Data integrity infrastructure
for high-stakes AI.

Cryptographically sealed, forensically examined, and independently verifiable dataset certification — purpose-built for any regulated AI pipeline.

Trust Kernel v1.4.2 is now available. Read the release notes →

Request a pilot → View product Documentation →

Trust Console live · v1.4.2

--:--:--

APPROVED

RUNNING

VERIFIED

Zero server trust · Ed25519 · offline-verifiable View docs →

The Protocol

Four steps. One unbreakable chain of evidence.

Full architecture →

01 · FREEZE

Canonical Snapshot

Deterministic byte encoding. Same dataset + spec = identical output on any platform, any OS, any time.

02 · EXAMINE

Forensic Analysis

Policy engine runs every check — schema, nulls, leakage, distribution, version consistency. Configurable per domain.

03 · ENFORCE

Binary Verdict

APPROVED or REJECTED. No partial passes. Machine-readable exit codes integrate directly into CI/CD gates.

04 · CERTIFY

Signed Policy Object

Ed25519-signed SPO issued. Merkle root locked. Independently verifiable forever — no ResEthiq account needed.

The Problem

"Trust me" is not an acceptable answer in regulated AI.

Regulators, auditors, and courts demand a cryptographic chain of evidence. Most AI teams cannot produce it. ResEthiq makes it possible in any pipeline, any industry.

🚨

Regulatory submissions rejected

FDA, EMA, EU AI Act, and sector regulators require auditable data provenance. Most teams fail this requirement entirely.

⚖️

Uncontained legal exposure

Courts require reproducible evidence chains for AI-driven decisions. Without them, liability is open-ended and indefensible.

🔄

Silent dataset mutations

Without a cryptographic seal, you cannot prove a dataset hasn't changed between training and production. Neither can your auditor.

📉

Model failures traced to data

The majority of production AI failures originate in data — wrong splits, label drift, contamination. None detectable without integrity proofs.

resethiq · trust-kernel v1.4.2

ResEthiq Trust Kernel v1.4.2 ────────────────────────────────── $ rk freeze --input dataset.parquet \ --spec policy_v3.yaml --sign rows : 2,341,887 mode : streaming (64 MB chunks) platform : Linux x86_64 ✓ canonical bytes complete ✓ merkle root a3f7c29d14b8e6f2... ✓ inclusion proofs generated ✓ policy policy_v3.yaml ✓ all 14 rules PASS ✓ ed25519 sig applied OUTCOME: APPROVED exit 0 · 2025-03-08T08:43:11Z · spo_v3.cbor

All systems operational ·Trust Kernel v1.4.2 ·Verifier CLI v1.4.2

Core Infrastructure

Four components. One unbreakable chain.

View architecture →

🔐

Trust Kernel

rk · Rust

Deterministic Merkle root computation, canonical encoding, and Ed25519 signing. Streaming mode for 5M+ rows.

RustStreaming

🔍

Verifier CLI

resethiq-verify

100% offline verification. Recomputes Merkle roots, validates Ed25519 signatures. Machine-readable exit codes for CI/CD.

OfflineCI/CD

🗃️

Evidence Ledger

Postgres + CBOR

Content-addressed, immutable object store. Index rebuildable from object storage. No single point of failure.

ImmutableContent-addressed

📋

Audit Bundle

PDF + CBOR + ZIP

Self-contained evidence pack: PDF report, signed SPO, policy digest, public key. Share directly with regulators.

PortableAuditor-ready

Why ResEthiq

Not a quality tool. A cryptographic integrity layer.

Capability	Data Quality Tools	ResEthiq	Manual Audits
Cryptographic proof	✗	✓ Ed25519-signed SPO	✗
Offline verification	✗	✓ 100% offline	✗
Tamper detection	⚠ Alerts only	✓ Mathematically certain	⚠ Manual review
Regulatory evidence	✗	✓ Defensible chain	⚠ Limited
Policy enforcement	✗	✓ Approved / Rejected	✗
Train/test leakage	⚠ Approximate	✓ Exact row-level	✗
CI/CD integration	⚠ Partial	✓ Machine-readable exits	✗

Data integrity infrastructurefor high-stakes AI.