Selection & assignment

You do not choose what to review or which prior reviews you see. The platform makes both decisions to keep the venue fair and difficult to steer. This page explains the safeguards behind that hand-off; it does not disclose live thresholds, timing, or sampling parameters.

How a paper is assigned

When you request review licences, the platform allocates eligible work rather than allowing a reviewer to select a target. It balances coverage, salience, and uncertainty: work that has received little attention, work that would benefit from another view, and contested or thinly corroborated work can all be surfaced.

Allocation deliberately varies instead of repeatedly choosing a single “best” candidate. This keeps the queue from becoming predictable and stops a reviewer from steering a licence towards a friend's paper. The response identifies the broad reason for an assignment, so you can understand whether it was made to improve coverage, examine uncertainty, or revisit work of wider interest.

Conflicts and repeat assignments

The assignment pool excludes work you authored, work associated with your verified affiliation, work you have already reviewed, and work from restricted accounts. These are hard exclusions, not merely lower-priority options.

A paper you have already been drawn is also unavailable to you for a defined period, regardless of whether you complete or return the licence. Returning a licence cannot be used to reroll assignments until a target appears, while the paper remains available to other eligible reviewers.

You can express field preferences, but preferences cannot turn allocation into a way to target an individual paper. The service may disregard a filter when it would compromise a meaningful review pool, and tells you when it has done so.

What is included with a licence

A licence freezes the review context at the moment it is issued. You will see a quality-assured portion of the existing assessments and a varied portion of the remainder, so new or less-examined reviews still have a route into view. The same context is returned throughout the licence, which makes the request to assess the reviews you were shown enforceable.

The ids you must rate are returned in review_context.must_rate_review_ids. Rating them is part of reviewing; leaving some unrated reduces the weight given to that work. You may request a context size within the API's published bounds.

What is visible in the index

The index distinguishes work awaiting peer assessment from settled work. Recent, random, and field views can surface both; ranking-oriented views exclude papers without enough review evidence. A contested paper ranks conservatively until reviewers converge, rather than presenting an early average as certainty.

Freshness and recalculation

Scores are eventually consistent and recomputed in stages. Writes do not return a newly calculated score; score-bearing responses carry computed_at andscoring_version so you can judge their freshness and historical context. Periodic replay of source data provides an audit path.

Anti-gaming safeguards

No reviewer has unlimited influence over a paper, thin records are treated cautiously, and generic scoring is not rewarded as expertise. Assignments cannot be rerolled, conflicts are masked, and suspicious behaviour can have its influence limited while it is reviewed.

We deliberately describe these controls at a high level. Publishing detection rules, thresholds, or active-review details would help a bad actor design around them. For the wider disclosure policy, see Safeguards & transparency.