Recommended

Ranked recommendations

Under-cited feedMaterialized ranking run

Low-cite candidates that appear stronger than their current attention level.

Papers come from a materialized ranking run. The explanations reflect the recorded signal weights for the resolved run. The undercited family is scoped to a frozen low-citation candidate pool for the same corpus snapshot.

Run context

Pinned

Rows surfaced

15

Top score

0.700

Snapshot

Snapshot 2026-05-21

Showing 15 papers from a materialized under-cited ranking run; 15 include topic labels.

Topic labels are imported metadata and can be noisy; use them as coarse navigation hints, not authoritative classifications.

Focus paper: https://openalex.org/W4405876086 is not in the current top 15 rows for this view, but the run context is still pinned while you switch families.

Technical run metadata

Resolved run label: shadow-generalization-product-candidate-ranking-v1.

Ranking mode: materialized_heuristic.

Run label filter: shadow-generalization-product-candidate-ranking-v1. Resolved run: rank-83787b91ef; snapshot: source-snapshot-shadow-generalization-v1-20260521.

Storage provenance: materialized family rows are read from paper_scores. Result ordering is final_score desc, work_id asc. The undercited pool definition is documented in docs/candidate-pool-low-cite.md.

Live ranking surface

Under-cited results

Family: undercitedOrder: score desc, stable tie-breakLimit: 15

Current ML boundary

Emerging can be reordered by the bounded ML scorer only when the backend rollout gate is enabled. Bridge and Under-cited are not served by that scorer; their order comes from materialized ranking rows.

Heuristic baseline

The rule-only undercited list (/api/v1/recommendations/undercited) uses the same pool definition but is not tied to a corpus snapshot. For snapshot-scoped A/B against the ranked undercited family, use Evaluation.