ANIME: Source Sets the Audience Floor

Death Note has 1,610,561 MyAnimeList members, more than any other anime in the archive, and scores 8.66. Hanada Shounen-shi, a 2002 series from the same studio, Madhouse, scores 8.11 and has 8,929. A MyAnimeList archive of 13,631 titles, from 1917 to an early-2019 snapshot, suggests that much of a gap like that is set before anyone draws a frame, starting with where the story comes from.

On television, light-novel adaptations draw a median of 153,184 platform members. Manga adaptations draw 37,500 and original series 4,738, which puts light novels about 32 times ahead of originals. Quality follows a different pattern. Only 4.6% of original TV series and 0.5% of TV game adaptations score 8.0 or higher as of the early-2019 snapshot, compared with 17.0% of manga adaptations. Across all scored titles, a title's score and its audience move closely together, with a rank correlation of 0.74. Every comparison below is measured against the overall median score of 6.38.

This report reads the archive as the record of a production system, not a list of favorites. In that system, a production committee (seisaku iinkai) of publishers, broadcasters, music labels, and merchandisers pools the budget and splits copyright in proportion to each member's investment. The animation studio is typically a contracted manufacturer that doesn't own the IP (Hernández Hernández, 2018, §4.3; Mihara, 2018). The figures here are observed fan behavior on one platform. They show which projects found audiences, not which ones made money.

RESEARCH QUESTION

Which decisions made before production (format, source material, studio, and whether to make a sequel) are associated with a title's reach and fan rating, and where does the platform's popularity record miss strong work?

Each format wave lines up with a new distribution channel

Anime output is not one growth curve. It's a series of format waves, and each wave lines up with a new channel. Theatrical film dominates the early catalog. Original video animation (OVA), sold directly on tape and disc, ran at roughly 100 titles a year from 1990 to 1993 and peaked at 115 in 1993. Television then became the core of the catalog, with 4,260 TV titles in the archive. Original net animation (ONA), made for web distribution first, grew from 40 titles in 2009 to 104 in 2013 and 216 in 2018, for 1,321 in total.

The archive records release format, not financing. Reading these waves as responses to new distribution channels is our editorial frame, not a measured finding, and the data can't show who paid for each wave. The context literature documents one wave directly: starting with Tetsuwan Atom (1963), TV anime was tied to sponsor merchandising (Steinberg, 2012).

Source material sets the audience floor

The source of a story, its gensaku, is the biggest single dividing line in the TV catalog. Light-novel adaptations have the highest median score (7.35) and by far the largest median audience. They reach about four times as many members as manga adaptations and about 32 times as many as originals. That gap measures audience size, not quality or staying power. It also compares typical shows, not the best ones: 62 of 1,012 original TV series (6.1%) draw more members than the median light-novel adaptation, while only 3 of 283 light-novel series fall below the median original. Manga adaptations are close behind on score (7.29), and they produce the highest share of top-rated shows, at 17.0% scoring 8.0 or higher (210 of 1,232).

The mechanism is the media mix. A story is built across print, animation, music, and merchandise at once (Steinberg, 2012). On our reading, a published book also serves as a live test of demand before a committee commits money to animation. This is selection, not proof that adapting a book lowers risk. Committees adapt titles that already have readers, so some of that audience is arriving with the book.

Originals face the opposite problem. Their median score is 6.63, and only 4.6% reach 8.0 or higher. With no print run to test demand first, an original has to prove itself on air. Game adaptations are the weakest group by fan rating. Their median is 6.63, and as of the early-2019 snapshot only 0.5% of TV game adaptations score 8.0 or higher (1 of 210). For a committee, one practical reading is that a game adaptation is better budgeted as part of the game's marketing than as the start of a stand-alone animated franchise. That reading is an editorial frame, not a result the file measures.

Consistency is a studio trait

In a system where the committee owns the project, the animation studio is usually the prime contractor (motouke), and much of the drawing is subcontracted (shitauke) to other studios and freelancers (Hanzawa, 2004). Inside a studio, production managers (seisaku shinkou) coordinate the work but depend on key creatives, whose drawing skill routinely gives them the upper hand (Morisawa, 2015). That makes a studio's typical output a better guide to what a commission will deliver than its single biggest hit.

Bones has 115 titles, a median of 7.46, an interquartile range (IQR, the spread of the middle half of its scores) of 0.95, and 23.5% of its titles scoring 8.0 or higher, the tightest spread among the top-scoring studios (only 7th-tightest of all 20). Kyoto Animation's 110 titles have a median of 7.43 and an IQR of 1.22. Kyoto Animation employs mostly salaried in-house staff and trains animators through its own school, which is unusual in an industry built on subcontracting (Creamer, 2015; Moore, 2017; Kyoto Animation training school). OLM is the counterexample that makes the point. Its 209 titles have a median of only 6.64, but its IQR of 0.70 is the tightest of any studio with 100 or more titles. It delivers a predictable product at a modest level.

