Cryptographically-signed certificates, public verification endpoints, published methodology, an auditable revocation registry. Everything a Fortune 500 procurement team, auditor, or standards reviewer needs to cite LabelSets in a contract.
The procurement-side verification surface. Any auditor, regulator, or buyer can call them from curl, a browser, or any HTTP client without creating an account.
Look up and verify a cert by its SHA-256 hash. Returns valid / revoked / tampered / not_found with the full payload.
Ed25519 public key in PEM format. Use for offline signature verification in air-gapped environments.
Liveness probe for SLA monitoring. Plug into your uptime dashboard.
Every LQS cert is signed with a single LabelSets Ed25519 keypair. Fetch the public key once, cache it, and re-verify any cert offline without trusting our infrastructure. If we go offline tomorrow, every cert ever issued stays verifiable.
Loading public key from /api/lqs-public-key…
The labelsets pip package ships three operations: online cert lookup, public-key fetch, and fully offline Ed25519 signature verification.
pip install labelsets import labelsets r = labelsets.verify_lqs_cert(cert_hash) print(r.valid, r.composite, r.tier) # True 87 gold
pip install labelsets[offline] import labelsets pk = labelsets.get_lqs_public_key() ov = labelsets.lqs_offline_verify(cert, pk) # True — signature verified locally
Drop-in GitHub Action that fails CI if training data doesn't meet LQS thresholds. Also available for Airflow, MLflow, Weights & Biases, DBT, and Kubeflow.
- uses: labelsets/lqs-verify@v1 with: cert_hash: ${{ vars.TRAINING_DATA_LQS_CERT }} min_score: 85 min_ci_low: 82 reject_contamination_tier: contaminated,severe reject_oracle_tier: brittle,divergent
A cert is a canonical-JSON payload + base64 Ed25519 signature + SHA-256 payload hash. Sorted-keys + whitespace-free UTF-8 canonicalization is identical between signer and verifier.
{
"v": "1.0",
"issuer": "Labelsets LLC",
"methodology_url": "https://labelsets.ai/lqs-methodology#v3.1",
"dataset_id": "d9e75396-2629-44dc-a609-4280d5571038",
"scored_at": "2026-04-23T00:59:45.219Z",
"scorer_version": "3.1",
"composite": 87,
"composite_ci_95": { "low": 85.2, "high": 88.8, "se": 0.9 },
"tier": "gold",
"task_conditional": { "task": "binary_classification", "task_source": "data_driven" },
"oracle_consensus": { "agreement_score": 92, "tier": "consensus", "kappa_fleiss": 0.84 },
"contamination": { "score": 95, "tier": "clean", "worst_benchmark": "MMLU" },
"downstream_projection": { "tier": "learning", "headroom_score": 45 }
}
Legal and ML teams: we'll walk you through the Ed25519 cert, the methodology whitepaper, BAA/DPA templates, and how to reference LQS in your SR 11-7, EU AI Act Art. 10, or HHS §1557 paperwork.