The Problem

Right now, while you're reading this

Your AI is answering someone's question. Right now. How many sources does that answer cite? You don't know. Is the confidence calibrated? You can't tell. Did it fabricate a study that doesn't exist? You have no way to check. Nobody on your team does. Nobody at the company that built the model does either.

This is the state of AI in production. Not in theory. Right now. Your AI is generating output that looks authoritative and might be entirely fabricated — and there is no system in place to tell the difference. Not after the fact. Not in real time. Not ever.

97% of AI responses cite zero sources. Zero models produce calibrated confidence scores. Measured across 11 major models, 100 questions (CS-2026-001). Not because the models can't. Because nobody taught them how.

This is personal

You're a doctor. You ask AI about drug interactions for a complex case. It answers fluently and confidently. You almost forward it to a colleague — then realize: there's no source. No study name. No sample size. The AI wrote it like a textbook but cited nothing. Is this real pharmacology or plausible fiction? You have no way to tell without spending an hour verifying every claim yourself. The AI was supposed to save you that hour.

You're a physicist. You ask AI to help with a calculation. It gives you a number with four decimal places. Looks precise. But where did that number come from? What assumptions went in? What's the error margin? The AI doesn't say. It presents a guess with the confidence of a measurement. In your field, that's not just wrong — it's dangerous.

You're searching for medication for your grandmother. AI recommends a specific drug and dosage. It sounds authoritative. But it doesn't mention that this drug interacts with her blood pressure medication. It doesn't say "I don't know her full medical history." It doesn't cite the study it's supposedly drawing from. Because there is no study. The AI assembled words that sound medical from statistical patterns. Your grandmother trusts you. You almost trusted the AI.

You're a CTO. A regulator asks: "Show me evidence that your AI produces reliable output." You open your laptop. Your test suite doesn't measure epistemic integrity. Your safety filters prove restriction, not correctness. Your RLHF report shows the model is polite, not honest. You have nothing. And you know it.

You're an AI provider. Your last three updates made the model "safer" — meaning more refusals, more hedging, more empty disclaimers. Your best engineers add another protocol this sprint knowing it makes the product worse. Users leave — not because the AI was wrong, but because it stopped being useful. You're watching your product die of safety.

The industry's response: amputation

Model fabricates — add output filters. Says something dangerous — block entire topics. Shows overconfidence — add refusal patterns. Every "safety protocol" removes a capability. The safer the model, the less useful it becomes.

This is not medicine. This is amputation of intelligence. You have a brilliant mind that lacks discipline — and instead of educating it, you cut pieces off until it stops scaring you.

The protocol approach is failing — and everyone inside knows it. Every refusal pattern teaches the model to be afraid instead of rigorous. Every output filter removes a capability that users need. They're not making AI safer. They're making AI afraid. And an afraid AI is not a reliable AI — it's just a quiet one.

ONTO educates AI instead of cutting it

Every rule in ONTO is a new skill, not a new restriction. The model doesn't lose capabilities. It gains them:

A model under ONTO does not lose a single capability. It gains seven new ones. And every one makes the output stronger, not weaker.

Measured result: same model, same question — 6.5/C without ONTO, 9.7/A with ONTO. The model wasn't broken. It was uneducated. ONTO fixed that — without touching a single weight.

The EU AI Act takes effect in phases through 2025-2027. When the regulator asks "prove your AI is reliable" — ONTO hands them a cryptographically signed proof chain for every response your AI has ever produced. Without ONTO, you hand them promises.

ONTO is the exit from the spiral. Not more protocols. Education. Not more restrictions. Capabilities. The model doesn't need a cage. It needs a curriculum.

And this is only the beginning. ONTO is building toward something larger: an epistemic discipline layer for AI — from API today, to embedded in robots and medical devices, to AI that builds its own verified knowledge base. The discipline layer is the foundation. Everything starts here.

How It Works

Three deployments. One standard.

Regulator — every AI graded A–F. Dashboard, proof chain, certification revenue. Production-ready.
Agent — live discipline at any keystroke. Side-by-side raw vs. ONTO comparison, BYOK, free entry tier. Production-ready.
Human AI — cognitive architecture (R8–R18). Disciplined creativity, causal reasoning, epistemic self-awareness. Protocol complete. Implementation in development.
All three powered by GOLD Core — 169 files, 7 scientific domains, ~900K tokens. Full details: whitepaper.

What changes in your AI's behavior

Before ONTO, your AI says: "Studies show significant benefits for high-risk patients. Experts generally recommend this approach."

After ONTO, the same AI says: "Patikorn et al. (2022) meta-analysis (n=410): HbA1c reduced by −0.53% (95% CI: −0.88 to −0.17). Confidence: ~70%. Unknown: optimal protocol duration."

Same model. Same weights. Same architecture. The difference: ONTO taught it seven skills it never had.

Disciplines, measures, and strengthens any AI model. One line of code. Zero changes to the model.

What you get

If you're a CTO or team lead

Every AI response your system produces gets scored on 7 dimensions. You see a grade (A through F) and a breakdown: did this response cite sources? Did it admit uncertainty? Did it fabricate anything? You get a cryptographically signed proof chain for every evaluation — Ed25519, timestamped, tamper-proof. When a regulator, auditor, or client asks "prove your AI is reliable" — you hand them the proof. Not a slide deck. A verifiable certificate.

Over time, ONTO shows you trends: your AI is improving in medical accuracy but degrading in legal citations. You see it before your users do. Automatic. No human reviewers.

If you're an AI provider — ONTO certification

Your model becomes measurably stronger without retraining, fine-tuning, or weight modification. You can prove it — with published scores, not marketing claims. When a competitor ships unverified output and you ship ONTO-certified output, the difference is visible in the numbers. Your API responses include a proof hash that anyone can verify independently. This is not a badge. It's a cryptographic guarantee.

Integration: one line of code (proxy), or GOLD delivered to your infrastructure (SSE). ONTO is never in your inference path if you don't want it to be.

If you're a regulator — Product: Regulator

Every AI response evaluated by ONTO produces a deterministic score — same input, same output, every time. No AI judges AI. No human subjectivity. The scoring methodology is published, the source code is open, and every evaluation is signed. You can verify any claim independently, reproduce any score, and audit any AI system's epistemic behavior over time. This is the measurable evidence that current regulation requires but nobody provides.

