Why this exists
People correct AI every day, but most corrections disappear with the conversation or return as unbounded memory. Meraki exists to test a stricter alternative: keep the evidence, propose a small change, apply it only to matching work, and show whether it helped.
The public product is early. The local Core loop is real; the hosted experience and independent evaluation are not finished. We publish that boundary because trust begins with knowing which part is a working mechanism and which part is still a thesis.
The relationship
Catalyst asks the question. Meraki is the user-facing iterative layer. Meraki Core is the independent engine trying to make that iteration real through evidence, scoped retrieval, comparison, and rollback.
The principles
- Evidence before inference.
- A proposed change before an active rule.
- Scope before retrieval.
- Comparison before a learning claim.
- An undo path for every accepted change.