Published June 10, 2026
I was reviewing a neuron — one of the persistent memory units that holds everything the system knows about a specific project — and the count said 156 atomic facts. That’s wrong. A healthy neuron carries 20, maybe 40. I opened it and started reading.
Facts about three different projects. Financial data sitting next to system logs. Names from contexts that have nothing to do with each other. A mess, and the worst part is I don’t know when it started.
The autonomous curator routes every incoming fact to the neuron where it belongs, and the matching was based on content similarity. If a fact mentioned a word that appeared in the neuron, it scored as a candidate. The problem: that ignores who the neuron is about. A fact about one project mentioned a term that also appeared in a completely different project’s neuron. The curator matched on the word, not the entity. That one misroute made the neuron’s content more ambiguous, which attracted more wrong facts. And those made it worse. Classic cascade — silent, gradual, invisible until the neuron is bloated beyond recognition.
We added a subject-entity check before content similarity runs. Now the system asks “does this fact belong to the same entity this neuron tracks?” before it even looks at how similar the content is. If the subject doesn’t match, rejected. Doesn’t matter how relevant the words look.
I cleaned the contaminated neuron by hand — relocated the 120+ misrouted facts to where they actually belonged. Then watched. 72 hours, zero misroutes. The neuron settled at 34 facts and stayed there.
Content similarity alone is a terrible routing strategy for memory governance. The system needed identity checks from day one.
— Javier
EIDARA v2 is free. SUPER DARA is what comes next.