Rights & fingerprinting
Nothing publishes without being checked against the whole network
A duplicate found after publication is a claim, a strike and an argument. A duplicate found at the gate is a decision — with the matched asset, the matching segments and the partner who owns them attached to it.
- Videos
- 25.4K
- Content ID claims
- 602.0K
- Backed by a match
- 41.9K
- Channels reached
- 185.0K
- Awaiting a decision
- 27
earning under the content owner
262,755 still active
measured against a reference
outside the network
disputed or in review
Matching is performed on video and audio segments. Never on the thumbnail, never on the title — those match each other constantly and mean nothing.
The pre-publish gate
Fingerprint first, publish second
Every upload is scanned against the network catalogue before it can reach YouTube. The thresholds are policy values, seeded and configurable per deployment — not constants buried in the code.
- 1
Scan on video and audio
The asset is fingerprinted segment by segment on both tracks, producing a match percentage and the exact time ranges that matched on each side.
- 2
Resolve to a verdict
The match percentage falls into a configured band and resolves to blocked, review or auto-approved. The band that decided it is shown alongside the verdict.
- 3
Route the decision
Blocked uploads stop and notify. Review goes to an operations queue with the diff. Auto-approved publishes and joins the catalogue.
- 4
Record it
The decision, the reviewer and the note are written to the audit log and dispatched by email, WhatsApp and in-app, so the creator learns it from the platform rather than from a takedown.
Threshold bands
configurable in service settings- Above 90%Blocked
The upload is terminated at the gate. An infringement alert is logged against the channel and the partner is notified with the matched asset named.
- 50–90%Needs review
Routed to the operations soft-review queue with a segment diff, the matched asset and the owning partner attached, for a human decision.
- 25–50%Needs review
The band the source requirements left unspecified. We resolve it to review rather than auto-approve: a wrong auto-approve costs a strike against the CMS, a wrong review costs a few minutes.
- Below 25%Auto approved
Published, and added to the network hash catalogue so that the next upload is checked against it too.
Anatomy of a match
A claim you can argue with
A verdict on its own invites a dispute. The platform hands over the evidence with it: what matched, how much of it, against which recording of ours, and how long after the upload the claim landed.
- The reference is named, with its own length and the day it was fingerprinted, so a match can be weighed against what it was measured from.
- Coverage is given twice, because the two answers differ: how much of their video is ours, and how much of ours appears in it.
- The claim’s policy is shown beside the asset’s default, and any divergence is marked — the money follows the claim, not the asset.
- YouTube’s feed carries no segment offsets, so the exact in-video positions are not shown here and are not invented. Every claim links to its own page in Studio, where they exist.
A real claim
ExpiredNeeli Neeli Akhiyan || Raj Bhai Video || Awanish Babu & Shilpi Raj
HP Music Bhojpuri · outside the network
One of 41.9K claims produced by comparing a video against one of our references. The rest were made by uploading, and were never compared against anything.
How claims were made
602.0K claims- Uploaded by the owner291.3K
- Short made with our audio239.1K
- Fingerprint match36.6K
- Delivered over SFTP18.0K
- Uploaded through the CMS2.6K
Only a fingerprint match compares against anything. A claim on a video the owner uploaded, or on a Short built with our audio, was never measured — so its evidence page says so rather than showing an empty comparison.
Network-wide reuse
The same asset, on how many channels
Reuse inside a network is not always infringement. It is often syndication nobody wrote down. Either way, operations needs to see the whole cluster rather than one uploaded copy at a time.
- 602.0K live claims across 185.0K channels, each naming the channel hosting the content, what matched, and how much of the video it covers.
- Clusters are classified as exact, near duplicate or segment match, and can be filtered by kind and by a minimum number of occurrences.
- The dedup matrix supports a batch action across a cluster — make private, with the master protected by default.
- Analytics carries the same intelligence per asset, so a re-upload count can be read from the revenue side as well as the rights side.
Content ID
Claims and disputes worked in one place
Claims arrive against the network's assets from outside, and are raised by the network against others. Both are fetched against the root CMS token and worked in the same queue.
- Every claim carries its policy — monetise, track or block — beside the asset’s own default, with any divergence marked. The money follows the claim, so that is where a payout stops matching what the asset was set to do.
- Disputes and appeals are states on the claim, not a separate spreadsheet. Release and uphold are actions with a confirmation that names the asset and the consequence.
- Territory restrictions are visible in the same view, so a video blocked in one market is not read as a global failure.
- Privacy resolves to the most restrictive position: where partners disagree about an asset, the result is private.
Where claims ended up
602.0K totalStill open
needs a personAuthorisation across the estate
255 channelsA channel with no token is reported as not evaluated, never as earning zero. The two look identical on a payout line and mean opposite things: one is a channel that made nothing, the other is a channel nobody can see. Live token state needs a YouTube API client, which is not yet configured — so this says what the rule is rather than showing a count it cannot stand behind.
Tokens and governance
No authorisation means no evaluation, not a zero
A channel whose partner has not granted a token cannot be measured. Reporting it as zero revenue is a lie that looks like data.
- A channel holding no token is excluded from revenue, from RPM and from every leaderboard, and shown with a request-authorisation action rather than a zero.
- A token that is expired, revoked, failing to refresh or short of the scopes it needs is named as its own state rather than lumped into one broken bucket — each needs a different action.
- Token health and monetisation are independent. A demonetised channel can hold a perfectly good token, and a connected token proves nothing about whether the channel earns.
- Tracking freezes when a channel is delinked, so a trailing-window figure never quietly includes a channel that stopped reporting.
- The YouTube API allowance is finite and spent by bulk operations, so a run that would exceed what is left has to check the remaining budget before it starts rather than failing halfway.
- The $250 tracking floor is an advisory policy about how finely a channel is tracked, not a payment rule and never a block.
See the rights surfaces
Suspected reuse, claims and disputes, token health and the creator-side gate are all live in the platform with the network's own catalogue behind them.