Skip to content

Methodology · working draft

How a number is made, and what it can't claim.

A number you can't check is only a sales claim. Here's how we arrive at ours. The method will be versioned, and every report will name the version it used.

Principles

Four rules that override everything else.

Not measured isn't zero
If a source doesn't answer or a page has no counter, we record a gap and leave it out of the totals.
No media handling
We work from public text, metadata and tracker counts. We don't fetch video or thumbnails from monitored sites.
A person confirms
Automation finds candidates. A match counts as confirmed only when a named reviewer on your side, or a licensed matching service, says so.
Show the range
Every estimate comes as a range with its confidence, plus the reasons for any caveats.

Matching

Three tiers, and only one counts as confirmed.

Confirmed
A rightsholder's reviewer confirmed the match, or a licensed reference-matching service did.
Likely
Exact title and studio, with duration inside a set tolerance. Queued for review.
Candidate
Partial match. Report only.

Known failure modes: a title can claim content a file doesn't contain, titles can be generic, and compilations and trims break duration checks. A licensed copy can look exactly like an unauthorized one, which is why each client supplies an allowlist. We publish how often reviewers reject each tier, so you can see how good the “likely” tier really is.

Reach

A range, built from public counters.

reach_low  = Σ counter growth, confirmed URLs
reach_high = reach_low + Σ (counter growth × p), likely URLs
p = upper bound of reviewer-confirmed precision for the tier

With too few reviewed matches to estimate p, we publish only the low figure and mark the high figure as not measured. Candidates are never counted. Counter growth starts when we first see a URL, so earlier views are missed. Counters are self-reported by the site and can be inflated. This measures listed attention, not people.

Torrent swarm size is reported on its own line and never added to reach. We use the largest count from any one tracker rather than a sum, because peers overlap between trackers.

Tracking

Every confirmed URL is re-checked.

Each confirmed URL is re-checked by HTTP status and page text at 1, 3, 7, 14 and 30 days after we first see it, and again for every report. The states are live, gone, soft-removed (the page loads but the content is gone), geo-blocked, inconclusive and reappeared. We report that a URL is gone, not why: we can't tell from outside whether a notice, the uploader or the site removed it.

Evidence

Evidence you can check without trusting us.

Each finding can be bundled with the observations behind it, hashed, and signed with a published public key, so anyone can confirm later that nothing changed. Bundles hold URLs, timestamps, status and scrubbed text, and no images. A signature shows the record hasn't changed since it was signed. It doesn't show that a court would accept it, which is a question for your counsel.

What we won't claim

Limits we'd rather state ourselves.

  • Complete coverage. Our findings are a sample of the public web, and private channels are invisible to us.
  • Lost revenue as fact. Any dollar figure is a scenario on your assumptions.
  • That a site is breaking the law. We report URLs observed with matched content.
  • That a URL that's gone will stay gone. That's why we re-check.

Have your counsel read this.

We'd rather answer hard questions about the method before you rely on a number.