ecommerce_video_brief_quality_checklist

# E-commerce Video Brief Quality Checklist This dataset provides 25 structured quality checks for reviewing an e-commerce product-video brief before production or generation begins. Each row represents one check that a reviewer can answer from the brief itself. The checks cover objective, product identity, scene construction, motion, composition, style, copy, output requirements, and final quality control. The dataset is intended for workflow prototyping, rule-based validation, annotation exercises, and small supervised-learning demonstrations. It is not a benchmark of model performance and it contains no generated-video ratings. Instead, it turns common brief-review questions into a compact tabular schema that can be used to build a checklist interface, compare human review decisions, or test a classifier that predicts which omissions are most likely to block production. ## Columns - `check_id` is the stable integer identifier for each check. - `stage` groups checks by the part of the workflow they govern. - `check_name` is a machine-friendly label. - `question` contains the review question in plain English. - `evidence_type` describes the expected form of evidence, such as text, asset, enum, list, sequence, number, or boolean. - `severity` uses high, medium, or low to indicate how strongly a missing answer can affect execution. - `required` marks the baseline checks that should normally be answered before work starts. - `example_pass` gives one concise example of an acceptable answer. Examples illustrate structure only; they are not universal requirements. ## Suggested Uses A simple validator can convert the `required` column into blocking rules, then use `severity` to rank unresolved items. Teams can also add columns such as reviewer answer, pass/fail, or revision note and use the resulting table as an annotation template. For machine-learning exercises, the categorical fields can support classification or clustering demonstrations, while the question and example fields can support text-embedding experiments. One practical workflow is to complete this checklist before moving a brief into an image-to-video or text-to-video tool. For example, a browser-based workflow such as [Kling 3.0 AI Video Generator](https://kling3ai.co/) can consume prompts and source images, but the checklist remains tool-independent: it focuses on whether the human brief clearly defines the goal, references, camera direction, visual constraints, duration, and acceptance criteria before any generation attempt. ## Scope And Limitations The rows are editorial rules authored as a reusable template, not observations collected from people or production systems. Severity labels are heuristic and should be adjusted for the product category, channel, legal requirements, and review process. The dataset does not establish that completing every row improves conversion, visual quality, or model accuracy. It also does not replace factual claim review, brand approval, rights clearance, accessibility review, or a comparison between generated output and the real product. Physical accuracy deserves special attention. Product geometry, color, labels, controls, and scale should be checked against approved reference assets. The checklist therefore separates product identity, reference assets, visual accuracy, negative constraints, and acceptance criteria instead of treating them as one generic prompt-quality field. The file is released under CC0 so it can be adapted into internal review forms, teaching examples, or validation utilities. When extending it, keep check identifiers stable or document any remapping so downstream annotations remain comparable.