We describe RIAN, a federated architecture of small Kuramoto-reservoir "brains" that achieves verifiable cross-domain reasoning without self-attention. Each brain is a fixed reservoir of ~1024 phase oscillators with a trained linear readout, specialised in one cognitive domain. Brains communicate by phase-cascade cross-talk: each brain carries a natural frequency ω and a vocabulary whitelist; a token activates only those brains whose ω resonates with the query phase and whose receptive field contains the token. The output of each activated brain becomes the probe of the next cascade level. Sentence formation is performed by a 50-line "speak brain" that maps cascade trees to natural-language templates. The system covers more than 30 cognitive domains with 109 productive brains, runs on a laptop, and produces natural-language sentences (e. g. "japan: su capital es tokyo, está en asia, su moneda es yen, su idioma es japanese") with 0. 1–7 s latency. It never hallucinates: queries outside the union of brain vocabularies stay silent. We document the architectural limits (ridge readout collapse on vocab-rich tasks, no copy mechanism, catastrophic interference of larger reservoirs), the OEIS / Wikipedia auto-scrapers that grow the corpus by verifying each new fact, and a stress test in which the system computes 9⁸1 exactly to 78 digits in 643 ms. This is paper 3 of the RIAN line. Bilingual EN+ES.
Sebastián Vázquez Andrade (Mon,) studied this question.