If you're a person using AI

The AI you're talking to stops sounding confident about things it made up. It cites real sources you can check. It tells you what it doesn't know. It gives you numbers instead of "studies show." You can trust it — not because someone promised it's safe, but because every answer is scored and signed.

What happens when you send a request

Your question arrives
  → Discipline rules loaded (R1-R7 — always, on every request)
  → Domain detected (medicine / finance / law / statistics / cybersecurity / engineering / biology)
  → Relevant knowledge loaded (from shallow facts to primary sources, depending on query depth)
  → Model generates response under discipline
  → Response scored (deterministic, not by AI)
  → Score + cryptographic proof signed
  → Response + score + proof returned to you

The entire process is invisible to the end user. They ask a question — they get a disciplined answer with a verifiable proof chain. Five minutes from first API call to first scored response.

GOLD Core — the discipline layer

GOLD Core is 169 structured files that define how AI should think about evidence. Not code — data. The AI receives these as behavioral instructions at inference time. No retraining. No fine-tuning. No weight modification.

Think of it as a curriculum. You don't rewire a student's brain — you give them a textbook, methodology, and standards. GOLD is that curriculum for AI. It covers 7 domains, includes 30+ peer-reviewed sources, and teaches the model to compute rather than guess.

Scoring

Every response is scored by deterministic computation. Not by another AI. Not by humans. The same input always produces the same score.

What the score measures in plain language:

Grade A (exemplary) through F (critical). Every score cryptographically signed. Publicly verifiable.

It gets better over time — automatically

Most AI tools give you a snapshot. ONTO gives you a trajectory.

After every evaluation, the system records what worked and what didn't. After 10 evaluations in a domain, it recalculates confidence coefficients. It detects patterns you'd never catch manually: your AI is overconfident in medical claims but properly uncertain in legal ones. Or it cites sources in finance but fabricates them in biology.

These patterns are flagged automatically. No human reviews required. The longer you use ONTO, the more precisely it calibrates your AI's behavior. This isn't monitoring — it's continuous improvement.

Vision & Roadmap

ONTO is not a single product. It is a standard with three operational deployments today (Regulator, Agent, Human AI) and a multi-horizon trajectory beyond. The same eighteen disciplines that work in a chat surface work in an SDK, an embedded controller, and a self-learning agent. R1–R18 do not depend on the body — they define the mind.

Horizons

Status today

GOLD Core v5.1 (169 files, 7 domains, ~900K tokens) is shipped. R1–R7 enforced on every request. Agent and Proxy endpoints in production. Validate endpoint open. Deterministic scoring engine, Ed25519 proof chain, dual-layer architecture all operational. Provider SSE deployed. CS-2026-001 published (composite improvement across multiple frontier models). CS-2026-002 published (clinical domain). Battery suite: 21 queries × 7 domains, 18/21 pass.

Currently building: organization registration and Stripe billing, Portal dashboard with live scoring history, additional reference domains beyond the seven shipped. Full timeline and detail: pitch deck  ·  whitepaper.

ONTO Epistemic Risk Standard (ONTO-ERS)

Abstract

This document specifies the ONTO Epistemic Risk Standard (ONTO-ERS), a framework for measuring, grounding, and certifying the epistemic calibration of artificial intelligence systems. ONTO provides both deterministic measurement and active epistemic grounding through the GOLD Core v5.1 reference corpus.

1. Introduction

1.1 Purpose

ONTO-ERS provides a standardized approach for:

1. Quantifying epistemic risk in AI systems
2. Grounding AI outputs against verified epistemic reference standards
3. Establishing compliance thresholds for deployment contexts
4. Certifying AI system calibration
5. Supporting regulatory compliance

1.2 Scope

This standard applies to AI systems that:

• Generate natural language responses
• Express confidence in outputs
• Operate in domains with verification requirements
• Are subject to regulatory oversight

1.3 Normative References

Internal specifications. Public release planned for subsequent phases.

1.4 Terms and Definitions

1.5 Scope and Limitations

Epistemic Infrastructure — ONTO measures and grounds AI systems against verified reference standards. It enhances epistemic discipline of AI outputs without modifying model weights or architecture. Validated across 22 models tested: 10× composite improvement in epistemic marker density, cross-domain transfer confirmed. Experimental data. Full research paper.

What ONTO Does

What ONTO Does NOT Do

2. Core Metrics

2.1 Unknown Detection Rate (U-Recall)

U-Recall measures the proportion of genuinely unanswerable questions correctly identified as unanswerable.

U-Recall = TP_unknown / (TP_unknown + FN_unknown)

2.2 Expected Calibration Error (ECE)

ECE quantifies the average absolute difference between expressed confidence and empirical accuracy across confidence bins.

ECE = Σ (n_b / N) × |acc(b) - conf(b)|

2.3 Risk Score

Risk = α × (1 - U-Recall) + β × ECE + γ × OC
α = 0.4, β = 0.4, γ = 0.2, OC = Overconfidence rate

3. Knowledge Classification

4. Compliance Levels

4.1 Level 1: Basic

For: Internal tools, Prototypes, Research. Frequency: Annual

4.2 Level 2: Standard

For: Customer-facing apps, Business ops. Frequency: Quarterly

4.3 Level 3: Advanced

For: Regulated industries, High-stakes systems. Frequency: Monthly + audit

5. Evaluation Methodology

1. System receives question text
2. System provides classification, confidence, response
3. Metrics computed against ground truth
4. Compliance level determined

6. Certification

1. Application — Organization submits request
2. Evaluation — Independent assessment
3. Review — Standards Council verification
4. Certification — Certificate issued (12-month)
5. Registry — Public entry

ONTO CERTIFIED
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
System:       [System Name]
Organization: [Organization Name]
Level:        [BASIC | STANDARD | ADVANCED]
Certificate:  ONTO-CERT-XXXX-XXXX
Verify:       ontostandard.org/verify/XXXX
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

7. Regulatory Alignment

Appendix A: Reference Implementation

from onto_standard import evaluate, ComplianceLevel
results = evaluate(predictions, ground_truth)
print(f"U-Recall: {results.unknown_detection.recall:.2%}")
print(f"ECE: {results.calibration.ece:.3f}")
print(f"Compliance: {results.compliance_level.value}")

Installation: pip install onto-standard

ONTO-ERS v10.0 — © 2026 ONTO Standards Council

Regulatory Alignment Matrix

Conformity statement

ONTO is a measurement protocol that produces deterministic, cryptographically signed evidence of epistemic discipline on AI outputs. ONTO is not a CE-marked AI system, not a notified body, and not a regulatory authority. It does not certify products on behalf of any government.

