The full per-video composite for a single video: identity, internal metrics (GMV, views, engagement), and the UVI creative analysis (hook, transcript, on-screen text, structure, detected products, vision tags, and scored features).

A video the shop owns but the pipeline has not analyzed yet returns with uvi_analyzed: false; a video the shop does not own returns 404. Set include_full_features=true to also receive the full scored-feature block.

Recommended fields to weigh (also returned as field_guide):

• hook_summary — How the video opens — the first seconds decide whether viewers stay, so a weak hook predicts weak reach.
• transcript — What the creator actually says — check the product claims are accurate and the messaging fits the brand.
• on_screen_text — On-screen captions/text (OCR) — catches claims, prices, or disclaimers shown but not spoken.
• structure_outline — The beats of the video (hook, demo, CTA) — a clear structure signals a creator who can sell, not just post.
• detected_products — Products the analyzer saw on screen — confirms the creator actually featured the product, i.e. product fit.
• vision_tags — Visual tags for setting, style, and content type — a fast read on production quality and brand-safety.
• features.hook_present_first_3s — Whether a real hook lands in the first three seconds — the single strongest retention signal.
• features.has_product_demo — Whether the product is actually demonstrated — demos convert far better than mentions.
• features.solution_shown — Whether the video shows the product solving a problem — the core of a persuasive sample video.
• features.has_voiceover — Whether the creator narrates — voiceover content tends to feel more authentic and engaged.
• features.video_pace — Editing pace — matches (or clashes with) the brand's desired energy and the platform's norms.
• features.content_summary — A one-line summary of the video — the quickest gist before opening the full analysis.
• metrics.views / metrics.video_gmv — Actual reach and sales this creator's recent content drove — the outcome the sample is betting on.
• creator.sample_received_num / creator.curr_status — How many samples this creator already got and where they stand — flags repeat requesters and stalled relationships.