I came to AI research through information design and publishing rather than machine-learning engineering. For more than three decades, my work has focused on how information is structured, presented, and made usable by people. That background led me to ask a different question about AI:

If the reader has changed, why are we assuming the source should remain exactly the same?

My work now focuses on what happens when AI becomes an intended reader and reasoner over material written for humans. Human readers routinely reconstruct relationships, boundaries, qualifications, scope, causation, hierarchy, and uncertainty from context. AI must reconstruct many of those same distinctions from prose, and it does not always do so reliably.

That led me to the idea at the center of the work: a bounded analytical record that sits beside the original source.

The source remains the evidence. The record preserves selected reasoning-relevant structure around it: what is supported, how elements are related, where a claim stops, what remains uncertain or unresolved, and where the evidence came from. Records can be created at different boundaries and levels of resolution and accumulated over time as a reusable analytical layer for later AI reasoning.

Producing those records responsibly turned into an architecture problem. Different kinds of source structure require different forms of examination, and source preparation, framing, routing, analysis, record production, and review need to remain distinct. That became the AI Source Analysis Framework, a working proof of concept.

The Framework is the process. The record is the intended outcome.

I work independently from Elko, Nevada, and am currently focused on developing, testing, and publishing this source-side research. The next stage is formal evaluation, machine-readable representation, selective automation, specialized analytical development, and independent technical testing to determine how far the approach can go.