What the numbers mean—and what they do not.
This is an evidence-linked archive of one channel’s filmed experience, not a controlled safety study or a claim about every Tesla on every road.
1. Source record
The core record is DirtyTesla’s public long-form YouTube catalog. YouTube Shorts are excluded because they typically repeat moments from full videos and lack the surrounding drive context needed for reliable extraction. Video metadata, captions or transcripts, and visible or narrated drive details are used to identify software versions, mileage, locations at a broad level, and notable moments. Public DirtyTesla posts may add rollout context.
2. Event extraction and review
Machine-assisted extraction finds candidate disengagements, interventions, near misses, and highlights. Each retained event includes a timestamp link back to the video so viewers can inspect the underlying moment. Community “flag incorrect” votes create a review signal; they do not silently rewrite the record.
3. Mileage and version matching
Mileage can come from drive logs, totals stated in a video, and reviewed matches to private trip records. The public site receives aggregate mileage—not precise private trip endpoints. Software strings are normalized into version buckets so point releases can be compared at a useful sample size.
4. Rate metrics
- Miles per disengagement is observed filmed miles divided by on-camera driving disengagements.
- Miles per intervention uses accelerator taps, nudges, or other manual assists identified on camera.
- Miles per takeover combines disengagements and interventions.
- Rate charts require at least 40 logged miles in a version bucket. Counts without enough mileage can still appear elsewhere, but should not be read as a rate.
- Parking demonstrations, deliberate stress tests, and other mismatched observation windows are excluded from driving-rate calculations when identified.
5. Limitations
- This is one creator’s routes, vehicles, driving style, weather, and filming choices. It is not a random sample of the fleet.
- A video is not a uniform unit of exposure: duration, road difficulty, and mileage vary.
- Older captions and metadata are less consistent, and some drives do not have reliable mileage.
- Software behavior can vary by hardware, region, map data, vehicle, and incremental build.
- A missing event means it was not identified in the reviewed public record—not that an issue could not have occurred.
6. Privacy and corrections
Precise private trip endpoints and raw trip-review records are kept out of the public website. Public output is limited to the detail needed for aggregate analysis and video verification. If you spot a bad timestamp, category, version, or description, use the event flag so it can enter the review queue.
7. Affiliate, referral, and sponsor policy
Commercial links are labeled as affiliate or referral links near the recommendation and use sponsored-link attributes. DirtyTesla may receive money, store credit, service credit, or another benefit from a qualifying action. Sponsorship pays for disclosed placement—not a favorable conclusion. Partners cannot alter metrics, suppress corrections, or receive private trip data.