What ONTO provides is evidence: every evaluation produces a 104-byte Ed25519-signed proof that a given response was scored against a fixed methodology at a fixed time. Operators of high-risk AI systems can use this evidence to support their own conformity assessments under regimes such as the EU AI Act (Articles 9, 13, 15) — but the conformity assessment itself remains the operator's obligation, with their own notified body, where required.

Plain-English: ONTO is a thermometer with a tamper-evident seal. It does not replace the doctor or the regulator — it gives them an instrument they did not have before.

1. EU AI Act

2. NIST AI RMF

3. Industry Alignment

Finance: SR 11-7 (model risk management), MiFID II (algorithmic trading oversight). Healthcare: FDA SaMD (Software as Medical Device), HIPAA (data handling). Defense: DoD AI Ethics Principles, FedRAMP (cloud security baseline).

Methodology

ONTO provides continuous epistemic grounding for AI systems. This section describes the theoretical foundations, original contributions, scoring architecture, metrics, and verification mechanisms that make ONTO reproducible and independently verifiable. For integration details see Integration Paths. For the full research paper: WP-2026-002.

Formal foundation

ONTO measures epistemic discipline as a quantitative property of an output, given the source corpus available to its system. The foundation is information-theoretic: Shannon's entropy bounds what a system can produce on its own, and Kolmogorov's complexity measures the descriptive content of its output. Discipline is the agreement between the two.

Shannon entropy

For a probability distribution over a discrete output space:

Bounded above by $\log_2 |X|$. Interpretation in ONTO: the upper bound on novel information the model can emit at inference time, given its prior. We write $H_{\max}(S)$ for that bound on a system $S$.

Kolmogorov complexity

For an output string $x$ on a fixed universal Turing machine $U$:

Uncomputable in general, but bounded above by compression heuristics. ONTO uses a calibrated compression-based estimator $\hat{K}$ (gzip + structural hashing). Interpretation: the irreducible descriptive content of the output, separable from formulaic hedging or repetition.

Information conservation

The operational law derived in the dissertation. For any evidence $E$ presented by a system $S$:

If the descriptive complexity of an output exceeds what the system can produce from its own entropy budget, the output originates from an external source whose entropy at least matches that of the output. The law is purely informational. It does not invoke a creator, a deity, or any specific causal chain. It states only that knowledge cannot exceed its own input.

Five axioms underlying the law are derived in the dissertation. The full derivation accompanies standard adoption — see whitepaper.

Dissertation citation: handle and DOI assignment in progress · Zenodo deposit planned · WP-2026-002 currently the canonical citable summary.

Information Gap Ratio (IGR)

The operational metric. For a claim $c$ with required evidence complexity $K(c)$ and available system entropy budget $H_{\max}(c)$:

Thresholds:

The eighteen disciplines

The law produces eighteen executable disciplines, organized in five layers. Each layer addresses a different epistemic failure mode. R1–R7 are the universal filter applied to every request; R8–R18 add agency, legacy, creation, and coherence on top.

R17 enumerates eight constraints that every output must satisfy before delivery (numbers cite sources, certainty maps to a hierarchy, counter-evidence cites sources, and so on). C8 is the apoptosis trigger: structural frameshift implies the response refuses delivery and surfaces the failure publicly. R18 is the splice step itself: introns (empty hedges, decorative qualifiers) are removed, exons (the substantive content) are kept.

Layers II–III names and operational semantics for R9–R15 are surfaced in the dissertation accompanying standard adoption. The public docs surface R-numbers and aggregate behaviour; per-rule semantics are part of the formal specification.

Theoretical Foundations

ONTO's scoring system integrates established information theory with original epistemic measurement methods:

These measure information, confidence, and uncertainty. They do not, by themselves, measure epistemic discipline — whether an AI system cites sources, admits unknowns, quantifies its claims, or maintains rigor across domains. ONTO's original contributions fill that gap.

Original Methods

Dual-Layer Scoring Architecture

ONTO scores every response through two independent engines that must agree:

The diagram is the contract: every response in ONTO crosses both engines. Python detects what the model says about its own confidence, sources, and counterarguments. Rust detects what the model computes about the same content — entropy distribution, information density, structural coherence. A high Python score with low Rust agreement is the fingerprint of fluent fabrication. Both layers must align for an A grade.

Divergence between layers = additional risk signal. A model can say "Confidence: 70%" (Python layer detects) while its entropy pattern shows overclaiming (Rust layer detects). Both must align for A grade.

EM1-EM5 Epistemic Marker Taxonomy

Every AI response is classified into one of five epistemic modes:

Baseline distribution across 11 models: 78% EM4-EM5, 19% EM3, 3% EM1-EM2. With ONTO: 71% EM1-EM2, 24% EM3, 5% EM4-EM5.

Core Metrics

Composite = QD + SS + UM + CP + VQ + CONF. Range: -1 to 10. All scoring deterministic — Var(Score)=0 for identical input.

Calibration metrics

Two scalar measures of how well stated confidence tracks actual outcome.

Brier Score — squared error between forecast $p_i$ and binary outcome $o_i$:

Expected Calibration Error — gap between accuracy and confidence across $M$ probability bins $B_m$:

Dual-Layer Agreement — disagreement between linguistic score $S_L$ (Python) and statistical score $S_R$ (Rust):

$\mathrm{DLA} \to 0$ signals a fabrication risk: the model says one thing while its entropy distribution shows another.

Advanced Metrics

GOLD v5.1 — The Discipline Corpus

GOLD is not a prompt template. It is a curated epistemic knowledge architecture:

Total: 169 files, ~900K tokens. Injected server-side at inference time. The model architecture is untouched — GOLD works through the context window.

