GenPage: Towards End-to-End Generative Homepage Construction at Netflix¶

Fidelity: core mechanism. Paper results, public-data measurements, and omissions are separate.

Paper information¶

Field	Details
Paper	arXiv 2606.31031
Organization	Netflix
First public release	2026-06-30 (arXiv v1)
Authors' code	The paper did not provide official or author code when checked on 2026-08-24.
Adapter / method	`genpage`
Local reproduction code	`src/auto_research/reproductions/genpage/`

Original paper¶

Motivation and method¶

Generate the entire homepage with one model rather than cascading retrieval and ranking. After WBC/SFT, RL post-training uses long-term production rewards while handling business rules and cold start.

flowchart LR
 A["Homepage prompt"] --> B["Autoregressive whole-page generation"] --> C["WBC, RL, and rules"]

Key figure from the paper¶

Figure 6: Training flow and key optimization steps. From the original paper; copyright remains with the authors.

Core equations¶

\[ \pi^*=\arg\max_\pi\mathbb E_{Y\sim\pi(\cdot|h)}[R_{page}(Y)] \]

Paper results¶

The main online result is core engagement +0.24% in A/B (p<0.001). Offline tables and online metrics use private or paper-specific protocols. The private paper protocol is not directly comparable with local MovieLens measurements.

Local reproduction¶

Local comparison: The baseline is the shared transition + content scorer. The treatment adds only genpage on the same users, items, full-catalog candidates, and seed.

MovieLens-100K, 220 users and 360 items, seed 42: NDCG@10 0.0540 → 0.0519 (-3.98%); Hit@10 0.1091 → 0.1045. Validation selects mixture weights only; the test set is not used for tuning.

auto-research reproduce --paper genpage --dataset-dir data --seed 42

See the fixed metrics.

Reproduction boundaries¶

The core mechanism runs on MovieLens-100K with the same split, fixed full-catalog candidates, and seed. Private features, the production foundation model, and online traffic are unavailable. The implementation has independent model state and scoring, rather than mapping a paper name to a shared weighted formula. See the adapter's omitted_core_components for omissions. Local changes and paper A/B gains are reported separately.