Bayesian weighted rating engine

The score behind
the score.

A raw IMDb number lies. A cult thriller with 40k votes and a 7.9 isn't really beating a blockbuster with 900k votes and a 7.8. True Rating runs both through the same Bayesian weighting so you compare like with like - then puts them in the ring.

Shared parameters

These apply to both films. Lower the threshold to trust niche ratings more; raise it to favor proven hits.

25,000

How many votes a film needs before its own rating is trusted over the baseline.

6.8

The mean score every film is pulled toward when it has few votes.

Contender AWinner

True Rating

7.27

vs raw 7.3

-0.03

How much its own rating counts94%

The remaining 6% is pulled toward the 6.8 baseline.

Contender B

True Rating

7.24

vs raw 7.4

-0.16

How much its own rating counts74%

The remaining 26% is pulled toward the 6.8 baseline.

The true winner

Hereditary

Takes it by 0.03 points once both scores are weighted by their vote counts.

7.27
7.24
Why this works

One thriller, ten thousand votes

The weighted rating

WR = ( v / (v + m) ) · R + ( m / (v + m) ) · C

R

The movie's own IMDb rating (1 - 10).

v

How many people actually voted on it.

m

The trust threshold - currently 25,000 votes.

C

The baseline every film leans toward - now 6.8.

01

Few votes? Stay humble

A film with a handful of votes barely outweighs the baseline C, so a fluke 9.5 from 200 fans gets dragged back toward the average. No brigading to the top.

02

Many votes? Earn your score

As votes climb past the threshold m, the film's own rating R takes over almost entirely. A proven hit is trusted to be exactly what the crowd says it is.

03

Now they compare fairly

Both films are measured against the same baseline and threshold, so a niche cult favorite and a mainstream blockbuster finally sit on the same scale.

The trick is the balance between v / (v + m) and m / (v + m) - they always add up to 1. Low-vote films borrow confidence from the crowd average; high-vote films keep their own. That's the whole reason a 7.4 from 71k votes can quietly beat a 7.3 from 380k, or lose to it, depending on how much proof you demand.