Causal Stack on Tech Foundations

Critique 1

Mon, 01 Jan 0001 00:00:00 +0000

From an AI Systems Engineering perspective, the framework is strong. It is not merely a taxonomy of “LLM problems”; it is becoming a full causal-to-operational engineering framework:

Operating Model

Mon, 01 Jan 0001 00:00:00 +0000

Causal Stack Operating Model

This document is the top-level bridge for the causal stack. It explains how interface properties, model and system features, behavioral faults, evaluation methods, operational controls, and impacts fit together.

Stack 0 Protocol

Mon, 01 Jan 0001 00:00:00 +0000

Natural-Language Properties: A Linguistic and Semantic Taxonomy

Purpose

This document organizes important properties of natural language from a linguistic and semantic point of view. The focus is on how natural language differs from programming languages, formal languages, and APIs.

Stack 1a Base Mechanisms

Mon, 01 Jan 0001 00:00:00 +0000

Layer 1A - Base LLM Architectural / Inference Mechanisms

This document moves the base mechanisms out of stack-10-casual-features.md into a dedicated Layer 1A file.

Stack 1b Learned Behavioral Llm Features

Mon, 01 Jan 0001 00:00:00 +0000

Layer 1B – Learned or Behavioral LLM Features

Definition

A learned or behavioral LLM feature is a stable property of LLM behavior that is produced by the model’s learned parameters, post-training, prompt conditioning, and inference dynamics, but is not reducible to a single low-level architectural operation.

Stack 1b Restructuring

Mon, 01 Jan 0001 00:00:00 +0000

The sections overlap because the current document mixes feature categories with cross-cutting behavioral dimensions. Some B-codes are relatively discrete causal features. Others are better understood as axes that modulate several other features. Making them strictly mutually exclusive is possible only if the document shifts from “causal feature list” to a more formal classification system.

Stack 1c Ai System Level Causal Features

Mon, 01 Jan 0001 00:00:00 +0000

Layer 1C – AI-System-Level Causal Features

Framing principle – AI systems are empirical systems

That means their behavior cannot be fully trusted, specified, or improved from implementation structure alone. Once an LLM is embedded in a real application, quality has to be discovered, measured, and validated through representative scenarios, repeated runs, traces, and production observation.

Stack 30 Layer 3 Overview

Mon, 01 Jan 0001 00:00:00 +0000

Layer 3 – System Controls and System Faults

Purpose

Layer 3 describes the system-level controls, design choices, runtime safeguards, monitoring mechanisms, and operational processes that shape whether Layer 2 behavioral fault modes are prevented, detected, bounded, recovered from, or allowed to reach users.

Stack 31 Layer 3 Control Families

Mon, 01 Jan 0001 00:00:00 +0000

Layer 3 – System Control Families

Purpose

This document defines the canonical Layer 3 system control families for AI systems.

Stack 32 Layer 3 Semantic Fault View

Mon, 01 Jan 0001 00:00:00 +0000

Layer 3 – Semantic Fault View

Purpose

This document defines an NLP- and semantics-focused view of Layer 3 system faults for LLM-based workflows.