v1.1.0 — SSRN prep

Verified claims, consistent numbers, sourced citations. Prepared for SSRN submission.

v1.2 — Corrected corpus count and bio

• Bio: twelve years → thirteen years
• Corpus count: verified at 6,000+ conversations / 300,000+ messages
• URLs updated to /research/ path
• PDF regenerated

v2.1 — Corpus Compounding

Section 9 addition: The corpus compounds.

Unstructured corpora are not single-use. The same conversations can be traversed from different conceptual angles, each traversal producing a different compiled document. Those documents feed back into the data source, making the corpus richer without new input from the maker. Structured prompts are consumable. Unstructured corpora are generative and renewable.

v2.0 — Bidirectional Accommodation + Copy/Clarity Pass

Changes

• Bidirectional accommodation frame applied throughout
• Pour reframed as accommodation (the act), not just material (the corpus)
• Stat lines reduced, voice violations fixed
• Classroom paragraph rewritten for specificity
• Implications section varied from mechanical parallelism

v1.1 — FormWork/LensArray Correction

What changed

• Section 4.2: "Evaluation lenses (Formwork)" corrected to "LensArray" with proper link
• Applied Tools section updated: FormWork listed as coordination harness, LensArray as separate evaluation tool
• Cross-references updated to current site URLs (/practice/ instead of /governance/)
• Added missing prosthetic cognition reference
• PDF included

v1.0 — Initial Publication

Input Inversion: Why Unstructured Human Thinking Produces Better AI Output

Initial publication of the whitepaper proposing input inversion as a formal design principle: reverse the burden of structure from the human to the tooling.

Key claims:

• Unstructured human thinking produces better AI output when purpose-built tools handle the translation
• Converging industry evidence supports the principle: declining prompt engineering, agentic workflows outperforming optimized prompts, voice input producing richer material, RAG infrastructure processing unstructured data at scale
• Three years of applied evidence: 1,643 ChatGPT sessions + 700+ Claude sessions of unstructured thinking compiled into a production site assessed as "unequivocally human-written"

Author: Peter Salvato
License: CC BY 4.0