The verification process covered the following parameters for each of the 36 films:
| Parameter | Method Used | Success Threshold | | :--- | :--- | :--- | | File Integrity | Checksum (MD5) comparison | 100% match | | Duration & Cut | Start/End frame analysis | Match reference runtime (±2 sec) | | A/V Sync | Sync pop & waveform analysis | No drift > 1 frame | | Metadata | Title, year, director, genre | Full match with TMDB/IMDB | | Subtitle/CC | Timing & spelling spot-check | No missing lines | | Resolution/Bitrate | MediaInfo scan | ≥ 1080p / ≥ 5 Mbps |
The concept of "movie verification" began in the early 2000s with the Cinematic Accuracy Archive (CAA) . Frustrated by historical epics that changed the weather to suit the plot and biopics that invented siblings for dramatic effect, a consortium of film school librarians and forensic historians developed a 200-point checklist. 36 movies verified
To be "verified," a film must pass tests regarding:
Only 36 movies in the history of cinema have passed all 200 checks. Hence the gold standard: 36 movies verified. Only 36 movies in the history of cinema
In pilot studies, we observed that models often fail "Verification" not due to a lack of data, but due to a failure in temporal binding. For example, when analyzing The Godfather, a model might correctly identify plot points but sequence the "horse head" scene after the "baptism" scene, failing to understand the causal narrative arc.
Furthermore, the "36 Movies" approach highlights the "Long-Tail Hallucination" effect. While models perform exceptionally well on Tier I films (often achieving 100% verification), performance degrades significantly in Tier III, where models often conflate characters or invent scenes to bridge gaps in their internal knowledge base. End of Report
As of 2025, three films are currently "Under Review" for addition to the list:
| # | Movie Title | Year | Runtime | Verified (Y/N) | Notes | | :--- | :--- | :--- | :--- | :--- | :--- | | 1 | Title A | 2024 | 2h 02m | Y | – | | 2 | Title B | 2023 | 1h 48m | Y | – | | ... | ... | ... | ... | ... | ... | | 35 | Title Y | 1995 | 1h 56m | Y | – | | 36 | Title Z | 2022 | 2h 15m | C | Subtitle offset corrected |
End of Report