Skip to content

Architecture

How CORA Remembers: an append-only log and no silent decisions

A record you can edit is not really a record. Overwrite a stored value and its past is gone: what it used to say, and why it changed, are simply unrecoverable. Yet those are exactly the things a beamline needs back, long after the fact. Which recipe did this run follow? Who approved the energy? Why was the center of rotation set there? CORA can answer questions like these for the same reason it can be trusted at all: it was built never to overwrite anything in the first place.

Agents as Principals: one identity for humans and AI

Two of the operators at the 2-BM beamline are not people. One of them reads a finished scan and records whether it completed cleanly, ran degraded, or aborted on a fault. The other watches for terminal run events and proposes a caution for a human to review. Both are software, and both raise the question every team adopting AI is now asking: how do you let an automated actor do real work without giving it a special, weaker, or unaccountable identity of its own? CORA's answer is that there is no special path. An agent is a principal, recorded, authorized, and audited exactly as a person is.