For anyone choosing a prime contractor, the decision metric is a studio's median and spread, not its best-known title. Across the 20 studios with 100 or more scored titles, median scores run from 7.46 (Bones) down to 5.40 (DLE). DLE (n = 153) mostly makes short-form titles, with a median episode of 3 minutes, so its median score isn't directly comparable to studios making 24-minute episodes. Bones and OLM have a 24-minute median episode. The largest studio, Toei (737 titles), sits near the bottom of the table at 6.68. Of those titles, 2.8% score 8.0 or higher (21 of 737).

Each card links its numbers to its history. The links are our reading, not measured effects. The section takes its numbers from the league table above and adds no new chart.

Sequels score higher, consistent with mostly well-received shows getting one

Of the 4,240 scored TV titles, 867 are sequels. We count a title as a sequel when its chain of prequels, every one of which aired earlier, reaches an earlier TV series. That rule keeps promotional specials from making a first season look like a sequel. Sequels have a median score of 6.99, against 6.76 for the other 3,373 titles, and 14.8% of them score 8.0 or higher (128 of 867), against 6.5% (218 of 3,373). The median gap ranges from 0.15 to 0.30 points depending on how "sequel" is defined, but every definition we tested points the same way.

The pattern is consistent with selection, not with any effect of being a sequel. Committees mostly order second seasons of shows that did well, and the viewers who come back already liked the first one. Among the top TV sequels by audience are Sword Art Online II, Naruto: Shippuuden, Code Geass R2, and Clannad: After Story (9.00). For a buyer, a sequel's score says more about the strength of the property than about the sequel itself.

Popularity tracks quality, with a short list of exceptions

Member counts are extremely skewed. A rank (Spearman) correlation gives 0.74, and correlating score with the log of member count gives 0.72. On this platform, well-rated titles are usually well watched.

The exceptions are rare and specific. Of 603 titles scoring 8.0 or higher, 78 sit beyond popularity rank 3,000. Requiring at least 1,000 user ratings cuts that to 43. Of the 43 titles with at least 1,000 ratings, 12 continue an earlier TV series (5 of them as TV seasons); we count sequels by following each title's prequel chain back to an earlier TV series. Limiting the list to TV series and films with no earlier prequel, parent story, or full story of any type leaves 11.

The Buried Canon is an editorial shortlist, not a measured ranking: score of 8.0 or higher, popularity rank past 3,000, 1,000 or more ratings, TV or film, and no earlier prequel, parent story, or full story of any type. Related titles with missing start dates do not count.

LIMITATIONS

Coverage window. The data is a Kaggle snapshot of MyAnimeList taken in early 2019 and published through TidyTuesday on April 23, 2019. Nothing released after that is included, and 2019 itself is only partly covered, which is why the format series ends at 2018.

Platform audience. MyAnimeList is an English-language enthusiast site. Its scores reflect that audience's expectations. Kids titles (median 5.86, n = 2,158) and music titles (median 5.44, n = 1,510) score lowest, which more likely reflects a mismatch between audience and content than weak production. Thriller (7.50, n = 102), Mystery (7.27, n = 605), and Psychological (7.23, n = 279) score highest.

No commercial data. The file has no revenue, broadcast ratings, disc sales, box office, or merchandise figures. A low score doesn't mean a franchise lost money.

Selection, not causation. Source material, studio, and sequel status are all chosen with an eye to expected demand, so none of the gaps above should be read as the effect of that choice alone. The format chart lines each wave up with a channel. It does not show that the channel shaped what got made.

Deduplication. The source file repeats each title once per studio and genre (77,911 rows for 13,631 titles). All title-level figures here use one row per title.

Context after the snapshot. Events after early 2019, including the 2019 Kyoto Animation arson, later Netflix commissions, and current streaming availability, come from outside sources. The data can't reflect them.

CONCLUSION: Choose the source, judge the studio by its median, and read sequel scores as property scores

The archive doesn't show what makes anime good. It does show where the industry's early decisions leave a mark. Adapting a light novel or manga brings an audience that is already there, while an original has to find one. Studios differ clearly in their typical output, which is the better guide to a commission. Sequel scores mostly reflect the shows that earned them. Beyond those patterns lies a short list of strong titles the platform's audience never found. The next useful question is commercial. Matching these fan signals to revenue data would show whether the platform's audience floor is also a revenue floor.

EDITOR'S NOTE