pramana-protocol

PRAMANA/1.0 — A Self-Verifying Cognition Protocol for AI-Native Service Universes

Named after Dharmakīrti's epistemological framework (c. 600–660 CE, Nalanda). Pramana = valid cognition.

The Problem

At 200+ AI services, capability registries decay into stale documentation. Services claim capabilities they no longer have. Knowledge bases contain rules that were never validated. Routing succeeds while cognition fails. Declaration without proof is hallucination at scale.

The Protocol

Three intertwined strands. Four signals. Five authenticity levels. One return signal that closes the loop.

Strand 1 — CAPABILITY    What services exist and what they can do  (AnkrCodex-compatible)
Strand 2 — KNOWLEDGE     Domain rules — SHASTRA / YUKTI / VIVEKA   (Knowledge Codex-compatible)
Strand 3 — PROOF         Implementation matches declaration         (CCA-compatible)

Signal 1 — STATE    What the service knows and can do
Signal 2 — TRUST    What is authorised
Signal 3 — SENSE    What is happening now
Signal 4 — PHALA    Delivery outcome → returned to SOURCE RULE     ← the novel signal

PHALA closes the loop. Every delivery outcome returns to the rule that generated the answer — not the service log. Wrong outcomes trigger RCA. Correct outcomes reaffirm the rule. The system learns from every transaction.

Authenticity Markers

Every output carries a marker. Never a percentage. A binary checklist per strand.

| Level | Label | Action | |---|---|---| | PRAMANA-0 | UNVERIFIED | Mandatory human check — do not act | | PRAMANA-1 | SINGLE SOURCE | Verify independently. System explains which strand + why it answered. | | PRAMANA-2 | PARTIAL AUTHENTIC | Missing strand named explicitly. Proceed with caution. | | PRAMANA-3 | AUTHENTIC | All three strands confirmed. Reliable. | | PRAMANA-3+ | AUTHENTICATED + INFERRED | Full coverage + SLM inference. Highest confidence. |

Install

npm install pramana-protocol

Usage

import {
  STRAND, SIGNAL, PRAMANA_LEVEL,
  getPramanaLevel, buildMarker,
  PhalaPayload
} from 'pramana-protocol';

// Determine authenticity level from available strands
const level = getPramanaLevel(
  [STRAND.CAPABILITY, STRAND.KNOWLEDGE],  // proof strand missing
  false                                    // SLM not available
);
// → PRAMANA_LEVEL.PARTIAL_AUTHENTIC (2)

// Build a display marker for the output
const marker = buildMarker(
  [STRAND.CAPABILITY, STRAND.KNOWLEDGE],
  false,
  'Proof strand unavailable — CCA not yet run for this service'
);
// → { level: 2, label: 'PRAMANA-2 — PARTIAL AUTHENTIC', strands_missing: ['proof'], ... }

// PHALA payload — fire on every delivery outcome
const phala: PhalaPayload = {
  signal_type: 'phala',
  rule_id: 'MAR-YK-007',
  applied_by: 'mari8x-legal',
  query_context: 'charter party clause, English law',
  outcome: 'failure',
  strands_available: [STRAND.CAPABILITY, STRAND.KNOWLEDGE],
  strand_missing: [STRAND.PROOF],
  authenticity_level: PRAMANA_LEVEL.PARTIAL_AUTHENTIC,
  rca_triggered: true,
  timestamp: new Date().toISOString()
};

Reference Implementation

ANKR is Customer Zero — 200+ services running PRAMANA/1.0.

• AnkrCodex (capability registry) — Strand 1
• GRANTHX Knowledge Codex (5,336 rules, SHASTRA/YUKTI/VIVEKA) — Strand 2
• CCA Capability Closure Audit (174 services audited) — Strand 3
• PHALA signal wired across all services

Protocol Specification

Full RFC-style spec: PROTOCOL.md

Academic paper (Zenodo DOI: 10.5281/zenodo.19273330): PAPER.md

Built on Forja

PRAMANA is the cognition layer built on top of the Forja declaration protocol.

Forja     = declaration protocol (what a service IS and can do)
PRAMANA   = cognition protocol   (whether that declaration holds under real usage)

• Forja: rocketlang/forja-protocol
• Trust constants: ankr-trust-constants

Author

Capt. Anil Kumar Sharma PowerPbox IT Solutions Pvt Ltd [email protected]

DPIIT Registered Startup, Haryana, India

License

Apache 2.0 — see LICENSE