Zovda
Datasets are often packaged in tiers, and the difference between them is usually undocumented. A buyer compares two numbers and assumes the gap is quality, when it might be coverage, freshness, or support.
A tier is only meaningful if it names what changes.
The three axes worth tiering
- Density: how much of the set passed a stricter review
- Freshness: how recently the rows were collected or verified
- Breadth: how many categories, languages, or sources are included
Tiering on all three at once produces a matrix nobody can compare. Pick one per product.
Say what the top tier adds
"Premium" is not a specification. If the top tier has double review, say so, give the acceptance rate for each tier, and let the buyer price the difference.
Keep a shared core
If each tier is built from a different collection, they are three products, not three tiers. A shared core, plus additions, keeps the comparison meaningful.
Publish the acceptance rates
The number that makes tiers honest is the pass rate on a stated check, per tier. Without it, the price difference is a marketing decision rather than a measured one.
Zovda ships each collection through Datuva with its review rate and its coverage stated per tier. The corpus catalogue shows how each family is described.
Tell us which axis matters to your training run. We reply within one business day.