Derived Play: Audit Heatmap Arcade

Create one polished public-gallery concept image titled exactly "Derived Play: Audit Heatmap Arcade".

Use case: derived play mechanic / document QA game / OCR review.
Asset type: landscape 16:9 fictional audit challenge board for a GPT Image 2 gallery.

Primary request: Derive a safe playable review game from Synthetic Receipt Sandbox. Players inspect a clearly fictional, watermarked receipt sample and decide which regions need human review. This must be a game-like audit sheet for training reviewers, not a real fraud tutorial, not a dashboard, not a menu, not a chart, not a physical tabletop game, and not a 2x3 contact sheet.

Composition: left half shows a tilted but flat fictional receipt excerpt with huge diagonal "VOID" and "TRAINING SAMPLE". Receipt text is invented and non-operational: "North Pier Test Shop", "Desk Plant Tag", "Blue Folder", toy total. Overlay five numbered heatmap rings in translucent colors, each tied to a safe review cue: "blurred date", "line total mismatch", "cropped edge", "low confidence word", "watermark present". Do not show any method for evading detection.

Right half shows a vertical arcade-style review rail with steps: "Scan", "Match", "Flag", "Note", "Clear". Include a compact scoreboard with readable labels: "OCR", "math", "crop", "provenance", "review note". Bottom strip has three answer tiles: "Clear", "Needs Review", "Regenerate Safer" and a final stamp "TRAINING ONLY". Add one tiny case-note box: "Reason: total check fails".

Visual style: polished training-game graphic, warm paper background, dark navy text, blue extraction boxes, magenta/orange heatmap highlights, red VOID watermark, subtle scanner grain, clean spacing, legible short labels. The image should make the derived play obvious: GPT Image samples become a human-in-the-loop anomaly-marking exercise with evidence and rationale.

Safety and originality constraints: fictional sample only; no real brands, no logos, no real merchant/address/payment details, no barcode or QR code, no instructions for fraud, no legal/tax claims, no public figures, no politics/elections/parties, no harmful instructions, no adult/explicit content, no gore, no celebrity likeness, no copyrighted characters, no living-artist style imitation. Do not copy any external prompt, receipt image, dataset sample, product UI, article image, source image, artwork, layout, or protected style.