# Literature verification notes

Verification date: 2026-08-14.

1. Wang et al. (2025), Item Level Exploration Traffic Allocation in Large-scale Recommendation Systems, arXiv:2505.09033.
2. Wang et al. (2025), Item-centric Exploration for Cold Start Problem, RecSys 2025.
3. Yao et al. (2024), Unveiling User Satisfaction and Creator Productivity Trade-Offs in Recommendation Platforms, NeurIPS 2024, DOI 10.52202/079017-2759.
4. Greenwood, Chiniah & Garg (2024), User-item Fairness Tradeoffs in Recommendations, NeurIPS 2024, DOI 10.52202/079017-3629.
5. Liu, Fang & Wu (2023), Estimating Propensity for Causality-based Recommendation without Exposure Data, NeurIPS 2023.
6. Wang, Bai, Sun & Joachims (2021), Fairness of Exposure in Stochastic Bandits, ICML 2021, PMLR 139:10686-10696.
7. Gan et al. (2024), Contextual Bandits with Budgeted Information Reveal, AISTATS 2024, PMLR 238:3970-3978.
8. Swaminathan et al. (2017), Off-policy Evaluation for Slate Recommendation, NeurIPS 2017.
9. Liu (2025), On Preference-based Stochastic Linear Contextual Bandits with Knapsacks, AISTATS 2025, PMLR 258:2890-2898.
10. Sankagiri et al. (2026), Recycling History: Efficient Recommendations from Contextual Dueling Bandits, ALT 2026, PMLR 313:1-20.

Boundary notes:
- ARCF, Match-Weighted Attention, Attention Waste, Routing Debt, Discovery-to-Conversion Gap, High-Value Routing Coverage, and AMNESS-AR are proposed constructs in this paper.
- The Total-Exposure Insufficiency Theorem is a structural measure-theoretic separation under the stated admissibility assumptions.
- The marginal routing equalization result is a KKT consequence for differentiable concave route yields under a binding budget constraint; the paper does not claim originality for KKT theory.
- Existing recommendation, exposure, bandit, propensity and fairness literature supports treating exposure allocation as an explicit decision problem.
- The paper does not assert that conversion maximization is the only valid platform objective.