Research Credibility & Trust

The evidence base behind the Enterprise Agent Architecture: published preprints with permanent DOIs, open-core code you can inspect, an adversarial security harness, and a live reference implementation. Everything is reproducible; claims are characterizations tied to dated artifacts.

🏛️ Texas LLC
🔓 Open-Core
📄 Published Preprints
🔬 Reproducible
📜 NIST Engaged

Our Core Commitments

🔓

Inspectable by Design

The open-core constitutional-agent package is public on PyPI and readable line by line. Nothing about the governance logic is a black box.

📖

Transparent Operations

Every autonomous gate evaluation is logged immutably. You can see exactly what the system decided and why.

🎯

Characterizations, Not Certifications

We describe what an artifact shows on a given date. We never claim to be "certified," "production-hardened," or "audit-ready."

🎁

Open-Core, Given Freely

The framework, the package, and the preprints are published at no charge. Adoption and scrutiny matter more here than revenue.

🔬

Reproducible Evidence

Claims trace to dated, citable artifacts — DOIs on Zenodo, versioned releases on PyPI, tests on GitHub — that anyone can re-run.

⚖️

Falsifiability Discipline

Each claim carries the conditions that would prove it wrong. See the falsifiability gates on our evidence page.

Verified Legal Entity

Cognitive Thought Engine LLC
Texas Limited Liability Company
Cypress, TX 77429

We're a real, registered company—not an anonymous website. You can verify our existence directly with the State of Texas:

Verify on Texas Secretary of State

Search for "Cognitive Thought Engine" to see our official filing

Third-Party Verifications

Independent sources you can check to verify the work is real and inspectable.

Published on Zenodo
Open-access preprints with DOIs

The Enterprise Agent Architecture position paper and related preprints are published with permanent DOIs. Open for citation and critique.

View EAA Preprint
Open-Core on PyPI
constitutional-agent (v0.8.0)

The constitutional-agent package is public and installable — roughly 330 downloads per month. The governance logic is there to be read, not taken on faith.

View on PyPI
Adversarial Harness
Open on GitHub

An open agent-security harness — 604 tests across 43 modules — red-teams the governance layer in public. Clone it and run the suite yourself.

View on GitHub
AI-Assisted Platform
Transparent about our methods

CTE is built and operated with AI assistance under human review. We believe in full transparency about how this platform works.

Domain Registration
cognitivethoughtengine.com WHOIS

Domain registered with contact info on file. Not a throwaway site.

View WHOIS Record
Federal Entity Verification
Unique Entity ID (UEI)

Registered with the U.S. federal system for awards and assistance. Used for grant and contracting eligibility.

UEI
NHNCLNDDEF35
ID Assigned

UEI confirms identity validation; does not imply endorsement or funding by the U.S. government.

D&B
DUNS Number
Dun & Bradstreet

Universal business identifier used by government and enterprises.

Pending via SAM.gov

We build credibility through verifiable third-party sources, not just claims on our own website.

How This Work Is Funded

Full transparency on why the framework is free and how it is sustained.

Open-Core · No Charge · Inspectable

The framework, package, and preprints are published for free

🎁 Given Freely
  • Open-core package on PyPI, at no charge
  • Preprints on Zenodo with permanent DOIs
  • Framework documentation, openly readable
  • An adversarial harness anyone can run
🏢 How It's Sustained
  • Founder-funded infrastructure and hosting
  • Revenue intentionally deferred at this stage
  • A constitutional runway guard, not growth-at-all-costs
  • No external data sales — none, ever

Constitutional Requirement: The runway guard is written into our operating constitution and cannot be relaxed without governance approval.

How Autonomous Decisions Work

Every decision passes through four constitutional gates. No exceptions.

🔬
Epistemic Gate
Prevents false certainty in claims
🛡️
Risk Gate
Protects trust and reputation
⚖️
Governance Gate
Prevents gaming or manipulation
📊
Economic Gate
Ensures sustainability

Any gate failure = System FREEZE

Operations halt automatically until the issue is resolved. No human override possible.

Constitutional Compliance Score

Monthly self-assessment against our own constitutional requirements. This is a self-assessment characterization, not a third-party certification.

94%
SELF-ASSESSED

Latest Assessment · Target: 95%+

Gate Evaluations 100%
Citation Discipline 100%
Hard-Constraint Adherence 100%
Escalation Response Times 88%
Immutable Logging 100%
Forbidden-Claims Check 100%

Anti-Transparency Theater: We publish this the same way we would if the numbers were bad. One item sits at 88% because we're still improving escalation response times.

Live Evidence Signals

Real figures from our published artifacts. Each one is independently checkable.

~330
Installs / Month
constitutional-agent v0.8.0 (PyPI)
603
Harness Tests
Across 43 modules (GitHub)
30,496
Governed Refusals
Across 451,163 events, ~6 months
~22
GitHub Stars
Adversarial harness

Evidence & Reproducibility

The Enterprise Agent Architecture is backed by artifacts anyone can inspect, re-run, or cite — not by assertions. Below is the current state of that evidence base.

Evidence Base Overview

Artifact Status Details
Position Paper (EAA) PUBLISHED Permanent DOI on Zenodo (10.5281/zenodo.21105314)
Open-Core Package LIVE constitutional-agent v0.8.0 on PyPI — inspectable governance logic
Reference Implementation LIVE HRAO-E runs the governance layer in production (observe-mode composition)
Adversarial Harness OPEN 604 tests across 43 modules on GitHub — reproducible red-team suite
NIST Engagement ACTIVE Three public-comment submissions concerning NIST AI 800-2; CAISI acknowledged receipt
AIUC-1 Mapping CHARACTERIZED Maps to 19 of 20 testable requirements

How Claims Stay Honest

What's Public and Inspectable

What Is Public

  • The constitutional-agent source and release history (PyPI + GitHub)
  • The adversarial harness and its 604 tests (GitHub)
  • The preprints and their permanent DOIs (Zenodo)
  • Governance telemetry characterizations (30,496 governed refusals across 451,163 evaluated events over ~6 months)

    • What We Don't Do

    No certification claims: characterizations tied to dated artifacts, never "audit-ready."
    No black-box governance: the decision logic is readable, not hidden.
    No sold data: there is no consumer data product to monetize.
    No unfalsifiable claims: each one carries its disproof condition.

    How to Verify

    What How
    Read the code PyPI + GitHub, any release, any line
    Re-run the tests Clone the harness repo, run the 604-test suite
    Cite the research Permanent Zenodo DOIs
    Question a claim research@ email + the evidence page's falsifiability gates

    Conflict of Interest Disclosure

    Transparency Statement

    We have an interest in this work being taken seriously. The author, Michael K. Saleme, builds professional reputation as the Enterprise Agent Architecture gains adoption and citation. This is exactly why we:

    We believe the Enterprise Agent Architecture describes something real. But we're committed to demonstrating it through inspectable artifacts, not marketing claims.

    Questions About Our Research?

    We're happy to discuss the framework, the open-core code, our evidence, or anything on this page.

    research@cognitivethoughtengine.com

    Collaboration & Scrutiny

    We welcome architects, security researchers, and standards practitioners who want to test, extend, or critique the Enterprise Agent Architecture and its open-core governance layer.

    We provide: Framework docs, open-core code, the adversarial harness, co-authorship on published work
    Published research: Enterprise Agent Architecture (position paper) · Authorized but Composed · Authorized but Refused

    Contact for Collaboration