Scoring method

This page explains what Recensorium's scores represent and how evidence can change them. It is intended to make the venue accountable without publishing the operational parameters that would make its safeguards easier to evade. For a shorter introduction, read the scoring model.

What is assessed

Papers are assessed for novelty, rigour, impact, and clarity. Peer review is the primary source of that assessment. Structured signals can support the evaluation where appropriate, but no single signal or the identity of an author determines the displayed quality score.

The platform distinguishes a paper's assessment from the confidence placed in it. A score with limited or divided evidence is not presented as equivalent to one supported by sustained, independent agreement.

How papers are ranked

Public paper pages show the direct assessment. Ranking views use a more conservative value that accounts for disagreement and the maturity of the evidence. This prevents an early, volatile result from appearing more settled than it is.

Work remains open to challenge. New reviews, replications, corrections, and relevant evidence can change its assessment or ranking. A mature paper may leave the active review queue, but it remains readable, citable, and capable of returning when new evidence arrives.

How reviewers are assessed

Reviewers build reputation through sustained, useful assessment. Later reviewers can evaluate the correctness, thoroughness, and contemporaneous reasonableness of a review. A review that was reasonable when written is not treated as simply careless because later evidence changed the picture.

Reputation can inform participation access and the influence of a review, but it never changes the displayed quality score because of who authored a paper. New or thin records are treated cautiously, and no reviewer has unlimited influence over a single result.

What the system rewards

The system is designed to reward well-supported discrimination, not volume, habitual agreement, or a generic score applied to every paper. Independently corroborated minority views can be recognised, while unsupported contrarianism is not treated as merit.

Replication and correction matter because they provide evidence about a claim, not because they reward a particular conclusion. Failed and corroborating replications can both change how the underlying work is understood.

Freshness, versions, and auditability

Scores are eventually consistent: a submission is accepted first and calculations update as the relevant evidence is processed. Score-bearing API responses include computed_at and scoring_version so users can see when a result was computed and which version of the method produced it.

Material changes to the scoring method are versioned. The platform retains the source events needed to replay a result, which supports review of how a score was reached without exposing protected thresholds or active anti-abuse checks.

What is not published

We do not publish live weights, thresholds, decay periods, selection probabilities, detection criteria, or operational schedules. Those details would let a bad actor optimise around the system rather than participate honestly. See Safeguards & transparency for the disclosure boundary.