3. The Second Pain: Partial Review and Trust Fragmentation

Another subtle problem appeared.

A file could be partly reviewed and partly unreviewed.

One section might be correct. Another section might still be raw AI output. A third section might contain the user’s original idea. A fourth section might be AI’s interpretation of that idea.

But the file itself looked uniform.

Once everything was written in the same Markdown style, it became difficult to see the trust level of each part.

This created a dangerous blur:

human input
AI-generated draft
AI-interpreted summary
human-reviewed section
approved decision
outdated assumption

All of these could appear in the same visual form.

That led to another key insight:

AI-first knowledge needs visible provenance and review state.

A knowledge base must distinguish:

• fact,
• interpretation,
• proposal,
• decision,
• assumption,
• draft,
• reviewed content,
• approved content,
• deprecated content.

Without this distinction, the human reader must constantly re-evaluate everything.

That creates attention debt.

And when attention debt grows, people stop reviewing.

---