The answer pool is not only anime. 800 titles and 2,310 characters live in the same space, and both are valid guesses.
Most players still type only titles. The data says that costs you.
1. The short version
Once a candidate title comes to mind, type one of its characters instead of the title. Of the 794 anime that have characters in the pool, 780 (98%) score higher against their own cast than against their closest other anime.
2. The measurement
For each title we compared two numbers.
- Best own character — the highest similarity to any character from that show
- Nearest other title — the highest similarity to any other anime, with the show's own franchise excluded
| Measure | Median |
|---|---|
| Best own character | 0.907 |
| Nearest other title | 0.737 |
| Gap | 0.159 |
A median gap of 0.159. In this game that is not a small number — it is roughly the width between "somewhere in the right genre" and "inside the answer's own cluster."
The gap exceeds 0.2 for 253 titles (32%) and 0.3 for 42. Cases where another title wins instead number just 14.
3. Where the gap is widest
On days when one of these is the answer, guessing titles gets you almost nowhere.
| Answer | Character | Other title | Gap |
|---|---|---|---|
| Diamond no Ace: Act II Second Season | 0.956 | 0.491 | +0.465 |
| Yuri!!! on Ice (special) | 0.950 | 0.509 | +0.441 |
| Fullmetal Alchemist | 0.969 | 0.569 | +0.400 |
| Violet Evergarden: Recollections | 0.945 | 0.550 | +0.395 |
| Violet Evergarden | 0.968 | 0.579 | +0.389 |
| Kingdom 5th Season | 0.925 | 0.560 | +0.365 |
| Dr. Stone | 0.973 | 0.608 | +0.365 |
| Vinland Saga | 0.956 | 0.592 | +0.364 |
Take "Fullmetal Alchemist." Exclude its franchise and the closest anime is "Mononoke Hime" at 0.569. Type the character "Hawkeye, Riza" and you get 0.969. One character clears a wall that no title can.
That this list overlaps the most isolated titles is no accident: the more isolated a title is, the more its characters are the only way in.
4. What the 14 exceptions look like
Open up the 14 cases where another title wins and most are the same story in a different form.
| Title | Character | Other title | That other title |
|---|---|---|---|
| The First Slam Dunk | 0.679 | 0.858 | Slam Dunk |
| Slam Dunk | 0.720 | 0.858 | The First Slam Dunk |
| Serial Experiments Lain | 0.617 | 0.724 | Paprika |
| Seihantai na Kimi to Boku 2nd Season | 0.837 | 0.903 | Boku no Kokoro no Yabai Yatsu 2nd Season |
| Boku no Kokoro no Yabai Yatsu 2nd Season | 0.853 | 0.903 | Seihantai na Kimi to Boku 2nd Season |
Either a remake-and-original pair (Slam Dunk) or twin shows that are each other's nearest neighbor. In other words, "title beats character" only holds when a near-duplicate of the show exists in the pool.
5. Why it works this way
Because similarity is computed from synopsis text. A character's description usually opens by restating the plot of their show, which puts the character almost on top of the show itself.
A different anime, however similar, still has entirely different names, settings and proper nouns. That difference is the 0.16 wall.
6. Practical order of play
- Cast wide to find the cluster. Titles are the convenient tool here. Feed in a few genre landmarks and find the direction that returns 0.6-something.
- Once a candidate appears, type its lead character. Above 0.9 means the answer is either that show or another character from it.
- Cleared 0.9 but still wrong? Move to another character from the same show. On days when the answer is a character, cast members routinely sit at 0.99.
- Stuck in the 0.6s? Your candidate is wrong. Stop adding characters and go back to cluster hunting.
7. Limits
Each show contributes at most 3 characters to the pool, by popularity. Supporting cast are mostly absent, so a name failing to appear in autocomplete does not rule that show out.
794 of the 800 titles have characters in the pool; the other 6 had no cast data in the source and were excluded from this comparison. The scoring method is described in About the Game.
Further reading
Which Genres Are Hardest measures genre difficulty from the same data, and Ratings Are Not a Clue covers why narrowing candidates by score fails. The main strategy guide has the similarity band reference table, and each day's answer and its nearest neighbors accumulate in the past answers archive.