The CompareDataBuyers format snapshot reports four row counts: source code 8 of 14, audio 3 of 14, video 3 of 14 and images 2 of 14. Every row uses the same base, the 14 researched organizations it classified across four selected data categories. Each row is easy to quote alone. Read together, the rows need some arithmetic, because each organization is counted once within a category and one organization can appear in several rows.

Step 1: add the rows. 8 + 3 + 3 + 2 = 16. That is 16 tags, not 16 organizations, and it is more than the base of 14.

Step 2: count who carries a tag. The record list marks four organizations as not listed in these four categories: Handshake AI, micro1, Scale AI and SuperAnnotate. That leaves 10 of the 14 with at least one tag.

Step 3: find where the extra tags come from. Sixteen tags over 10 organizations means some carry more than one. Defined.ai carries all four (4 tags). Sieve carries audio, video and images (3). Protege carries audio and video (2). The other seven carry source code only (7). 4 + 3 + 2 + 7 = 16.

Step 4: check the overlap between code and the media rows. Defined.ai is the only organization in both the source code row and any audio, video or images row. So the 8 source code profiles and the 3 organizations in the media rows add up to 10 organizations, not 11.

What this changes when you quote a figure
- "8 of 14" is a correct proportion of the 14 profiles for source code. For audio, video and images together, the count is 3 organizations, not 3 + 3 + 2.
- No row adds up to the 14. Four of the 14 carry none.
- A row counts tags under the snapshot's definitions: source code means a code or repository category, rather than documentation alone, and audio, video and images refer to the corresponding profile categories. Sources can also describe different program tracks.

The table and every record: https://comparedatabuyers.com/research/data-buyer-formats/

Published by CompareDataBuyers. Program requirements can change; confirm current terms with the program.