Building Curio: Tests Before Claims
Did the improvement reach the person using it?
A plausible system is not yet a working product. These build notes follow each claim to the layer a reader can actually see or hear.
A server can return the right item while the phone shows something else. A voiceover can sound better while two variables change together. A convincing model role can remain only a proposal until complete workflows are compared.
These notes follow the path from idea to implementation, then ask whether the visible result—not the sophistication of the pipeline—earned the claim.
Curio Will Compare Two AI Workflow Bundles on Fixed Briefs
Can splitting idea work from implementation improve complete work under fixed conditions?
Boundary: No Curio result exists yet. The comparison will test complete workflows, not prove that either displayed model label is universally better at one job. Read the evidence → Building CurioI Changed Sentences and Pauses. The Voiceover Felt Less Like a List.
I judged one Curio rewrite less list-like.
Boundary: It does not show that fewer sentences always improve narration or that pitch resets cause list-like delivery. Sentence structure and pause timing changed together. Read the evidence → Building CurioMy Pinned-Card Check Stopped Before the Phone Screen
My check proved which card the server sent first, not which card the phone finally showed.
Boundary: It did not prove which card a person finally saw, whether the position changed, or whether position affected reader behavior. A live screen test was still required. Read the evidence →