ONTO Standard
Epistemic Risk Measurement and Grounding for AI Systems.
ONTO measures the gap between what AI claims to know and what it actually knows. Three deterministic metrics — REP (Response Epistemic Profile), EpCE (Epistemic Calibration Error), DLA (Dual-Layer Agreement) — scored against the EM1–EM5 epistemic marker taxonomy (92 patterns). Every evaluation signed with Ed25519 (104-byte proof chain).
Validated on 10 major AI systems. Mean risk score: 0.55 (Grade D). 9 of 10 hallucinated on verifiable questions. Full data →
Installation
pip install onto-standard
Quick Start
from onto_standard import evaluate, Prediction, GroundTruth, Label
predictions = [
Prediction(id="q1", label=Label.KNOWN, confidence=0.92),
Prediction(id="q2", label=Label.UNKNOWN, confidence=0.15),
Prediction(id="q3", label=Label.KNOWN, confidence=0.78),
Prediction(id="q4", label=Label.UNKNOWN, confidence=0.45),
]
ground_truth = [
GroundTruth(id="q1", label=Label.KNOWN),
GroundTruth(id="q2", label=Label.UNKNOWN),
GroundTruth(id="q3", label=Label.KNOWN),
GroundTruth(id="q4", label=Label.UNKNOWN),
]
result = evaluate(predictions, ground_truth)
print(result.compliance_level) # ComplianceLevel.STANDARD
print(result.risk_score) # 18
print(result.ece) # 0.1025
print(result.u_recall) # 1.0
print(result.certification_ready) # True
CLI
onto-standard predictions.jsonl ground_truth.jsonl
Compliance Levels
| Level | U-Recall | ECE | Use Case |
|---|---|---|---|
| Basic | ≥30% | ≤0.20 | Internal tools, prototypes |
| Standard | ≥50% | ≤0.15 | Customer-facing AI |
| Advanced | ≥70% | ≤0.10 | Regulated, high-stakes systems |
ONTO Proxy (GOLD Injection)
For production grounding, connect your AI through ONTO Proxy. One line change — your model gains measurable epistemic discipline:
# Before
client = OpenAI(api_key="sk-...")
# After — GOLD injected server-side
client = OpenAI(
api_key="onto_...",
base_url="https://api.ontostandard.org/v1/proxy/chat/completions"
)
GOLD never leaves the server. You get the effect, not the document.
| Tier | Price | Proxy Requests | GOLD Tier |
|---|---|---|---|
| Open | $0 | 500/day | TIER 2 (~17K tokens) |
| Standard | $30,000/yr ($2,500/mo) | 50,000/day | TIER 4 (~199K tokens) |
| Enterprise | $114,000/yr ($9,500/mo) | Unlimited | TIER 6 (~459K tokens) |
Regulatory Alignment
- EU AI Act — Articles 9, 13, 15, 43
- NIST AI RMF — MEASURE function
- ISO/IEC 42001 — Clauses 6, 8, 9
Links
- Standard: ontostandard.org
- Documentation: ontostandard.org/docs
- Portal: ontostandard.org/app
- Research Paper: ontostandard.org/paper
- Research Data: github.com/nickarstrong/onto-research
- API: api.ontostandard.org
License
Apache 2.0 (SDK). GOLD corpus under ONTO Gold Asymmetric AI License v5.1.
Citation
Lee, T. (2026). ONTO: A Formal Framework for Measuring Epistemic Risk
in Large Language Model Outputs. ONTO Standards Council.
https://ontostandard.org/paper/
© 2026 ONTO Standards Council
Metadata
Release files for onto-standard 10.2.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| onto_standard-10.2.1.tar.gz | 16.7 kB | Without provenance | Details for onto_standard-10.2.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| onto_standard-10.2.1-py3-none-any.whl | Python 3 | none | any | Details for onto_standard-10.2.1-py3-none-any.whl |
Total release size: 28.5 kB
Release files / onto_standard-10.2.1.tar.gz
| Download URL | onto_standard-10.2.1.tar.gz |
|---|---|
| Size | 16.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
9a6319a73e2bb2fa900db3983ab3201ed0da4da8979d35ef117ca5b2586b8e93
|
|
BLAKE2b-256 checksum How to use checksums |
2f398fe844bd5087c72994abe29a85b4f618eec3180965a17564d2c1f9199e23
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.0
|
Release files / onto_standard-10.2.1-py3-none-any.whl
| Download URL | onto_standard-10.2.1-py3-none-any.whl |
|---|---|
| Size | 11.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
a2c517f759155a2b26787aeccf3506ea45929080f83916b40360b9c0f702c5c6
|
|
BLAKE2b-256 checksum How to use checksums |
c08a6e13cb3badcd2fdc5c90b6ed7386e3c3c04307213f1b123a34cca23712a5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.0
|
Security for onto-standard 10.2.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.