CONTEXT

CONTEXT is an evidence-first investigation workspace for trust and safety reviewers. Instead of scoring one post in isolation, it places the actual social content, conversation and timeline beside an explainable Meta AI recommendation. Four complete fictional cases are ready to explore: a carousel appeal where additional context changes the recommendation; a playable cooking Reel with object inspection and speech transcription; repeated contact after a request to stop; and investment posts with observable coordination. Reviewers can compare reported-item and full-context assessments, open every cited source, inspect alternative explanations and uncertainty, save a human decision, export the case, or upload authorized media. Muse Spark 1.3 performs multimodal reasoning. SAM 3.1 returns real segmentation geometry. Muse Voice Transcribe supplies actual speech transcripts. Recorded API responses are dated and labeled; live reruns are available. No invented assessment is presented as a live model result. No private Instagram history is accessed and no enforcement action is taken on Instagram. Video reasoning uses four sampled frames plus a separate transcript. The application, case data, interface and integrations were built during the hackathon. Pre-existing components include the Sites/Vinext starter, React/Vite/Cloudflare, dependencies and design skills. AI coding and image-generation assistance were used. Demo photos are generated; cooking footage is licensed from Gary Barnes/Pexels with synthetic narration added. The three-minute walkthrough uses actual application captures and model results, edited with narration. This is a working prototype and a simplified original policy exercise, not official Meta policy or an accuracy benchmark.