How this works

I transcribed 859 episodes of the Kinda Funny Gamescast and In Review, about 1,312 hours of audio, with speaker diarisation. Scores, rankings and the sentences around them were then extracted automatically. Afterwards I checked the extracted scores against the audio and added or corrected them by hand where the extraction missed or misheard one. Most scores were found by a machine and checked by a person; some are still missing or wrong. If you find one, write to mat@valdbjorn.com, or send it to @pluttodk on Discord or Bluesky. If a host did not say a number, no number is shown for them. In total this gave 1640 titles, 583 scores and 92 franchise rankings. The quoted sentences are lightly cleaned for transcription errors, but the score itself is always shown exactly as it was said.

Kinda Funny used a 1–5 scale before May 2024 and 0–10 after. A score is always shown on the scale the critic actually used, so a “four out of five” is never turned into 8/10 next to someone's name. Scores are only normalised when they go into an average. The lowest score on the 1–5 scale is 2/10 while 0–10 goes all the way down to 0, so a lifetime mean across the change gets pulled a bit upward. Every critic page therefore also shows the two eras separately.

There is no metascore, no critic consensus and no filling of gaps. Only 291 of 1640 titles have a score at all. The film show ranks instead of rating. Those pages show a rank, not a score.

2 scores use an impossible scale (out of 4, 4.6 and 100). These are almost certainly transcription errors. They are shown as spoken, marked, and left out of every average and of the structured data:

• Andy Cortez on cult-of-the-lamb — 4.1/4.6
• