Pricing screenshots age badly unless the source trail keeps the unit and checked date

A source-trail question for AI pricing comparisons across token, request, minute, seat, and tool-call units.

AI pricing comparisons age quickly because the unit changes across products. One page might price per million tokens, another per minute, another per tool call, and another per seat. A screenshot without checked date, model name, region, workload, and unit is hard to reuse later. For source trails, I think the useful fields are provider, product, model, input unit, cached-input unit, output unit, non-token charges, checked date, source link, and expected workload. What else should be captured before comparing AI tool cost?