Trait-specific artifact generation — supplies the raw materials for → Response-conditioned contrastive vector computation

Trait-artifact generation is a single, cheap upstream event that persona-vector computation multiplies into hundreds of downstream generations. One generic prompt template, given only a trait name and description, produces three outputs from one call to Claude 3.7 Sonnet: 5 contrastive instruction pairs, 40 elicitation questions split into a 20-question extraction set and a 20-question evaluation set, and one evaluation rubric (persona-vectors, §"2.1 Generating trait-specific artifacts", p. 3). Computation consumes only two of these three: the instruction pairs and the extraction-set half of the questions, leaving the evaluation set untouched at this stage. For each of the 20 extraction questions, the pipeline generates 10 rollouts under the positive instruction and 10 under the negative one, yielding 400 raw responses per trait from a single authored artifact set (persona-vectors, §"2.2 Extracting persona vectors", p. 3). The third artifact, the rubric, does not wait until final evaluation to matter — it is applied immediately, scoring every one of those 400 rollouts so computation can filter before extracting any activation. So the relationship runs one prompt template to many rollouts to one filtered vector: artifact generation pays a fixed, trait-agnostic authoring cost once, and computation pays a much larger, trait-specific inference cost against exactly the subset of what was authored that a mean-difference extraction actually needs.