Skip to main content

The Scoring Jury

Early discovery

This page describes what we want a contributor jury to achieve and the direction we are currently exploring to get there. Specific mechanisms — panel size, term length, how questions reach the panel — are not yet settled.

The most consequential parameters in CORPUS are the ones that decide which contributions become economically visible: the scoring dimensions, their weights, the integrity rules, the temporal decay curves. We do not want those parameters to be set by the people who run CORPUS day-to-day. We want them to be set with the participation of the contributors whose work they govern.

What we want this body to do​

A few outcomes define the goal:

  • Review scoring outcomes in aggregate and ask whether the system is producing the distribution it ought to produce — or whether it is reproducing biases that the algorithm has encoded but not declared.
  • Recommend recalibration of scoring dimensions, their weighting, the temporal parameters, or the boundaries between integrity flags. Concretely: if noise music or other deliberately lo-fi work consistently scores low on a "production quality" metric that quietly equates clarity with value, this is the body that flags it for correction.
  • Set constraints, not code. The jury should not modify the algorithm directly. It should set the constraints within which the algorithm optimises — and any change to those constraints is logged and explained.

That is the goal. The mechanism by which we get there is less settled.

The direction we are currently exploring​

We want to avoid two failure modes at once: a panel that produces opinion noise rather than meaningful input, and a panel captured by the loudest or best-connected voices in the contributor base.

The direction we are currently exploring is a small contributor panel selected by stratified lottery (sortition) — stratified by region, genre cluster, catalogue size, and time of entry, so that no single demographic dominates. The reasoning behind sortition rather than election: in networks with power-law participation, elections reproduce that topology. Lottery breaks it.

Members would be compensated from the revenue pool. Governance work is a contribution on par with annotation or metadata curation, and should be recognised as such.

What is still open​

A great deal:

  • Panel size, term length, rotation. Open.
  • How questions reach the panel. Who decides what the jury reviews, on what cadence — open. We are building a deliberation tool alongside the protocol that will likely shape this answer.
  • What happens to recommendations. Whether the jury's output is binding within its mandate, advisory to a separate decision-maker, or something in between — open.
  • The mandate itself. Scoring is the natural starting point. Whether the same body extends to licensing principles, the operational/royalty split, or other parameters is part of what the pilot is meant to discover.

The honest framing is that we know what we want this body to achieve and we have a leading direction for how. The details will be worked out with the first cohort of contributors who sit on it.