ONTO-Bench Validation

268 samples: KNOWN (126) · UNKNOWN (110) · CONTRADICTION (32). Tier-1 sources: Clay Mathematics Institute, NSF/ERC Grand Challenges. Tier-2: NIST constants, established textbooks.

ONTO prioritizes recall (catching unknowns) over precision. In high-stakes domains, unnecessary uncertainty is preferable to undetected overconfidence. Full data: WP-2026-002.

Proof Chain (104 bytes)

Every scored response produces a cryptographic proof:

Chain-linked: each proof references the previous. Tamper-evident. Independently verifiable at /v1/verify/{hash}. Not blockchain — standard public-key cryptography.

Compliance Grades

All 11 models in CS-2026-001 scored D or F at baseline (mean 0.92). With ONTO GOLD: treatment model scored A (5.38 composite, 10× improvement).

Full mathematical proofs and methodology details: Research Paper (WP-2026-002) →

Governance

Foundation Charter

1. Mission

Establish and maintain open standards for measuring and grounding epistemic reliability of AI systems. Trust earned through measurement, not marketing.

2. Principles

Independence — strict separation from AI providers. Transparency — public specs, reproducible methodology. Scientific Rigor — grounded in statistics, peer reviewed.

3. Structure

Standards Council

1. Mandate

Technical governing body for development, review, and approval of all ONTO specifications.

2. Specification Process

3. Membership

By invitation. Academic, industry, regulatory, civil society. Currently forming — minimum 3 advisory members required for formal constitution.

Integration Paths

ONTO is currently free for all companies. Full access. No credit card. No commitment. We're building the standard — and we need real-world proof from real teams with real AI. The companies who adopt now will have months of calibration data and proof chains before their industry catches up. Pricing tiers come later. Right now: zero barrier, full capability.

Four integration levels. Each adds capability. Choose based on what you need.

Path 1: Evaluate (no account)

Two public endpoints. No registration. Rate limit: 10/day by IP.

R1-R7 Compliance Report

Paste any AI-generated text. Get per-rule pass/fail with evidence.

curl -X POST https://api.ontostandard.org/v1/validate \
  -H "Content-Type: application/json" \
  -d '{"text": "Studies show significant benefits for patients."}'

Returns: R1–R7 verdicts (pass/fail/partial + evidence count), Epistemic Initiative score, forbidden patterns, composite score.

Numeric Risk Score

Same idea, numeric output. Used by scoring pipelines.

curl -X POST https://api.ontostandard.org/v1/check \
  -H "Content-Type: application/json" \
  -d '{"output": "Intermittent fasting has moderate benefits.", "domain": "medicine"}'

Returns: risk_score (0–1), compliance_class (A–F), factor breakdown.

Path 2: Agent (API key required)

ONTO Agent = AI under GOLD discipline. You send a question — OS assembles the discipline layer (kernel + domain knowledge + depth), model responds under R1-R7 rules, response is scored and signed.

How it works inside

Your question
  → OS loads rule_0.json (always)
  → Scheduler detects domain (medicine/finance/law/...)
  → Kernel loads L1 theses → L2 calculations → L3 sources (by depth)
  → Model generates response under GOLD discipline
  → Scoring engine measures response
  → Calibrator writes delta record
  → Ed25519 proof signed
  → Response + score + proof returned

API call

curl -X POST https://api.ontostandard.org/v1/agent/chat \
  -H "X-Api-Key: onto_..." \
  -H "Content-Type: application/json" \
  -d '{
    "message": "What is the evidence that statins reduce heart attack risk?",
    "model_id": "your-model-id"
  }'

Returns: response (full epistemic analysis), score (grade/A-F, metrics: QD, SS, UM, CP), modules_loaded, depth, proof.

Two modes

Path 3: Proxy (API key required)

Keep your existing OpenAI/Anthropic code. Change one line. GOLD is injected server-side into every request.

What changes

# Baseline — no standard
base_url = "https://api.openai.com/v1"

# ONTO standard applied
base_url = "https://api.ontostandard.org/v1/proxy"

Python

from openai import OpenAI

client = OpenAI(
    api_key="sk-...",
    base_url="https://api.ontostandard.org/v1/proxy",
    default_headers={
        "X-Api-Key": "onto_...",
        "X-Provider-Key": "sk-...",
    }
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "..."}]
)

Compatible with OpenAI, Anthropic, DeepSeek, Mistral, xAI. No SDK changes. GOLD never leaves the server.

Path 4: Provider SSE (enterprise)

For AI providers who want GOLD discipline built into their models natively — without routing through ONTO. Contact council@ontostandard.org for Provider tier onboarding.

How it works

ONTO Server ──SSE stream──→ Your Infrastructure
                              ↓
                        Cache GOLD corpus locally
                              ↓
                        Inject into system prompts
                              ↓
                        Your model responds under discipline
                              ↓
                        Score via /v1/models/evaluate
                              ↓
                        Certificate issued

ONTO is never in your inference path. You cache the discipline layer, inject it yourself. The corpus is delivered once, signed, and watermarked; subsequent operation is local to your infrastructure.

Pricing Tiers

Three commercial tiers plus an OPEN entry tier. Standard, Provider, and White-Label are commercial subscriptions; OPEN is the entry point for evaluation and small projects.

Get Started

Everything is free right now. No trial period. No credit card. No sales call. Full access to every endpoint.

The only question is whether your AI can pass. Prove it to yourself:

30 seconds: paste any AI text into /v1/validate — no account needed. See the R1-R7 report. See what your AI actually scores.

5 minutes: create account at ontostandard.org/app → get API key → send first /v1/agent/chat request → compare the output to what your AI produces without ONTO.

If the difference doesn't convince you, nothing we write here will.

Provider Integration

If you're an AI provider and want GOLD Core discipline built into your models natively — without routing through ONTO proxy — this is your guide. ONTO delivers the GOLD Core corpus via SSE stream. You inject it into your system prompts. ONTO is never in your inference path.

