- 01
HowHow Basic: a project-local research control plane
Why the durable records, hashes, append-only events and packaging live in a CLI rather than in the agent, and what that split buys.
Read note → - 02
Why a research agent needs one canonical truth
Provider state is not scientific truth. Events, approvals, evidence, runs and claims need one durable authority.
Read note → - 03
Failures belong in the scientific record
A system that keeps only its best run destroys the information a reader needs to trust the one that worked.
Read note → - 04
What READY FOR HUMAN REVIEW does not mean
Package checks and scientific correctness are separate judgements, and collapsing them into one badge is how a reader gets misled.
Read note →