Deterministic Systems Before Probabilistic AI

Technical systems often improve when the uncertain parts are isolated instead of made central.

This placeholder article exists to validate typography, metadata, routing, code blocks, lists, tables, blockquotes, images, captions, and diagrams. Replace it with a finished article when the writing is ready.

Start With The Stable Surface

Before adding probabilistic behavior, define what can be represented, checked, and transformed deterministically.

• Normalize input where rules are clear.
• Preserve source references and provenance.
• Use classifiers when the decision boundary is well understood.
• Reserve language models for tasks that need semantic interpretation.

The architectural question is not whether a model is powerful. It is whether uncertainty belongs at that point in the system.

A Small Decision Table

Problem shape	Better first option	Reason
Exact identifiers	Parser or rules	Deterministic behavior is inspectable.
Repeated categories	Classifier	Training data can make boundaries explicit.
Semantic synthesis	Language model	Meaning may be implicit in the text.

Example Pipeline

type SourceFragment = {
  id: string;
  text: string;
  sourceUrl: string;
};

export function normalize(fragment: SourceFragment) {
  return {
    ...fragment,
    text: fragment.text.trim().replace(/\s+/g, " ")
  };
}

Diagram Placeholder

The article system should support inline SVG diagrams with captions.

Footnote

This placeholder includes a footnote to confirm Markdown rendering support.¹

Footnotes

1. The final article can take a stronger position once examples and references are added. ↩