# Off-Policy Evaluation for Semantic ID Recommenders: Does the Model's Own Code Hierarchy Help?

arXiv: https://arxiv.org/abs/2608.28905 · HTML: https://mlsanity.com/paper/2608.28905/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 9 of its 17 references that are in the map.
- First posted on arXiv 28 Aug 2026.

## Builds on (most cited first, 9 of 9)

- [KuaiRand: An Unbiased Sequential Recommendation Dataset with Randomly Exposed Videos](https://mlsanity.com/paper/2208.08696/) (2022)
- [Off-Policy Evaluation for Large Action Spaces via Embeddings](https://mlsanity.com/paper/2202.06317/) (2022)
- [Adaptive Estimator Selection for Off-Policy Evaluation](https://mlsanity.com/paper/2002.07729/) (2020)
- [Off-Policy Evaluation for Large Action Spaces via Conjunct Effect Modeling](https://mlsanity.com/paper/2305.08062/) (2023)
- [Policy-Adaptive Estimator Selection for Off-Policy Evaluation](https://mlsanity.com/paper/2211.13904/) (2022)
- [POTEC: Off-Policy Learning for Large Action Spaces via Two-Stage Policy Decomposition](https://mlsanity.com/paper/2402.06151/) (2024)
- [Marginal Density Ratio for Off-Policy Evaluation in Contextual Bandits](https://mlsanity.com/paper/2312.01457/) (2023)
- [Learning Action Embeddings for Off-Policy Evaluation](https://mlsanity.com/paper/2305.03954/) (2023)
- [Automated Off-Policy Estimator Selection via Supervised Learning](https://mlsanity.com/paper/2406.18022/) (2024)

## Built on by (most cited first, 1 of 1)

- [Training with Missed Targets in Generative Recommendation: Separating Supervision from Probability Competition](https://mlsanity.com/paper/2610.10124/) (2026)

## Abstract (arXiv)

Generative recommenders increasingly emit semantic IDs (SIDs): each item is a short sequence of hierarchical discrete codes from a residual quantizer, decoded autoregressively. Before spending scarce A/B-test, a team may decide offline which decoder or reranking variants are worth testing - a job for off-policy evaluation (OPE). We ask a simple question: can the model's own SID tree serve as the action abstraction for that OPE? Our answer has three parts. (i) Under the near-argmax logging real recommenders use, per-item OPE is hopeless - as item-level effective sample size is usually small on production logs - but marginalizing items to code-prefix clusters restores estimable support and cuts error. (ii) This gain is thanks to coarsening, not to the hierarchy specifically; but the SID tree is what makes coarsening feasible in a generative system - each cluster's mass is exactly and cheaply returned by the decoder, whereas flat clustering requires enumerating item/leaf masses that a code-only decoder does not directly expose. (iii) Resolution depth is the operative knob - coarser under scarce support - and a conditional bias bound links the coarsening bias to the quantizer's worst-case reconstruction residual and the target-logging divergence.