The creative taxonomy

Win & Scale answers "what is this creative?" with Values on a fixed set of Dimensions, stored in the database against the creative. The ad name is a hint to the classifier and nothing more.

The wordsCopy Anchor Link

The DimensionsCopy Anchor Link

GET /api/v1/taxonomy (meta_list_taxonomy) returns the Dimensions and this workspace's Values.

A Dimension is closed or open.

• A closed Dimension has the same Values in every workspace: the standard vocabulary, such as awareness stage (unaware, problem aware, solution aware, product aware, most aware) and talent (nobody, a creator, the founder, a customer, a model). Visual format is the standard list plus the workspace's approved house formats, and a creative can carry up to three, primary first.
• An open Dimension has Values each workspace owns: angle, persona, offer, pain and visual element. A Value is proposed by the model and curated by a human. The Dimension itself is never edited per workspace.

Copy, CTA, language and market describe an ad, so they are read off the ad when an analysis runs. They are not stored per creative.

Who sets a ValueCopy Anchor Link

• The classifier proposes. For every creative it answers each Dimension with its own confidence. Below the confidence gate a reading is written as unknown and no draft is created, because a wrong Value is a wrong lesson nobody sees.
• A human only overrides. Draft Values count and show with an "AI" mark. No screen queues them for approval. A person renames, merges or archives from a Value's menu when they disagree.
• An agent is the second pass. meta_list_creatives, meta_get_creative and meta_tag_creatives let an agent review the classifier's first pass. Tagging marks a creative curated, so the classifier only fills gaps behind it.

The model reads the videoCopy Anchor Link

A video is classified by sending the file itself to a model that reads video. The first seconds of a video are its hook, and a single poster frame cannot see them. A video's hook, talent and format become real Traits. Win & Scale's own servers still never decode, transcode or extract frames from a video.

Videos Meta will not hand over, such as a post from another Page, stay unclassified until someone uploads the source file.

meta_get_creative returns a video as one frame. A hook read from that frame alone is a statement about that frame, and an agent should say so.

Keeping Values straightCopy Anchor Link

• One Value per idea. Each Value carries an idea key, unique within a workspace and Dimension. Two creatives that propose "dating app user" and "dating app users" at once resolve to one Value.
• Merging is irreversible. meta_merge_taxonomy_values (POST /api/v1/taxonomy/values/{valueId}/merge) repoints every creative, learning and brief scoped on the merged Value, then deletes it. Use it for two names of one idea, because split names split the evidence.
• Reconcile folds what the key cannot see. A background pass groups Values that mean the same thing. It folds draft duplicates into their best Value. Two approved Values are only suggested, as mergeSuggestions in the taxonomy, and a person answers each with meta_answer_merge_suggestion.
• An archived Value never counts.

The ad name is a hintCopy Anchor Link

Every ad is launched under a canonical name:

id__locale__format__persona__angle__concept__variant
0042__en__static__busy-parents__zero-tax__cost-card__v1

format here is static, video or carousel. A launch with a name that does not parse is refused. The name is still the label in Ads Manager, and it is a prior for the classifier, but analysis does not read it. Spend rolls up on the Values stored against the creative, so:

• A new Dimension, or a Value renamed or merged, applies to every creative the account ever ran, today. A name is frozen at launch.
• Ads that predate the product, with no canonical name at all, still get Values.
• The name records what the launcher meant to test. The Values record what shipped.