01
Watch history
Categories and creators you have actually watched. The longest-running, highest-weight signal — what you watch is the strongest predictor of what you want next.
How the
algorithm
actually works.
Published in compliance with the EU Digital Services Act, but useful regardless of jurisdiction — five inputs, three forbidden inputs, three controls, no dark patterns.
§ One
Five
signals.
Every recommendation rank you see is a weighted combination of these five inputs. Heavy > Medium > Light, but the weights are not static — we re-tune them periodically and document changes.
02
Engagement
Likes, saves, watch-to-completion, and re-watches. Engagement is a stronger signal than impressions — finishing a clip outweighs starting ten of them.
03
Recency
Newer content is weighted higher so the catalogue stays fresh. Older content can still rank — engagement signal can override recency penalty.
04
Popularity
View counts and engagement rates within your preferred categories. We use popularity inside your taste cluster — not site-wide popularity that would homogenise everyone.
05
Creator diversity
We deliberately surface content from a range of creators, not just the most popular handful. New creators get exposure to break the cold-start problem.
§ Two
What we don't
feed it.
Every signal we exclude is a deliberate choice, not an oversight. The catalogue of inputs we don't use is as much a part of the system as the ones we do.
No external profiling
We do not use personal data from outside your on-platform activity. No ad-network data, no third-party tracking, no cross-site behavioural sharing.
No sensitive-data profiling
No profiling based on ethnicity, political opinions, religion, health, or other sensitive categories. Recommendations are about content, not who you are.
No engagement maximisation
We do not optimise for time-on-site at the expense of viewer wellbeing. No infinite-scroll dark patterns, no escalating-intensity rabbit holes.
§ Three
Your
controls.
Three explicit ways to override or bypass the algorithm. The recommender exists to help; it's never the only path through the catalogue.
⌖
Hide extreme content
A profile-settings toggle that filters out specific category clusters from your recommendations entirely.
☷
Browse by category
Use the category index directly to bypass the algorithm — chronological, popularity-sorted, or filtered by tag.
🔍
Direct search
Search by keyword, creator, performer, or tag. Search results are not personalised; what you ask for is what you get.
§ Four — Transparency
Questions about how it ranks your feed?
Customer support routes recommender questions to the people who actually maintain the system. Material changes to the algorithm are documented in the changelog.