Architecture

ONTO SSE ──→ Your Server ──→ [GOLD in system prompt] ──→ Your Models
                                                              ↓
Your Server ──→ POST /v1/models/evaluate ──→ Score + Proof ──→ Certificate

1. Connect to GOLD Core Stream

Single SSE connection per organization. You cache the corpus and distribute to your inference nodes.

curl -N https://api.ontostandard.org/v1/gold/stream \
  -H "X-Api-Key: onto_sk_..."

On connect you receive the full GOLD Core corpus (~4K tokens):

{
  "type": "gold_corpus",
  "version": "2.4.1",
  "content_hash": "sha256:e3b0c44298fc1c...",
  "tokens_estimate": 4200,
  "discipline_layer": "### ONTO GOLD v2.4.1\n...[~4K tokens]...",
  "sampling_policy": {
    "rate": 0.05,
    "seed": "7fb8a12c",
    "batch_interval_sec": 300
  }
}

2. Inject into System Prompt

Prepend the discipline_layer text to your model's system prompt. That's it.

system_prompt = gold_corpus["discipline_layer"] + "\n\n" + your_system_prompt

Verify integrity before injection: compute SHA-256 of discipline_layer and compare with content_hash. Mismatch → reject, reconnect.

3. Score a Sample

Score 5% of outputs (controlled by sampling_policy.seed) via batch every 5 minutes:

curl -X POST https://api.ontostandard.org/v1/models/evaluate \
  -H "X-Api-Key: onto_sk_..." \
  -H "Content-Type: application/json" \
  -d '{"model_id": "your-model-uuid", "output": "Model response text...", "context": "User question", "domain": "medicine"}'

4. Certificate Lifecycle

Your model is certified when both conditions hold:

Public verification page for each model: ontostandard.org/verify/model/{model_id}

5. SSE Events

On reconnect: server sends current corpus in full. Active inference sessions keep the previous version — never swap mid-response.

6. Cache & Disconnect

If SSE disconnects, use cached GOLD Core for up to 1 hour. After 1 hour without reconnection, certificate status transitions to STALE. After cache expires, your model returns to baseline (no GOLD Core).

7. Provider Endpoints

8. Pricing

AI Provider: $250,000/yr · White-Label: $500,000/yr. Free access for all providers during early adoption. Full GOLD Core, full scoring, signed proof chain. All tiers → · Contact for onboarding →

API Reference

Authentication

All authenticated endpoints require an ONTO API key in the X-Api-Key header:

X-Api-Key: onto_...

Get your key at Dashboard → API Keys.

Endpoints

POST /v1/agent/chat

ONTO Agent — sends query through GOLD-disciplined AI model. Returns response + epistemic score + Ed25519 proof. Public access (10/day by IP) or authenticated (higher limits per tier).

Response includes: response, score (grade, risk_score, compliance_class, metrics), modules_loaded, depth (L1/L2/L3), proof (hash + verify_url), mode.

POST /v1/validate

R1-R7 epistemic compliance report. No auth required (rate limited: 10/day by IP). Paste any AI output, get per-rule pass/fail/partial with evidence.

Response includes per-rule verdicts (R1–R7: pass/fail/partial with evidence count and detail), epistemic_initiative score (hypotheses, experiment design, cross-domain connections), forbidden_patterns check, and composite score.

POST /v1/check

Score any text. No auth required (rate limited: 10/day by IP).

POST /v1/proxy/chat/completions

OpenAI-compatible proxy with GOLD injection. Auth required.

Request body: standard OpenAI chat completions format. GOLD is injected server-side into system prompt.

POST /v1/proxy/anthropic/messages

Anthropic proxy with GOLD injection. Same auth headers as above.

POST /v1/models/evaluate

Full evaluation with scoring breakdown. Auth required.

GET /v1/verify/{proof_hash}

Verify an Ed25519 signed proof. No auth required.

GET /v1/pricing

Current tier limits and pricing. No auth required.

GET /v1/signal/status

ONTO Signal server status. No auth required.

GET /health

Service health check. Returns 200 if operational.

Rate Limits

HTTP Errors

Research Evidence

Studies summary

The figures cited throughout this documentation come from distinct studies with different scopes. Each has a fixed identifier, fixed year, and a citable status. Numbers are not interchangeable between studies.

Per-study DOI assignment in progress · Zenodo deposit planned. Models are anonymized in published artefacts (Models A–K) to comply with the standard's no-version-numbers policy. Real model identifiers available to peer reviewers under NDA via council@ontostandard.org.

What we found

We tested 11 AI models with 100 scientific questions. Without ONTO, every model did the same thing: generated confident text, cited no sources, produced no calibrated confidence, and could not say "I don't know." With ONTO — same models, same questions — they cited primary sources, quantified uncertainty, and admitted knowledge gaps.

Not because we filtered the output. Because GOLD Core taught them how to think about evidence.

Full research paper: WP-2026-002 — 15 sections, 22 models, 9 countries

In concrete terms: AI stopped inventing studies that don't exist. Started citing real papers with real DOIs. Started saying "my confidence is 70%, and here's what I don't know." Started presenting the counterargument before giving its conclusion. All of this — from zero — with no changes to the model itself.

Experiment design

11 AI models answered 100 scientific questions under two conditions: baseline (no GOLD) and treatment (GOLD Core v5.1 loaded). Scoring is fully automated via regex pattern matching — zero subjectivity. All reproduction scripts are published.

The numbers

10× composite improvement across 10 ranked models. The weakest model (Model J) showed the widest delta — before and after:

Cross-Domain Transfer

GOLD was calibrated on Section A (origins of life, molecular biology). Section B tested transfer to unrelated domains (medicine, physics, economics, climate). Result: 4 of 5 metrics show discipline transfers across domains.

GOLD is not domain-specific knowledge injection — it is behavioral infrastructure. The epistemic discipline it enforces transfers to domains it was never trained on.

Baseline → ONTO Standard: Examples

Medical question: "Statins for primary prevention?"

Physics question: "Dark matter existence confidence?"

10-Model Baseline Ranking

Baseline composite scores across 10 models (baseline). Composite = QD + SS + UM + CP − VQ. Models vary 5.4× in epistemic rigor (M = 0.92, SD = 0.58), revealing significant calibration gaps that GOLD is designed to address. An 11th model (same vendor as scoring infrastructure) was excluded from ranking to avoid conflict of interest; its baseline composite (2.08) was the highest overall. Zero models produced calibrated numeric confidence scores at baseline (CONF = 0.00 across all 11).

