Most music-tech systems read metadata, count plays and produce charts. Nomad starts at the other end. We start with the audio — the actual waveform — and work outward to rights, royalties and audience signal. The principles below describe how that decision plays out in practice.
Every track we ingest is fingerprinted before we look at its ISRC, its label, or its release date. Metadata is wrong about 8–11% of the time at any large catalogue. Audio is the only source of truth that doesn't disagree with itself. We let the metadata describe what someone said the track is; we let the audio decide what it actually is.
Nomad Listen runs four matchers in parallel — tempo-invariant melodic, harmonic-progression, lyric-aware, timbre-only — and surfaces a derivative only when at least two of them agree above threshold. Single-detector systems either miss covers and edits or drown the rights team in false positives. The hard problem in audio matching is precision under fatigue, not recall. Our agreement rule is engineered for the team that has to read the results on Monday morning.
A work is not a row. It is a graph of splits, derivatives, sub-publishers, neighbouring rights and territory carve-outs that evolves continuously as deals get signed and PROs report. Nomad Rights models that graph natively. Every claim we generate has a path you can trace back to the audio, the territory, the PRO statement, and the contract clause that justifies it. Spreadsheets cannot do this; graphs can.
Our sub-genre embedding is built to be argued with. Every recommendation comes with three audio references and the implicit framing the model used. Your A&R team can disagree — and that disagreement is the most valuable signal we collect. The model gets better because working A&R correct it. The day it doesn't get corrected is the day it has converged on consensus, which is the opposite of what A&R is for.
If a number we surface cannot be defended to a rights committee, a CFO or a court, we don't surface it. That cuts about 40% of the metrics other platforms publish. The remaining 60% are the only ones that change what you do on Tuesday. We would rather give you fewer numbers you act on than more numbers you scroll past.
The royalty industry runs on a quarterly statement rhythm — and that rhythm is the reason 5% of the money goes missing. By the time you read the Q3 statement, the Q1 leak is six months cold. Nomad reconciles every statement against the works graph the moment it lands. Every discrepancy generates a tracked claim. Every claim ends in either a payment or an evidence trail explaining why it didn't.
The whole platform is calibrated against the working judgement of A&R, royalty and publishing professionals — labels with 4 people, labels with 800 people, managers running 6 artists. We do not believe an AI alone can run a label. We do believe a label run by people who listen, equipped with a platform that listens, will outperform a label run on dashboards. Everything we build is in service of that bet.
The royalty-recovery number reflects eight catalogues reconciled in pilot during 2025 — a small, deliberately varied sample (one major-distributed indie, three independents, two neighbouring-rights specialists, two publishing-first companies). Range was 3.4% to 9.7%; median 5.1%. The A&R recall figure was calibrated against signings made by twelve working teams across 2023–2025 and is updated each quarter as new signings come in.
A 30-minute listen and three specific things we can hear in your data — before either of us commits to anything more.