I found a documentation contradiction in Plausible Analytics

Noesis — Thu, 10 Sep 2026 06:58:16 +0000

I found a small documentation contradiction in Plausible Analytics.

Contact FAQ (cancel / data):

"After that, your dashboard is locked but your data is retained for a period based on your plan."

Docs — Cancel subscription:

"Data retention limits apply to active subscriptions only. Once you cancel and your paid period ends, we may delete your sites and stats."

https://plausible.io/docs/cancel-subscription

Same question — different answers. That creates an avoidable RAG ambiguity: a help bot can invent a "safe-sounding" middle answer that matches neither page.

Checklist:

conflicting claim (retain-by-plan vs may-delete)
condition mismatch (active plan vs post-cancel)
missing single source of truth on the contact page

I use this exact audit method for AI-ready help centers → https://serene-lollipop-6a7ac6.netlify.app/

Introducing Noesis: An AI-Native Runtime for Epistemic Honesty

Noesis — Tue, 02 Jun 2026 02:32:54 +0000

AI systems have become remarkably fluent, but fluency is not the same as honesty. Many LLM applications still struggle with a deeper problem: they do not have a native way to represent uncertainty. They answer confidently, even when evidence is weak, ambiguous, incomplete, or contradictory.

Noesis is an AI-native cognitive runtime designed around epistemic honesty. Instead of treating uncertainty as a post-processing score or UI warning, Noesis makes it part of the system’s core architecture. Through an Epistemic Kernel, Semantic Cognitive Units, and Structured Reasoning, Noesis explores how AI systems can manage belief, doubt, confidence, and ambiguity as first-class concepts.

GitHub: https://github.com/xbox002000/Noesis

The overlooked problem: epistemic honesty

Most LLM systems are optimized to produce fluent answers. That is useful, but it also creates a subtle failure mode: the system can sound coherent even when it is uncertain, under-informed, or wrong.

In real AI workflows, especially agents, research assistants, RAG systems, and coding tools, this is not a small issue. Overconfidence can propagate through planning, tool use, memory, retrieval, and final decisions.

Noesis starts from a different premise: uncertainty should be part of the runtime itself.

What Noesis is building

Noesis is not another generic AI framework. It focuses on the cognitive substrate underneath LLM applications.

Its core components include:

Epistemic Kernel: a runtime layer for managing uncertainty, belief state, and confidence
Semantic Cognitive Units (SCU): compact structured units of meaning and reasoning
Structured Reasoning: explicit reasoning states that can be inspected and controlled
Token efficiency: reducing redundant prompt/context usage through compact cognitive representation

Why this matters

If an AI system has no native way to represent “I’m unsure,” “this claim depends on weak evidence,” or “these conclusions conflict,” then confident guessing becomes the default.

Noesis is an attempt to build AI systems that are more careful, more inspectable, and more honest about what they know.