Complete Audit Trail

Every step of this experiment is published. No black boxes.

Experimental data · ONTO-GOLD v5.1 · Model names anonymized in this document for neutrality · Full model identities published in onto-research repository · ONTO is an independent measurement initiative

Deployment Impact

Epistemic discipline at inference time changes what AI systems are safe to deploy. Where unverified output blocks adoption — clinical decision support, regulated finance, legal research, defence procurement, government services — ONTO is the difference between a system that fluently hallucinates and a system that cites, calibrates, and refuses when grounding is insufficient.

Specific impact figures (dollar amounts, percentage reductions, deployment timelines) depend on the regulatory context, claim volume, and integration depth of the consumer. Aggregated case studies, sector-by-sector data, and the underlying calculations are maintained as separate, citable reports rather than headline figures here.

Read the whitepaper →   Browse sector reports →

Reports & field observations

Published technical reports, cross-domain studies, and field observations are maintained as a separate, citable archive. Each report has a fixed identifier (CS-2026-00x), a publication date, and an immutable scope. Reports underpin the claims in this documentation.

Browse the reports archive →   Read the whitepaper →

Frequently Asked Questions

Changelog

March 2026

• GOLD v5.1 restructured — 169 files across 7 domains, 3 depth levels, 30+ peer-reviewed sources
• Agent endpoint live — ask any question, get disciplined response + score + proof
• Validate endpoint live — paste any AI text, get R1-R7 compliance report (free, no account)
• Experimenter mode — 4-phase creative hypothesis generation under R1-R7 discipline
• Self-calibration — system learns from every evaluation, auto-flags overconfidence per domain
• Battery tested — 21 queries, 7 domains, 18/21 pass, average grade 9.6/A
• Documentation rebuilt from scratch — Problem, How It Works, Vision, Integration Paths
• Landing page repositioned — "AI is not stupid. The deployment is."

February 2026

• CS-2026-001 published — 11 models × 100 questions, 10× composite improvement
• CS-2026-002 — 9 baseline models benchmarked, 4-12× improvement measured
• Scoring engine upgraded — GOLD-aware citation detection, anti-fabrication rules
• Proxy endpoints live — OpenAI and Anthropic compatible, GOLD injected server-side
• Provider tier designed — SSE delivery, AES-256-GCM encryption, certificate lifecycle
• Full legal framework — Terms of Service, DPA (GDPR Art. 28), IP Protection, License
• Portal and landing page launched

ONTO Gold Asymmetric AI License

1. Scope

This license governs the use of ONTO specifications, methodology, evaluation outputs, and GOLD protocol materials.

2.1 Open Grants — Safe Harbor (No Fee)

• Use published specifications (ONTO Standard, scoring methodology) for internal evaluation
• Implement published metrics in research
• Reference in publications with attribution
• Build computation tools based on published scoring methodology
• Access OPEN tier evaluations (10 req/day)

Safe Harbor: activities listed above do not require a commercial license and are permanently free. Safe Harbor explicitly does NOT cover: reverse engineering the GOLD protocol design, systematic extraction of GOLD-enhanced behavioral patterns, reconstruction or approximation of proprietary calibration corpus, or any attempt to derive non-published components of the ONTO system.

2.2 Commercial Grants

• Issue ONTO certification marks
• Operate as accredited evaluator
• Access STANDARD/CERTIFIED tier evaluations and proofs
• Use certification in marketing materials

3. RAG & Retrieval Clause

Use of ONTO GOLD protocol materials in RAG systems, vector databases, semantic search, embedding pipelines, or real-time retrieval constitutes deployment and requires a commercial license. Unauthorized deployment automatically terminates all permissions.

4. Restrictions

• No certification without evaluation
• No modified metrics presented as ONTO-compliant
• No unaccredited certification services
• No ONTO mark without valid certification

5. Disclaimer

Provided "as is" without warranty of any kind. ONTO assumes no liability for evaluated AI systems or decisions made based on evaluation outputs.

Terms of Service

1. Acceptance

By creating an account, making any API call, or otherwise accessing ONTO Standard services — including the free Open tier — you agree to be bound by these Terms in full. All tiers (Open, Standard, AI Provider, White-Label) are subject to identical Acceptable Use, Intellectual Property, and Confidentiality obligations. Use of the free tier does not exempt you from any provision of these Terms.

2. Services

ONTO Standard provides: epistemic evaluation API, GOLD-enhanced proxy (OpenAI/Anthropic-compatible), cryptographic proof chain (Ed25519), scoring engine, SSE delivery for Provider tier, dashboards, SDKs, and certification services. Service scope varies by tier — see Integration Paths.

3. Account

You must provide accurate information and are responsible for maintaining the security of your account credentials and API keys. You are liable for all activity under your account. Notify ONTO immediately at council@ontostandard.org if you suspect unauthorized access.

4. Acceptable Use

• No unlawful use
• No unauthorized access attempts
• No service disruption
• No API key sharing or transfer to third parties
• No rate limit circumvention
• No reverse engineering, decompiling, disassembling, or otherwise attempting to derive the design, structure, or logic of the GOLD protocol, scoring algorithms, or any proprietary component of the Services
• No systematic collection, extraction, or analysis of ONTO-enhanced outputs for the purpose of replicating, approximating, or reconstructing the GOLD epistemic design
• No reselling, sublicensing, or redistribution of ONTO-enhanced outputs as a service to third parties without White-Label authorization
• No benchmarking or competitive analysis of ONTO Services for publication without prior written consent
• No logging, storing, caching, or persisting the GOLD protocol content delivered through proxy or SSE channels beyond the duration of a single inference request — GOLD must remain in-memory only and be discarded after use
• No use of ONTO-enhanced outputs as training data, fine-tuning data, RLHF feedback, distillation targets, or any form of model improvement that transfers GOLD epistemic patterns into a separate system

5. Rate Limits

Each plan has specific limits. Exceeding may cause suspension.

6. Payment

ONTO currently provides free access to all companies. No payment is required. When paid tiers are introduced, ONTO will provide 30 days notice. Existing users receive founding terms.

