Skip to content

Evidence papers

Who gets to say it’s good?

For HR teams and internal proposal authors. Software skills do not show who can judge work. See the research, proposals and responsibility.

Free · Full paper

6 pages
13 November 2026
Edition 1.1

Evaluating is not the same as producing. A cited briefing on qualified judgment, AI-assisted review and fair selection, with a proposed governance decision matrix.

Our view of innate human sense is separate from research. Experience, domain knowledge and fair review matter.

No sign-up or payment. Studies, interpretation and proposals are labelled separately. The decision matrix is not a tested hiring tool.

Further papers

Proposed editions · $15 each, one-time purchase. Not on sale yet. Checkout and paid access are not active.

$15

Proposed

The Qualified Reviewer

Proposed scope: recognising domain judgment, designing a fair reviewer audition and separating perceptual expertise from production-tool fluency.

$15

Proposed

A Decision Standard for AI Work

Proposed scope: evidence thresholds, accountable acceptance and the authority to hold, reframe or reject work without losing the decision history.

A useful starting point, not the last word.

The free paper draws on research by Jacobi and colleagues, Sackett and colleagues, Doshi & Hauser, Kranzle & Sharratt, and Maier and colleagues, alongside NIST’s Generative AI Profile. Source URLs and study-specific limitations appear in the PDF.

The burnt steak is an example, not a reported incident. Overcorrection, regression, drift and unchanged repetition are different. They mean going too far, breaking working parts, leaving the brief and repeating unchanged work. AI review with separate criteria can help. It is not independent proof and cannot replace qualified human approval.

Our evidence standardOpen the Decision Gate

Research cutoff: 2026-11-13. Research interpretation, not legal or clinical advice. Hiring benefits and rework savings are unmeasured.