Back to Blog

The Number That Is Not the Film: How Three Rating Systems Turned One Movie Into Three Scores

Niu Lai carries three very different numbers — Douban 6.0, IMDb 6.5 weighted, and IMDb 8.2 raw. The gap between them says more about how each site counts than about what audiences watched.

November 10, 20269 min read#["ratings"#"douban"#"imdb"#"analysis"#"data"]

A Chinese animated film with 86 minutes of runtime, a deliberately rough look and no advertising budget has three scores attached to it, and the three do not agree. Douban opened at 5.9 on August 19, 2026 and later settled at 6.0, with the distribution curving into a shape the Chinese press quickly named after the film's own title: 35.8 percent of raters gave five stars, 37.8 percent gave one star, and the middle bands between them accounted for 26.4 percent. On IMDb the same title carries a weighted score of 6.5 while the unweighted average of the visible votes sits at 8.2. Two and seven tenths of a point separates the number a reader sees from the number the site actually computes, and the gap exists because of how the site counts rather than anything anyone watched.

There is a third column of numbers, and it is the one the industry reads. When Traces of Justice reached its fifth week in Chinese cinemas in November 2026, its Douban score had opened at 6.8 and eased to 6.7, while the ticket buyer ratings on Maoyan and Taopiaopiao were 9.6 and 9.5, with Lighthouse also showing 9.5. The film had drawn close to 18.8 million admissions. A title with a low critic facing score and near perfect audience facing scores is not a paradox, and understanding why requires looking at what each number is built out of.

What the mean hides

Douban's 5.9 came from a distribution that almost no commercial film produces. The press description was precise about it: a bull horn shape, with the two ends loaded and the middle thin. In comparable cases where a film became a curiosity through online mockery, the scores had gone the other way. Mermaid Summer, which had drawn crowds for the same reason, collected a low score overall, around 4.0, rather than an even split between love and hate. Creation of the Gods II, sitting at a similar 5.9, had a heavy middle. Pure Hearts: Into Chinese Showbiz holds the lowest score record on Douban, with 99 percent of raters giving it one star and the average falling to 2.0 out of 10. Niu Lai was the outlier, and the reason sits inside the five star column.

Commentators who examined the ratings reported that most of the five star votes were doing something other than praising the film. The phrases circulating were jokes in their own right, calling it the ceiling of Chinese animation, comparing it to large scale industrial production, and using inflated approval to deflate the seriousness of a rating system. Readers noted that the score was losing the ability to describe the work and had become an outlet for personal feeling. One professional writer on the platform put it plainly, saying the story was better than most other stories in that summer season. Others said that a film like this becoming successful would be a national disgrace. The one star column, meanwhile, was almost entirely about craft: models that looked like early computer games, stiff movement, visible clipping, a distance from what a cinema screen should carry.

Both columns were describing the same object accurately. The movie really was rough and the story really did hold together. The bull horn shape records a room where people disagreed with each other, and a single number thrown across that room lands on whoever catches it. The score moved from 5.9 to 6.0 in late August, and the reporting on that change was consistent about the cause. The audience had widened from core film people to walk in for the social experience, and the evaluation criteria followed them. The shift of a tenth of a point was a change in who was voting, not in what they had seen.

What weighting does

IMDb is explicit about its method in its own help documentation. Registered users rate a title from one to ten, and the individual ratings are aggregated into the single figure shown on a title page. The arithmetic mean is calculated and displayed on the breakdown page, but the score on the main page is a weighted average, because not every rating carries the same impact on the final number. When unusual rating activity is detected, a different weighting calculation may be applied to preserve the system. The company states plainly that it does not disclose the exact method used to generate the rating, and that this is deliberate, so the mechanism cannot be reverse engineered for abuse. The wider chart runs on a separate published calculation that blends a title's own average with the average across eligible titles, weighted by how many votes it has received, which pulls a title with few votes toward the site wide figure and lets only large and genuine samples climb on their own strength. Only votes from regular voters count toward that chart, and the definition of a regular voter is not public, which is why a title's chart figure can differ slightly from the figure on its own page.

That structure explains the 8.2 reading for Niu Lai. A small sample of enthusiastic raters, many of whom were there for the joke, produces a high raw average, and a weighting scheme designed to damp small and coordinated samples pulls the reported number down toward the middle. Whether a reader lands on 8.2 or 6.5 tells them more about the size and shape of the voting sample than about the film. It is worth being careful here, because the gap is routinely reported as a difference between Chinese and foreign audiences, when the more accurate reading is a difference between two counting methods applied to crowds of very different sizes and composition.

The third column

The industry number is neither a critic score nor a viewer score. It is the pre release interest count, and its influence runs through the most commercially consequential part of a film's life. Cinema managers in China consult the platform counts when they decide how many screens to give a title on its first day, and the counts are maintained by the ticketing platforms themselves.

Trade analysis published in 2024 laid out the limits. The number is useful for a film's opening, because a large count produces a high position in the apps' recommendation areas and reaches potential buyers directly. It is a poor guide to a film's final gross. The most watched film of that year, Successor with 80.4 million admissions, had 368,000 wants on Maoyan and 224,000 on Taopiaopiao, a combined figure that is a small fraction of the audience it eventually reached. An Inhabited Island, with 1.425 million combined wants across the two platforms, still finished with 20.345 million admissions, which is to say the wants were a rounding error against the result. A film that opened at 640 million yuan with 15 million admissions had fewer than 100,000 wants on either platform, because a good word of mouth does the work that an advance click cannot. What the count predicts well is the first weekend, and what it predicts badly is the shape of the curve after that.

The counting itself has an opacity problem that matters more than the predictive limits. In the 2026 summer season, the same title carried a wants figure of 340,000 on Maoyan Professional and 265,300 on Lighthouse Professional, a difference of more than 70,000, and the two platforms define the metric differently in terms of de-duplication and time window without disclosing either definition. One of the platforms was also an investor in the film in question. A metric that is a counter on someone's own server, defined by that server's owner, and used to set the industry expectation for screen counts, does not have the same standing as a ticket sales figure, which has a ticket behind it. Trade commentary has made this point directly: box office is counted from stub records while screenings are set by a definition, and the two books are not the same book.

What the three columns add up to

Put the three systems beside each other and the picture gets clearer rather than muddier. Douban produced a bull horn because a film built on a joke invites a crowd that argues. The reported figure moved as the crowd changed shape. IMDb produced two numbers because it publishes both the raw mean and a weighted score, and keeps the weighting private to protect it from manipulation. The ticket platforms produce a pre release count that shapes the opening and then stops predicting, and they differ from each other by tens of thousands on the same title for reasons that are not published.

For Niu Lai specifically, the split is unusually clean. The 4 million plus raters on Douban and the much smaller group on IMDb were looking at the same 86 minutes, and the differences in their numbers come from sample size, weighting and voting rules rather than from what the film is. Meanwhile the first film's own success remains a mechanical fact: 7,169 yuan across the first nine days, a low of 921 yuan on August 13, four screens at its worst, then about 6.09 million yuan on August 16 once the mockery went mainstream, and a final tally near 64.3 million yuan after three key extensions.

None of this settles the question the ratings were asked to settle, which is whether the film deserves the number. It does show why that question keeps producing different answers, and why a rating tells you less about a film than about the rules under which its raters were counted. The bull horn on Douban, the private weighting on IMDb and the platform disagreement over wants all describe crowds and counting methods. The film is 86 minutes long and has not changed since August.

--- This article is part of the Niu Lai fan site. We are not affiliated with the production company.