English
Traffic allocation
These settings decide how much volume head ads get, and how a new ad category is explored on a site before it has enough data.
Premium Ad Percentile Threshold
| Field | Value |
|---|---|
| Admin name | Premium Ad Percentile Threshold |
| Key | ad.premium.percentile.threshold |
| Control | Slider |
| Format | Integer 0–100 |
| Example | 25 |
What it does
Among currently matchable ads, this share of the top ranks is treated as the premium pool. 25 means the top 25% are premium; the rest are regular.
The premium pool then receives more impressions via the next setting.
Impact
- Lower (e.g. 10): a narrower head, harder for mid-tier ads to show.
- Higher (e.g. 50): half the ads count as premium, so the tilt weakens.
- Avoid 0; the tilt can fail or become extreme.
Premium Ad Traffic Weight
| Field | Value |
|---|---|
| Admin name | Premium Ad Traffic Weight |
| Key | ad.premium.traffic.weight |
| Control | Slider |
| Format | Integer 0–100 |
| Example | 70 |
What it does
When several candidate ads are needed, impressions are allocated with a power-law bias toward the premium pool. This is roughly the share of opportunities premium ads receive.
Example: threshold 25 and weight 70 → about 25% of ads take about 70% of impressions; the rest share 30%.
Impact
- Raise: head ads are more stable, eCPM more concentrated, long-tail ads learn slower.
- Lower: fairer split, better for testing, head advertisers may feel diluted.
- Always read it with the percentile threshold. Changing only one is hard to predict.
Category First Exposure Size
| Field | Value |
|---|---|
| Admin name | Category First Exposure Size |
| Key | category.first.exposure.size |
| Control | Text |
| Format | Positive integer |
| Example | 500 |
What it does
The system tracks, per site, how each ad category performs, then prefers categories that work on that site.
Until a category has this many impressions on the site, its real (often low) CTR is not used to punish it. It gets an exploration weight based on the better categories already measured, so new categories can still show.
After the floor, real CTR drives the recommendation.
Impact
- Raise: longer exploration, more even category mix, some volume may go to a poor-fit class.
- Lower: the site narrows to proven classes faster, and may drop a class before it was tested.
- Similar idea to CTR minimum sample size, but only for “this site × this ad category”, and it does not zero the overall ad or site score.