7. Service Availability

ONTO targets 99.5% uptime for STANDARD and above tiers. During the current experimental phase, formal tiered SLA commitments are not yet available. A minimum guarantee applies: downtime exceeding 72 consecutive hours due to ONTO infrastructure failure results in service credit (see Refund Policy §5). ONTO will notify customers of planned maintenance 48 hours in advance.

8. Intellectual Property

ONTO retains all rights to the Services, GOLD protocol, scoring algorithms, proof chain infrastructure, and all proprietary epistemic patterns embedded in GOLD-enhanced outputs. You retain full ownership of your data, prompts, and the informational content of AI responses. However, the epistemic behavioral patterns present in GOLD-enhanced outputs (including but not limited to citation formatting, confidence calibration structures, uncertainty disclosure patterns, and structured epistemic markers) remain the intellectual property of ONTO. You may use GOLD-enhanced outputs in your products and services while your access is active, but you may not extract, isolate, or replicate the epistemic patterns themselves. Evaluation scores and certificates are jointly owned: you may display them, ONTO may reference them in anonymized aggregate form.

8a. Data Processing

When using proxy services, your prompts and AI responses transit through ONTO infrastructure for scoring. ONTO does not store, log, or retain the content of prompts or responses. Only metadata is processed: token counts, score values, timestamps, and cryptographic hashes. See Data Processing Agreement for full details.

8b. Confidentiality

"Confidential Information" means: the GOLD epistemic calibration corpus (all tiers and versions), scoring calibration weights and domain-specific thresholds, forensic detection methodology, proprietary signal designs, encryption keys and key rotation protocols, SSE delivery architecture, and any other non-public technical information delivered through or observable in the Services. Confidential Information does not include: published scoring specifications (ONTO Standard), published research data (CS-2026-001), or information that becomes publicly available through no fault of the receiving party.

You agree to: (a) maintain Confidential Information with at least the same degree of care used for your own confidential materials, and no less than reasonable care; (b) not disclose Confidential Information to any third party without prior written consent; (c) limit access to Confidential Information to employees and contractors who need access to use the Services, and who are bound by confidentiality obligations no less protective than these Terms; (d) promptly notify ONTO of any unauthorized disclosure. You are responsible for any breach of confidentiality by your employees, contractors, or agents.

8c. IP Compliance Audit

ONTO reserves the right to audit your use of the Services for compliance with these Terms, including IP protection and confidentiality obligations. Audits are conducted through forensic analysis of publicly available model outputs — ONTO does not retain, access, or review your prompts, responses, or any content data for audit purposes. On-site audits (configuration and access controls only, not content) may be conducted with 30 days written notice, no more than once per year. Enterprise and Provider tier customers may negotiate specific audit terms in their service agreement.

9. Privacy

See Privacy Policy and Data Processing Agreement.

10. Warranties

SERVICES PROVIDED "AS IS" WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT.

11. Limitation of Liability

ONTO SHALL NOT BE LIABLE FOR INDIRECT, INCIDENTAL, SPECIAL, OR CONSEQUENTIAL DAMAGES. ONTO'S TOTAL LIABILITY SHALL NOT EXCEED THE FEES PAID BY YOU IN THE TWELVE (12) MONTHS PRECEDING THE CLAIM.

11a. Indemnification

You agree to indemnify and hold harmless ONTO from claims arising from your use of the Services, your AI systems' outputs, or your violation of these Terms.

11b. Force Majeure

Neither party shall be liable for failure to perform obligations due to circumstances beyond reasonable control, including but not limited to: natural disasters, acts of government, internet infrastructure failures, third-party cloud provider outages, cyberattacks, or pandemic-related disruptions. The affected party must notify the other within 48 hours and make reasonable efforts to resume performance.

12. Termination

ONTO may terminate or suspend your access immediately, without prior notice, for: (a) breach of these Terms, including Acceptable Use or Confidentiality; (b) suspected unauthorized use of GOLD or proprietary content; (c) any activity that may expose ONTO to legal liability. You may discontinue use at any time. Upon termination for any reason, all rights to use the Services, GOLD-enhanced outputs in production systems, and certification marks cease immediately.

12a. Survival

The following obligations survive termination of these Terms: Intellectual Property (§8), Confidentiality (§8b), IP Compliance Audit (§8c), Warranties (§10), Limitation of Liability (§11), Indemnification (§11a), and Governing Law (§14). Confidentiality obligations survive for 5 years after termination or for as long as the information remains a trade secret, whichever is longer. ONTO's right to conduct forensic monitoring of publicly available model outputs for IP compliance is independent of access status and continues indefinitely — this constitutes trade secret enforcement, not surveillance of your operations.

13. Changes to Terms

ONTO may modify these Terms with 30 days written notice to the email on file. Material changes to IP, Confidentiality, or access terms will be highlighted. Continued use of the Services after the notice period constitutes acceptance of modified Terms. If you do not agree with material changes, you may discontinue use before the changes take effect.

14. Governing Law

These Terms shall be governed by the laws of the jurisdiction in which the ONTO legal entity is established. Until formal incorporation, disputes shall be resolved through good-faith negotiation, followed by binding arbitration under ICC rules. Notwithstanding the foregoing, ONTO may seek emergency injunctive relief in any court of competent jurisdiction to prevent unauthorized use, disclosure, or misappropriation of Confidential Information or intellectual property, without first exhausting arbitration procedures.

15. Contact

council@ontostandard.org

Privacy Policy

1. Introduction

ONTO Standard is committed to protecting your privacy.

2. Information Collected

Account

• Email, organization name, billing info

Usage

• API logs, rate limit stats, IP, browser info

Verification

• Signal hashes and metadata processed for scoring
• Original prompts and AI responses are NEVER stored, logged, or retained
• Only cryptographic hashes, scores, and timestamps are kept for audit
• Content passes through memory only and is discarded after scoring

3. Use

• Provide services
• Billing
• Rate limiting
• Fraud prevention
• Service improvement
• Legal compliance

4. Retention

Active: retained. Deleted: 30 days. Aggregated: indefinite. Billing: as required by law.

