0 of 5

What actually makes agents useful: tools, checks, context, and evals. Built from real workflows, not demos.

1. what a harness is 2. tools & contracts 3. context engineering 4. evals 5. failure modes
ABSORBING LATELY
book
The Book of Why · Judea Pearl. Slowly, with a pencil.
papers
Agent evaluation writeups · anything honest about where agents fail.
rewatch
3Blue1Brown, linear algebra · it hits different the third time.