# Statistics

Generated from the released JSON. The test-split QA counts are the denominators
of QA Acc. Per-clip segment counts are part of the withheld ground truth.

## Test split (ground truth withheld)

| Track | Task | Clips | QA items | Hours |
|---|---|--:|--:|--:|
| Daily | IC | 100 | 507 | 5.16 |
| Daily | ST | 100 | 241 | 5.13 |
| Daily | LT | 100 | 252 | 5.71 |
| Science | IC | 30 | 239 | 1.80 |
| Science | ST | 35 | 104 | 2.63 |
| Science | LT | 35 | 157 | 2.05 |
| **Daily total** | | **300** | **1000** | **16.00** |
| **Science total** | | **100** | **500** | **6.47** |

## Dev split (ground truth released)

| Track | Task | Clips | Segments | of which open | QA items |
|---|---|--:|--:|--:|--:|
| Daily | IC | 14 | 31 | 0 | 69 |
| Daily | ST | 14 | 62 | 15 | 27 |
| Daily | LT | 14 | 57 | 24 | 34 |
| Science | IC | 3 | 14 | 0 | 25 |
| Science | ST | 3 | 71 | 1 | 7 |
| Science | LT | 2 | 15 | 5 | 7 |

Dev is a **subset** of test: its 50 clips are also scored in the test split.
Dev exists to validate a
harness end to end, not as a held-out set.