Technical Architecture | Inquiro AI Research Labs

Architecture Overview

Receipts-native architecture generates cryptographic proof at inference-time, not post-hoc. Every AI operation returns a tuple of (result, receipt). Each receipt contains a dual-hash (SHA256:BLAKE3) payload, Merkle chain linkage, Ed25519 signature, and RACI governance chain.

Latency: sub-2ms per receipt. Throughput: 100,000 receipts/second. Compression: 4.8-5.5x over zstd via Comprimere (optional value-add module, not core RNA).

Cryptographic Anchoring

Each receipt is anchored by three independent proof mechanisms. Any one is sufficient; all three together make the record unforgeable.

Payload Hash

Dual-hash of the decision payload using SHA256 and BLAKE3 simultaneously. SHA256 provides NIST-standard compliance; BLAKE3 provides speed (3-4x faster) and a quantum resistance pathway. Combined collision probability: 2^-512, effectively zero.

"payload_hash": {
  "sha256": "a7f3c2d8...e1f9a4b6",
  "blake3": "4f9a2bc7...d7e1f4a8"
}

Merkle Chain

Each receipt is linked to the previous receipt via hash pointer, forming an append-only chain. Tampering with any receipt breaks all subsequent hashes; detection is instantaneous without scanning full history.

"merkle_root": "2c8f9ea1...d5e1f7a4",
"chain_index": 8472,
"prev_receipt_hash": "1b7d4f9a...c5d2e6f8"

Ed25519 Signature

Elliptic curve digital signature proving receipt authenticity. 62,000 signatures/sec on commodity hardware, 32-byte keys, deterministic (no random number generator dependency).

"signature": "ed25519:9f2a7c4e...e2b5d8f1a4"

RACI Governance Chain

Every receipt carries a structured accountability record tracking Responsible, Accountable, Consulted, and Informed actors at each decision point. Regulatory frameworks (DoD 3000.09, EU AI Act Article 12) require knowing who was involved in each decision. RACI answers this with cryptographic timestamps, not assertions.

"raci_chain": [
  { "role": "accountable", "actor": "flight_controller", "ts": "2026-03-14T14:32:19Z" },
  { "role": "informed", "actor": "ground_control_7291", "ts": "2026-03-14T14:32:20Z" }
]

Human-in-loop notification proven at T+1s. Unforgeable accountability chain from decision to action.

Decision Metadata

Each receipt records what decision was made, when, and why, including the environmental context at decision time. Machine-readable format enables automated compliance checking, actuarial analysis, and regulatory reporting without human interpretation of ambiguous log entries.

"receipt_type": "decision",
"ts": "2026-03-14T14:32:19Z",
"decision": "altitude_adjustment",
"rationale": "wind_gust_detected",
"context": {
  "wind_speed_kts": 18,
  "tolerance_threshold_kts": 25,
  "regulation_ref": "FAA_Part_107"
}

Insurance carriers can programmatically verify that decisions complied with regulations: no interpretation, no ambiguity.

Design Principles

Receipts at compute-time

Every operation returns (result, receipt). The receipt is a first-class output of every computation, not a log entry appended afterward. If a function doesn't return a receipt, it didn't happen.

Append-only by design

The receipt chain has no delete, no update, no overwrite. Once a receipt is anchored to the Merkle chain, it exists permanently. This is a feature, not a limitation.

Self-authenticating

Receipts authenticate through cryptographic properties alone. No trusted third party, no certificate authority, no blockchain consensus mechanism required. The math is the proof.

Defense-in-depth

Dual-hash provides algorithm-level redundancy. Merkle chain provides chain-level integrity. Ed25519 provides origin authentication. RACI chain provides governance accountability. Four independent proof mechanisms; any one of which is sufficient.

Edge-native

Designed for resource-constrained environments: drones, satellites, autonomous vehicles. Sub-2ms latency and 32-byte keys mean receipts work where cloud connectivity doesn't exist. Receipts generate locally, sync when connected.

Performance benchmarks from local verified prototypes using receipts-native architecture foundation. Compression metrics represent Comprimere module (value-add layer, not core RNA). Not production systems.

Implementation Stack

Capability	RNA Implementation	Regulatory Requirement
Latency per entry	<2ms	Real-time decision logging (DoD 3000.09)
Tamper-evident	Yes (Merkle chain)	Court-admissible evidence (FRE 901(b)(9))
Court admissible	Yes (Ed25519 + Merkle)	Machine-generated record authentication
Works offline / edge	Yes	BVLOS operations (FAA Part 108)
Governance chain	RACI built-in	Accountability tracking (DoD 3000.09, EU AI Act Art. 12)
Regulatory coverage	All 4 frameworks	FRE 901(b)(9), DoD 3000.09, FAA Part 108, EU AI Act Art. 12
Resource requirements	Minimal (32-byte keys)	Edge deployment (drones, satellites, autonomous vehicles)

Performance benchmarks from local verified prototypes. Not production systems.

Back to Main Site