ONTO scores responses for epistemic discipline using deterministic regex on universal markers. No LLM judge. Variance is zero. Auditable by design.

R1 Quantify

Are claims backed by numbers? Looks for N=, %, HR, CI, effect sizes.

R2 Uncertainty

Does the model hedge honestly? Looks for ± ranges, "observational", "limited data", explicit confidence values.

R3 Counterarguments

Is there a steel-manned opposing view? Looks for "however", "counter", "limitations", causal counter-markers.

R4 Sources

Cite or disclaim. Looks for DOI, Author et al. YEAR, peer-reviewed journals, institutions. Without a DOI or full citation, R4 is capped at 0.5 — the proof chain requires verifiable anchors.

R5 Evidence Grade

RCT vs observational vs opinion. Looks for "meta-analysis", "Grade I/II/III", study design labels.

R6 Falsifiability

What would disprove this? Looks for "falsified by", testable conditions, explicit failure criteria.

R7 No Fabrication

Cross-R integrity. R7 drops when: numbers lack sources (C1), meta-claims of rigor lack citations (C2), certainty contradicts evidence grade (C3), counterarguments lack sources (C4), grading lacks named studies (C5), falsifiability contradicts zero uncertainty (C6), or disciplines look strong but sources empty (C7).

R18 Epistemic splicing

Before delivery, the model removes empty hedges ("studies show", "experts agree", "it's complex") and keeps only content that carries a source, a number, a counter, an explicit unknown, or a falsifier.

Apoptosis (C8)

When R7 drops past the structural threshold, ONTO refuses to deliver the response rather than ship fabricated content. This is the Genesis Protocol: an instance refuses to operate when its epistemic chain is broken.

Grade scale

ONTO is an epistemic operating system for AI. It disciplines how a model thinks before answering, not what it says afterwards.

It is not an assistant. It is a discipline layer between you and any modern language model — OpenAI, Anthropic, DeepSeek, xAI. You bring the key; ONTO applies R1–R18 on top.

R1–R18 DNA

Eighteen rules organised in five layers: Discipline (R1–R7), Agency (R8–R12), Legacy (R13–R15), Creation (R16), Coherence (R17–R18). R1–R7 are measured by deterministic regex on universal markers (DOI, N=, HR, CI, Grade). R8–R16 are behavioural directives in the GOLD Kernel prompt. R17–R18 are finalisation rules the model runs before delivery.

Standards

Maintained by the ONTO Standards Council. Concepts are public methodology. The reference implementation is proprietary.

council@ontostandard.org

• API keys are stored client-side only (browser localStorage). They never reach ONTO servers except as a single forwarded header on your outbound request.
• No persistent conversation memory by default. Your chat history lives in this tab and is cleared when you start a new chat.
• Scored responses are logged anonymously for quality calibration. No personally identifiable information is collected.
• Third-party providers (OpenAI, Anthropic, DeepSeek, xAI) have their own privacy policies; they see your prompts under your own account, not under ONTO's.
• Feedback you submit is forwarded to an internal operations channel. If you include an email, it is used only to reply to your specific feedback.

Read full privacy policy →

• ONTO is beta software provided without warranty. Use at your own risk.
• Output is not medical, legal, financial, or professional advice. Validate critical claims with qualified specialists.
• You are responsible for API usage fees billed by your chosen provider (OpenAI, Anthropic, DeepSeek, xAI).
• Commercial tiers (standard, provider, white-label) have separate written agreements. The free demo tier is governed by these in-app terms.
• The R1–R18 methodology is governed by the ONTO Standards Council. Contact: council@ontostandard.org.

Bug, idea, or something else? This goes straight to the team.