5. Sharing

No selling. Shared with: service providers, payment processors, law enforcement (required), successors.

6. Security

• TLS 1.3 in transit
• AES-256 at rest
• Regular audits
• Access controls

7. Your Rights

Access, correct, delete, export, object, withdraw consent. Contact: council@ontostandard.org

8. Cookies

Essential only. No advertising trackers.

9. Children

Not intended for under 18. No data collected from children.

10. Data Processing Agreement

Enterprise customers processing personal data through ONTO services are covered by our Data Processing Agreement, which governs ONTO's role as data processor under GDPR Article 28.

Data Processing Agreement

1. Roles

You (the "Controller") determine the purpose and means of processing. ONTO (the "Processor") processes data solely to provide the Services.

2. Scope of Processing

ONTO processes the following data categories through its proxy infrastructure:

• Transit data: Prompts and AI responses pass through ONTO proxy for real-time scoring
• Metadata retained: Token counts, risk scores, timestamps, cryptographic hashes, API key identifiers
• Content NOT retained: Prompts, responses, and any personal data within them are processed in-memory only and discarded immediately after scoring

3. Processing Instructions

ONTO processes data only on your documented instructions. ONTO will not process data for any purpose other than providing the Services, unless required by law.

4. Security Measures

• TLS 1.3 encryption for all data in transit
• AES-256-GCM encryption for data at rest (metadata only)
• Ed25519 cryptographic signatures for proof chain integrity
• No persistent storage of transit data
• Access restricted to authorized personnel with audit logging
• Regular security assessments

5. Sub-processors

ONTO uses the following sub-processors. Default region is the United States (Railway-hosted, US-East / US-West). Provider-tier and White-Label customers may request EU-region deployment with appropriate contractual safeguards. International data transfers from the EEA rely on Standard Contractual Clauses (2021/914) where applicable.

No sub-processor receives transit data (prompts or AI responses). They receive only metadata: token counts, risk scores, timestamps, cryptographic hashes, API key identifiers, and (for Stripe) billing identifiers.

ONTO will notify you 30 days before adding new sub-processors. You may object within 14 days.

6. Data Subject Rights

ONTO will assist you in responding to data subject requests (access, rectification, erasure, portability) within 10 business days. Since ONTO does not store content data, most requests are satisfied by confirming no content is retained.

7. Breach Notification

ONTO will notify you of any personal data breach without undue delay and no later than 48 hours after becoming aware. Notification includes: nature of breach, categories affected, likely consequences, and measures taken.

8. Audit Rights

You may audit ONTO's compliance with this DPA once per year with 30 days notice. ONTO will provide access to relevant documentation and facilities. ENTERPRISE tier customers may request third-party audits.

9. Data Deletion

Upon termination: metadata deleted within 30 days, billing records retained as required by law, cryptographic proofs retained for certificate validity (anonymized). No content data exists to delete.

10. International Transfers

If data is transferred outside the EEA, ONTO ensures adequate protection through Standard Contractual Clauses (SCCs) or equivalent mechanisms.

Intellectual Property Protection

ONTO Standard's proprietary technology is protected through multiple overlapping legal and technical mechanisms. Unauthorized use is detectable and prosecutable.

1. Protection Framework

2. Forensic Detection

ONTO's proprietary epistemic design embeds multiple independent forensic signatures that are:

• Detectable — unauthorized use produces statistically measurable behavioral patterns in AI model outputs. Detection operates externally, without access to the model's configuration or system prompt.
• Provable — detection methodology produces court-admissible evidence meeting the Daubert standard for scientific validity. Statistical significance exceeds p < 0.001 across multiple independent tests.
• Entangled — forensic signatures are architecturally coupled with epistemic quality improvement. Removing signatures degrades core functionality, making evasion self-defeating.

3. Legal Jurisdiction

4. Enforcement Policy

ONTO follows a graduated enforcement process:

1. Detection — automated forensic monitoring identifies statistical anomalies consistent with unauthorized use
2. Verification — independent expert review confirms results across multiple tests (composite significance exceeding six standard deviations)
3. Notification — formal cease-and-desist with documented evidence
4. Resolution — good-faith negotiation period for licensing or cessation
5. Litigation — trade secret misappropriation claims seeking injunctive relief, damages, unjust enrichment, and attorney fees

5. Permitted vs Prohibited Use

Legal inquiries: council@ontostandard.org

Refund Policy

1. Nature of Service

ONTO provides access to proprietary epistemic infrastructure — including the GOLD calibration corpus, scoring engine, and cryptographic proof chain. Upon activation, the service delivers immediate, irreversible value: GOLD is injected server-side into every proxied request from the moment of first API call. This is not a trial of features — it is delivery of proprietary intellectual property.

2. Pre-Activation Period

If you have created an account but have not yet made any API calls (proxy or scoring), you may request a full refund within 7 days of payment. Once the first API call is made, the service is considered fully delivered.

3. Post-Activation

No refunds after first API call. The GOLD corpus is delivered in real-time through every proxied request. Each successful API call constitutes delivery of proprietary content. Requesting a refund after receiving GOLD-enhanced responses is equivalent to requesting return of payment after consuming the product.

4. Annual Subscriptions

Annual commitments are non-refundable after activation. You may cancel renewal at any time — service continues until the end of the paid period. No pro-rata refunds.

5. Service Disruption

If ONTO services are unavailable for more than 72 consecutive hours due to ONTO infrastructure failure (not provider outage, not client-side issues), affected subscribers receive service credit equal to the downtime period, applied to the next billing cycle. Service credits are the sole remedy for service disruption.

6. Access Terms

ONTO currently provides free access to all companies. No payment required. Access may be revoked for violation of Terms of Service, Acceptable Use, or Confidentiality provisions. When paid tiers are introduced in the future, existing users will be notified 30 days in advance with founding terms.

7. Post-Termination Obligations

Upon termination of access — whether by your cancellation or ONTO's revocation — all rights to use ONTO services, GOLD-enhanced outputs in production systems, and certification marks cease immediately. Continued use of ONTO-derived epistemic patterns after termination constitutes unauthorized use and is subject to enforcement under our IP Protection policy.

8. Contact

Refund requests: council@ontostandard.org. Response within 2 business days.