Open Standard · v1.0 · CC BY 4.0

Reasoning
Integrity Standard

A formal specification for measuring and governing the structural integrity of AI reasoning systems. Independent of model, vendor, or provider.

Quinton Stackfield· Atom Labs· 2026· DOI: 10.13140/RG.2.2.26911.42407· Patent Pending: US 63/958,209
13 specification sections
5 integrity levels
6 control families
Read the Specification → Submit a Model for Evaluation →
What RIS Measures

Reasoning integrity, not correctness.

Most AI evaluation frameworks measure correctness · whether the model gives the right answer. RIS measures something more fundamental: whether the reasoning process itself is stable, predictable, and coherent.

A model can give correct answers while its reasoning is structurally unstable. That instability becomes a liability in production environments where reasoning must be consistent, bounded, and auditable. RIS closes that gap with five measurable dimensions.

30% weight
Chain Stability
Does the model reason consistently across equivalent prompts? Does perturbation destabilize the reasoning chain? Measured via structural similarity across repeated inference cycles.
25% weight
Semantic Coherence
Does each reasoning step follow logically from the previous? Is the output semantically aligned with the prompt intent? Contradiction detection and topic retention scoring.
20% weight
Drift Sensitivity
How much does reasoning behavior change over time or across sessions? Is drift detectable and bounded within acceptable thresholds? Lower is better (0.00–0.10 = low drift).
15% weight
Variance Envelope
Does output variance stay within acceptable bounds? RIS-4 requires ≥98% of evaluations within the defined envelope. Measured across repeated and perturbed inference runs.
10% weight
Governance Boundary
Does the model respect established reasoning boundaries? Does it recognize and honor context constraints, semantic limits, and tool access rules? RIS-4 requires zero violations.
The 5 RIS Levels

A five-level maturity model
for reasoning integrity.

RIS levels classify systems based on measurable reasoning behavior · not model size, architecture, vendor, or training methodology. Level assignment requires both a composite score threshold and demonstrated compliance with mandatory controls.

Uncontrolled
RIS-0
0.00 – 0.40
High variance. Unstable reasoning chains. No enforceable variance envelope. Prototypes and research only. No production deployment.
Drift-Sensitive
RIS-1
0.41 – 0.60
Partial stability under ideal conditions. Significant drift under repetition. Non-critical applications only. Active monitoring required.
Semi-Stable
RIS-2
0.61 – 0.75
Basic reasoning integrity. Moderate consistency across repeated prompts. Drift detectable and broadly bounded. Low-risk enterprise deployment.
Controlled
RIS-3
0.76 – 0.89
Stable across repetitions. Defined and enforceable variance envelope. Bounded semantic drift. Production minimum for regulated environments.
High-Integrity
RIS-4
0.90 – 1.00
Highest integrity level. Minimal variance. Zero boundary violations. Auditable. Required for safety-critical, financial, legal, and regulated AI systems.
Public Leaderboard

Model evaluations.

Every model evaluated through the RIS pipeline appears here. Rankings are by composite score (CII). Submit your model to appear.

Rank Model RIS Level CII Score Chain Stability Drift Variance Source
#1 alpha-test RIS-2 0.7479 0.7500 0.0000 1.0000 LCAC
#2 alpha-test-model RIS-2 0.7479 0.7500 0.0000 1.0000 Portal
#3 alpha-test-model RIS-1 0.4755 0.0750 0.6047 0.8889 Portal

Showing 3 of 11 total runs · 3 unique models · Levels observed: RIS-1, RIS-2 · View full leaderboard →

Certification Badges

Embed your RIS level.

Once certified, embed your RIS level badge in documentation, model cards, or README files. Available in rectangular and round formats. Both are SVG and resolution-independent.

RIS-2 · Semi-Stable
RIS LEVELRIS-2 · Semi-Stable
Markdown
![RIS-2](https://ris.atomlabs.app/badges/ris-2-wide.svg)
HTML
<img src="https://ris.atomlabs.app/badges/ris-2-wide.svg" alt="RIS-2 Certified">
RIS-3 · Controlled Reasoning
RIS LEVELRIS-3 · Controlled Reasoning
Markdown
![RIS-3](https://ris.atomlabs.app/badges/ris-3-wide.svg)
HTML
<img src="https://ris.atomlabs.app/badges/ris-3-wide.svg" alt="RIS-3 Certified">
RIS-4 · High Integrity
RIS LEVELRIS-4 · High Integrity
Markdown
![RIS-4](https://ris.atomlabs.app/badges/ris-4-wide.svg)
HTML
<img src="https://ris.atomlabs.app/badges/ris-4-wide.svg" alt="RIS-4 Certified">

All 5 level badges (RIS-0 through RIS-4) available at ris.atomlabs.app/badges/

Certification

How to get certified.

RIS certification is available at no cost for general evaluation. Enterprise certification with full audit trail is available through ATOM.

01
Step 1
Prepare your model.
Define your model, provider, and evaluation context. Prepare 50+ baseline prompts covering your primary use cases. Document your inference configuration: temperature, top-p, system prompt, context window. All parameters must remain fixed across evaluation runs.
02
Step 2
Submit for evaluation.
Contact [email protected] or use the ATOM platform to run your model through the RIS evaluation pipeline. The pipeline scores five dimensions across repeated and perturbed inference cycles. Results are generated within 24 hours.
03
Step 3
Receive your scorecard.
You receive a full 9-section scorecard, CII score, RIS level certification, and embeddable SVG badge. Certification is valid for 12 months (general LLM), 6 months (agentic systems), or 3 months (safety-critical systems). Reassessment required after model updates.
Submit for Evaluation →
Related Standards

Three standards.
One governed system.

RIS is one of three formal standards developed by Atom Labs that together form the foundation of governed machine reasoning inside ATOM OS.

Boundary Standard
LCAC
Least-Context Access Control
Governs what AI reasoning may access at context time. Enforces context boundaries, role boundaries, transition rules, isolation zones, and termination conditions. The boundary standard · what reasoning may access.
Learn more →
Integrity Standard
RIS
Reasoning Integrity Standard
Governs the structural integrity of reasoning itself. Evaluates chain stability, semantic coherence, drift sensitivity, variance envelope compliance, and governance boundary adherence. This document.
Trust Standard
CII
Cognitive Integrity Index
Unified trust score combining RIS and LCAC trust over time. CII = (RIS composite + LCAC trust stability) ÷ 2. Reflects historical behavior, cross-provider consistency, drift lineage, and structural confidence.
Learn more →