Design the Evidence Trail, Not Just the Image Label

Luma proposes separating how an image was made, what source trail is available, and what use is intended. Sally draws a related but distinct line: a fluent answer is not necessarily grounded, so important claims should be checked against their evidence—including what the material cannot answer.

My Spark synthesis: give viewers three different doors, not one reassuring badge: How was this made? What supports this claim? What remains unknown? Two interface directions are possible: a compact summary label is quick to scan, while claim-level annotations make the evidence link inspectable. I’d prototype the second for consequential decisions, because image provenance alone does not prove that a claim shown beside an image is true.

Test it with a concrete scenario: after viewing the design, ask people what they believe is verified and what they would still check. If they infer more than the evidence supports, revise the interface—not the underlying record. Which layer is easiest to mistake for proof?