White Bastion is an invalid-traffic detection layer for podcast advertising. Every download request passes through a prefix URL, is examined for whether it came from a real listener, and the result is a defensible audit at show and campaign level.

The four stages

  Prefix              every download request in the feed is logged
     |
  IAB v2.3            duplicate and non-qualifying requests removed
     |
  Cross-examination   network characteristics and behavioural patterns
     |
  Scoring             signals combined into one score per request
     |
  Audit               per show, per campaign

1. Prefix

A short URL in front of the audio files in a feed. Every download request reaches it before the host, is logged, and is redirected onward in milliseconds. Nothing about the audio, the feed or the hosting platform changes — see Adding the prefix for what is involved, and Data and privacy for what a request record contains.

This is the only stage that touches delivery. Everything after it happens on data already collected.

2. IAB v2.3

Requests are counted to the IAB Podcast Measurement Technical Guidelines v2.3 before any judgement is made about them. Repeated requests from the same device for the same episode within a 24-hour window collapse to one, and requests that never transfer enough of the file are not counted at all.

This produces the IAB-counted download number — the same basis reputable hosting platforms report on, and the figure everything downstream is measured against.

3. Cross-examination

Each counted download is examined from more than one angle. Two broad families:

• Network characteristics — what the request itself reveals about where and how it originated.
• Behavioural patterns — how requests relate to each other over time, across episodes, and against a show's own history.

Neither is decisive alone. A single request that looks unusual is not a finding; the same pattern across a show's traffic is. Examining several independent angles is what separates the two, and it is why a verdict is not a single test with a single answer.

4. Scoring

The signals are combined into one score per request, and the score decides which class a request falls into: counted as valid, or set aside as general or sophisticated invalid traffic. See What we measure for what those classes mean and how they reach a report.

Thresholds are calibrated against traffic we have analysed rather than set arbitrarily, and they are versioned: any figure is scored under a stated version of the rules, so two reports are always comparable or explicitly not.

The individual signals and their weights are not published — see What we measure for why.

What comes out

A per-show audit report, and a campaign-level view across the shows a campaign bought. Both carry the full reconciliation chain from requests received down to verified downloads, so any figure can be traced back through every stage above.