PRAMANA: A Self-Verifying Cognition Protocol for AI-Native Service Universes

As AI-native service universes scale beyond 100 services, a structural problem emerges: no single agent, router, or gateway can verify whether a response was generated from a valid capability, a valid rule, and a valid delivery outcome — simultaneously. Existing approaches (service meshes, API gateways, knowledge graphs) solve routing or observability, but not *cognition correctness*. We present PRAMANA — a self-verifying cognition protocol named after Dharmakīrti's (c. 600–660 CE, Nalanda) epistemological framework for valid cognition. PRAMANA binds three verification strands — capability, knowledge, and proof — through four signals: STATE, TRUST, SENSE, and PHALA. The novel contribution is PHALA: a delivery outcome signal that returns to the *source rule*, not the service log. Every delivery reaffirms or challenges the rule that generated the answer, creating a recursive RCA loop that can train a Small Language Model (SLM) when plugged in. PRAMANA is not theoretical — it is documented from a working implementation across 200+ services, 5,336 domain rules, and 174 audited capability declarations. The authenticity marker system (PRAMANA-0 through PRAMANA-3+) provides a binary-per-strand checklist that any downstream agent or human can interpret without probabilistic reasoning.