Pramana protocol standardizes claim verification for autonomous AI agents

Ravi Kiran Kadaboina's paper, "Pramana: A Protocol-Layer Treatment of Claim Verification in Autonomous Agent Networks," proposes a standardized wire format for audit trails in agentic AI systems. Current verification methods fall into two unstandardized camps: probabilistic verdict patterns (e.g., self-consistency voting, LLM ensembles) produce judgments but no replayable artifacts, while artifact-producing patterns (RAG, tool-augmented traces) create vendor-specific records that external auditors cannot reconstruct without bespoke integration. Pramana wraps every consequential agent output in a typed ClaimAttestation with one of four variants—measurement, inference, analogy, citation—each paired with a verify() operation against the recorded source. The four-way typology is drawn from classical Indian epistemology (pramana, meaning "valid means of knowledge"). For MeasurementClaim and CitationClaim, verify() is fully deterministic; for InferenceClaim and AnalogyClaim, it is conditionally deterministic (audit-replayable when backed by an LLM).

The protocol's lifecycle is formally specified in TLA+ and exhaustively model-checked with TLC across three symmetry-reduced models, hitting 38,563 distinct reachable states with zero invariant violations. A Python reference implementation passes all 84 tests. The paper also defines A2A and MCP wire-extension manifests enforcing three deployment-grade invariants: reachability, SLA bound, and offline re-verifiability. An exploratory pilot involving 100 code-generation samples and 2,275 LLM-as-judge reviewer calls did not aim to validate Pramana itself but found a striking 40-percentage-point raw false positive rate (FPR) delta across different corpora, suggesting that reference-solution quality heavily skews LLM-based evaluation. The structural argument and formal verification, not the pilot, substantiate Pramana's claim of enabling verifiable autonomous agent systems.