How ratings work
hüt abig aggregates ratings from IMDb, Rotten Tomatoes, TMDB, Metacritic, and TVmaze into one transparent 0–100 index, then ranks films and series available on Swiss streaming. We do not invent scores. Ratings measure opinions from a particular population under a particular aggregation rule — they are not interchangeable observations of an objective “true quality”.
Sources in the blend
| Source | Default weight | What it measures |
|---|---|---|
| IMDb | 40 | Large audience vote. Bayesian-shrunk inside Overall toward 7.0 with a 25,000-vote prior; the IMDb column stays raw. |
| TMDB | 15 | Community catalogue rating. Bayesian-shrunk inside Overall toward 7.0 with a 750-vote prior; the TMDB column stays raw. |
| Rotten Tomatoes critics | 15 | Tomatometer critic positivity, normalised to 0–100. |
| Rotten Tomatoes audience | 5 | Audience / Popcorn score, normalised to 0–100. |
| Metacritic critics | 20 | Weighted critic metascore. |
| Metacritic users | 5 | Metacritic user score. |
| TVmaze | 0 | Series community rating. Visible as a column; off in the default blend because coverage is thinner. |
| Awards (Oscars / Emmys) | 10 | Wins count 3× nominations, log-scaled to 0–100 within each medium. |
| Vote confidence | 0 | Sortable log-vote-volume column. Optional in the blend. |
| My taste | 0 | Personal favourites affinity. Joins only when favourites are active. |
These are the Consensus (default) weights. Presets Critics, Crowd, Equal, and Personalise retune the mix. Missing sources are omitted from the average, never treated as zero. Coverage for Rotten Tomatoes, Metacritic, and TVmaze is incomplete; we do not fill gaps with estimates, and we do not currently include Letterboxd. Scores are collected snapshots, not live page scrapes.
See the rating aggregator landing for the same sources in search-oriented language.
Kinds of signal
| Kind of signal | What it measures | Limitation |
|---|---|---|
| Audience ratings | User votes on a numeric scale, often with platform weighting. | Self-selection and popularity affect representation and precision. |
| Critic scores | Aggregated critic positivity or weighted critic averages. | Positivity is not always the same as intensity of praise; source panels differ. |
| Community ratings | Votes from catalogue and fan communities. | Small samples are volatile; coverage varies by title type. |
| Vote volume | How many people contributed to a rating. | Volume is not quality; it only speaks to sample size. |
hüt abig model
Adjustable Overall ranking
The main list’s Overall score normalises available rating sources to 0–100, then blends the sources that exist for that title with the weights above. Missing sources are omitted from the average, never treated as zero. Vote confidence and personal favourites relevance stay out of the default blend; both can be added with the weight controls. Awards use Oscar records for films and Emmy records for series (wins count 3× nominations, log-scaled to 0–100).
Catalogue eligibility
Ranked lists show only titles with an IMDb audience rating. The curated worldwide catalogue and each production-origin collection target 1,000 IMDb-rated titles where data allows.
Consensus quality
The detail view’s consensus is a separate, non-adjustable evidence score. Two primary rating sources (IMDb and TMDB) are each shrunk toward a prior mean with Bayesian vote priors, then combined 65% / 35% and mapped to 0–100 with quality = clamp((consensus − 6.5) × 40). A consensus rating of 6.5 maps to 0 and 9.0 maps to 100. It does not include critic aggregators, awards, relevance, freshness, or personal taste.
Bayesian shrinkage
adjusted = v/(v+m) × R + m/(v+m) × C. Small samples move toward a prior mean; large samples retain more of their observed rating.
Colour scale
Score colours run from red below 6, through orange at 6, yellow at 6.5, green at 7, and cyan at 8, with blue reserved for 10. Values between those anchors are interpolated continuously so high ratings stay visually distinguishable. Missing values are neutral.
Awards and availability
Major awards records represent institutional recognition, not direct audience quality. When enabled, the awards weight adds an Oscar/Emmy signal (wins ×3 + nominations, log-scaled within each medium). The ranking itself is worldwide quality among the curated catalogue; Switzerland refers to streaming availability on Swiss providers, which can lag catalogue changes.
Pick for me
The Pick flow asks a few taste questions, then ranks shortlists with a deterministic on-device matcher. When available, AWS Bedrock (Amazon Nova Lite) may re-rank the shortlist from structured catalogue metadata; if the model is unavailable, the client fallback keeps serving picks.
Trending
The Trending list ranks current popularity within the active filters and returns the leading 100 titles. Strict title matches from a Switzerland search-interest snapshot can add a boost when available; an empty match set does not make the list empty.
Age classifications
hüt abig preserves official source codes and adds plain-language guidance in title details. Examples: PG-13 warns that some material may be inappropriate below 13; an R-rated film requires an accompanying parent or adult guardian for viewers under 17; TV-14 indicates material many parents may find unsuitable below 14; TV-MA is intended for mature audiences and may be unsuitable below 17. Numeric classifications such as 12, 16, or 18 are displayed as explicit minimum-age guidance. Local provider guidance remains authoritative for a particular Swiss release.
Limitations
- Missing ratings disproportionately affect niche, international, older, and non-English titles.
- Critic and audience disagreement can be meaningful rather than error.
- Ratings drift and can be affected by publicity, fandom, and social influence.
- A single rank necessarily hides uncertainty and disagreement.