Why WatchLens
Several platforms already host online recommendation studies. WatchLens is purpose-built for video recommendation research where the UI, the policy, and the event measurement need to be varied independently. This page situates the design against the alternatives.
What other platforms optimise for
| Platform | Optimised for |
|---|---|
| Informfully | News articles. UI is a hard-coded React Native client, and recommendations are pushed into the platform's database by an external service. |
| RecBole | Offline benchmarking. 70+ algorithms, but no UI, no event capture, no live participants. |
| One-off platforms authored per study | Whatever the study needs. Reusability and cross-study comparability are not goals. |
These tools are excellent at what they target. They are not designed for a study where the question is "does this UI change participant behaviour under this policy?". In each one, either the UI is fixed, or the runtime is fixed, or the event schema is fixed.
Where WatchLens fits
WatchLens is for studies where you want to vary one axis at a time without rewriting the rest of the stack.
- The policy is a plug-in (Python class or external HTTP). A baseline swap is one config change. A custom policy is one file.
- The UI is a plug-in (bundled preset or in-browser TSX or visual block tree). All three emit the same event schema.
- The event schema is fixed (33 events across 6 categories). A study using policy A on UI X compares cleanly against a study using policy B on UI Y because the data shapes are identical.
When WatchLens is the wrong choice
- You only need offline benchmarking. Use RecBole directly.
- You need a fully customised UI that does not fit the feed-page-and-watch-page abstraction. Write your own platform, because the abstraction will get in your way.
- Your participants need a native mobile app. WatchLens ships a responsive web app with three device-tagged variants. There is no iOS or Android client.
Where to go next
- Architecture Overview. Read about the hybrid registry pattern that holds the plug-in axes together.
- Configurable Execution. See the recommender and UI plug-in axes in detail.
- Standardized Measurement. Read what the 33-event schema captures and why.