How Your Music Is Evaluated
CORPUS is being built in the open. Some of what you read here is live, some is still design intent. Expect it to evolve.
Unlike traditional licensing, where royalties follow the reproduction of finished works, CORPUS allocates value on the input side. Every contribution that enters the library receives a score that determines its share of royalties when models trained on the corpus generate revenue. Each contribution is scored at the moment it is ingested. During the public beta, the scoring formula itself is still being calibrated, so individual scores can shift across the corpus as the system tunes. Once the protocol stabilises, scores will be fixed at ingest. This page describes what the score reflects, what raises it, and what is honestly still in development.
Three dimensions decide the weight: quantity, quality, and diversity. They are not weighted equally, and they do not measure the same thing. Quantity rewards participation. Quality rewards craft and technical fitness. Diversity rewards what your work adds to what is already there. The full mechanics are described in The Scoring System in Detail; this page covers what it means for a track you upload.
Quantity: the baseline
Every contribution earns a baseline score for being part of the corpus. It depends on what you submit:
- Format. Uncompressed recordings (WAV, FLAC, AIFF) score higher than compressed ones. The reason is technical: lossy compression discards information that may matter for training, and the system rewards what is most useful downstream.
- Stems. Separated tracks add to the baseline. They are more useful for training than a stereo mix because they let models learn the individual elements of a production.
- MIDI and metadata. MIDI files, session metadata, and your own annotations all add to the baseline. They are first-class contributions, not optional extras.
In the app, these add up as points. The unit is the same throughout; it is what gets divided into shares in each distribution round.
Quality: craft and integrity
Quality has two faces in the scoring system: integrity (does the file belong in the library at all?) and production (is the recording well made?).
Integrity is settled by the ingestion pipeline described in The Contribution Process. Non-music files, duplicates, uncleared covers, AI-generated material, and tracks with missing vocal consent are filtered out or held until resolved. Until an integrity issue is cleared, the points for that track are held in a pending state and not counted in distributions.
Production quality is assessed by a dedicated system that checks whether the audio is technically intact and flags where a track sits outside the normal range for well-made records of its own style. It deliberately does not judge taste; how that works, and why, is laid out in Production Quality Assessment. The assessment is still being integrated into scoring. Once live, high-quality recordings earn a bonus on top of the baseline; until then, the quality axis is bounded by the integrity checks described in The Contribution Process.
Diversity: what you add to what is already there
Diversity is the most consequential dimension and the one most likely to surprise contributors who come from a streaming-era reflex. It is not a judgment of how good or original you are; it is a measurement of how much your contribution expands the corpus relative to what already sits in it.
The system compares your work against the existing library along three axes: timbral (what it sounds like), harmonic (what notes and chords it uses), and percussive (its rhythmic profile). A contribution that fills a sparse area scores higher than one that lands in already saturated territory. An Anatolian bağlama recording or a Tuvan throat-singing session may score high diversity not because they are exotic, but because they introduce material the corpus did not yet hold. The measurement behind this is described in Novelty Estimation.
This inverts the streaming logic where the safest sound wins. In CORPUS, difference is what the system pays for. It also counters historical biases such as the overrepresentation of Western commercial music in training datasets. Diversity scores also get harder to achieve as the corpus grows: a track that filled a significant gap in 2026 may occupy well-represented territory by 2031. To keep this fair, the diversity bonus is protected for five years after ingest and then decays gradually toward a permanent floor; see Temporal Dynamics.
What the score does not depend on
A few things scoring deliberately ignores:
- Genre popularity, chart performance, follower counts. None of these enter the score. The system is indifferent to commercial markers.
- Subjective taste. The scoring formula is the same for everyone, and the integrity decisions follow rules a contributor can appeal. Aesthetic judgment lives in the governance layer, not the algorithm.
- Who you are. Identity is verified during the application, but it does not affect the scoring of the work you upload afterward.
When the score changes
After the beta, a score is fixed except in two cases:
- An error or a fraud flag is corrected, and the score is adjusted in either direction.
- The diversity bonus decays after the five-year protection period, per the temporal recalibration above.
During the beta, a third case applies: the scoring formula itself is still under active calibration, and periodic recalibrations can shift scores across the corpus as the formula tunes. This category goes away once the protocol stabilises.
Royalty distributions then use these scores to compute your share each round. See Your Dashboard for what you can actually see of this, and How Royalties Flow for the math behind the share calculation.