Source Boundary

Allowed sources for this run:

• Official AI Engineer World's Fair 2026 schedule endpoints.
• Public AI Engineer YouTube channel metadata.
• Public YouTube caption availability metadata.
• Public YouTube transcript helper output where captions were available.
• Local Whisper fallback output when captions are missing.
• Local video frame extraction, Tesseract OCR, RapidOCR/ONNX repairs, and OpenCV reconstructed slide crops.

The run did not import private Miami notes, personal recordings, queue state, or diary content.

Confidence Rules

• Official schedule facts are canonical for titles, dates, times, tracks, speakers, and affiliations.
• Related YouTube videos are supporting context unless the video is confirmed as the exact session recording.
• YouTube transcripts are treated as transcript evidence, but speaker attribution and session matching still need review when the video is only speaker-matched.
• Slide OCR is best-effort. Prefer reconstructed slide crops when the full-stage frame OCR is weak, and preserve unreadable/non-slide classifications rather than inventing text.

Current generated scale: 560 sessions, 553 people, 340 companies, 117 resources, 310 slide pages, 11 topics, and 5 event/day pages.