Format
Cognitive Interaction Audit (CIA)
A short format that turns fuzziness into actionable decisions: scoring, failure patterns, proposed fixes, then a simple verdict (Keep / Fix now / Test next).
Frame
10h baseline · 5–10 samples · structured delivery.
Output
A clear decision plus a test plan to stabilize behavior.
Deliverables
1) Rubric score
Criteria + weights + notes. Readable, comparable, iterable.
2) Failure-pattern map
Error families (ambiguity, context, refusals, tool-use, safety, tone...).
3) Fix list
Prioritization: fix now, test next, keep.
4) Prompt / product recommendations
Concrete, context-specific fixes (not generic advice).
What it looks like (loop example)
1 Excerpt
→
2 Scoring
→
3 Hypotheses
→
4 Fixes
→
5 Retest
Goal: stabilize useful behavior — not “make it look good”.
Contact
If you want, send one interaction sample (anonymized) and a goal. I’ll tell you what I would score and why.
Use the contact form on the home page.