Misty Cripps · Vaquita DesignAuthor Normalization
Standardizing contributor presentation without changing what they submitted.
Every contributor writes their own bio, in their own voice, and uploads their own photo. An organization needs those to read and look consistent. Allowing an LLM to revise them crosses a line: it changes what a person said and alters how they chose to appear. The design problem was figuring out what could be standardized without fundamentally changing what the contributor submitted.
Where does normalizing end and fundamental change begin?
A bio is a small act of self-presentation. They chose what to mention, what to emphasize and how to sound. Their photo is how they chose to appear. Both belong to the contributor. If the organization silently rewrites the bio into a different voice, or “improves” the photo, it has changed the person. That’s a trust violation, and it’s invisible to the contributor until they see themselves rendered as someone slightly other than who they presented.
But the organization does need consistency. We want them in third person, with publication names formatted consistently every time. Photos should be set within a consistent frame. Twenty contributors’ formatting choices can’t become twenty different presentations on the same page.
The design resolves this with one principle: normalize the form, never alter the content. Change first-person to third-person, but don’t change what’s said. Crop the photo, but don’t retouch it. The boundary between normalization and editing is the entire design.
Parallel normalization paths that join at a single contributor record.
When a content manager approves a submission, the pipeline takes the contributor’s raw inputs — bio, photo, name, social links — and runs them through parallel normalization paths. The bio goes to Claude for a structural review and rewrite. The photo goes to Cloud Vision for face-aware cropping. Name and social links run through deterministic normalization. The results join at a single PublishPress author record, deduplicated by email, so the same person never becomes two contributors.
The normalized record becomes part of the draft a content manager reviews before publishing. A human always sees the result before a reader does. The automation removes the manual production work, but it doesn’t remove the oversight.
Everything that would have crossed from normalizing into editing.
The whole design lives on one line: normalize the structure, but never alter the content. Each rejection is a place where automation could have crossed the line while appearing to be an improvement.
Three normalization paths, one deduplicated record.
Claude rewrites the bio from first-person to third-person, italicizes publication names, and strips inline social handles, which are captured separately as structured links. The prompt is tightly scoped to formatting transformations. Claude is explicitly instructed not to alter wording, change emphasis, or “improve” anything.
The handle is removed from the prose and normalized into the contributor’s structured social links. Nothing about what the contributor said is changed.
Cloud Vision detects the face in the uploaded photo and the pipeline crops centered on it, so the contributor stays framed regardless of how the submitted photo was framed. When no face is detected, it falls back to WordPress’s default crop rather than forcing a guess.
Email-based deduplication decides whether to merge into an existing contributor record or create a new one, so the same person never becomes two authors. A three-field name system prioritizes pseudonyms, so they can choose to use that rather than their legal name. Social URLs are normalized across six platforms. Handles, full URLs, and partial inputs all resolve to a single canonical format.
Consistent presentation, untouched voice.
Preserved
- The contributor’s voice. What they wrote stays what they wrote
- What the contributor chose to emphasize about themselves
- The contributor’s actual photo. Just cropped, never retouched
- Pseudonyms, handled with priority
Now possible
- Consistent contributor presentation
- No manual avatar cropping or bio reformatting
- Duplicate-free author records
- The pattern ports to any system needing to normalize user-submitted profiles, like CRM contacts, directory listings, marketplace seller profiles