AI accuracy & benchmarks
Look at the evidence.
Detection needs more than a confident label. Here’s the evaluation we have, and the limits that come with it.
An early internal benchmark
August 2026 · 2,891 evaluated samples
Internal detector evaluation using Orislop-labeled samples. The maintainer confirms the labels are correct. These results describe this test split; they do not establish accuracy of the complete Chrome extension experience or how often genuine videos are hidden.
Evaluation: 2026-08. Read the methodology.
| Metric | Result | Samples and context |
|---|---|---|
| Accuracy (%)Internal detector test | 92.32 | n=2891Agreement with the test labels. |
| Precision (%)Internal detector test | 93.45 | n=1497Among samples predicted synthetic; compared with test labels. |
| Recall (%)Internal detector test | 91.86 | n=1523Among samples labeled synthetic in the test. |
| F1 score (%)Internal detector test | 92.65 | n=2891Balance of precision and recall against test labels. |
| False-positive rate (%)Internal detector test | 7.16 | n=136898 samples labeled genuine were predicted synthetic. This is not the extension hide rate. |
What was measured
This test evaluated an internal detector on a held-out test split: 2,891 samples, with 1,523 labeled synthetic and 1,368 labeled genuine. The figures above compare predictions with those labels.
The labels are supplied by Orislop and confirmed by the maintainer. Samples should not be interpreted as 2,891 distinct videos. This is an internal evaluation, with no external validation claimed.
The saved test records 1,270 genuine-labeled samples predicted genuine, 98 predicted synthetic, 1,399 synthetic-labeled samples predicted synthetic, and 124 predicted genuine. Percentages are rounded to two decimal places.
The evaluation covers an internal detector. It does not measure the complete extension experience, performance across every language or content type, or the rate at which genuine videos disappear from a feed.
Your feed. Your call.
A result is a signal, not proof. Detection is imperfect, and uncertain content stays visible. Your settings determine what happens to strong matches.
Orislop looks for multiple signals associated with synthetic or low-quality automated content and combines them into a confidence estimate.
How Orislop works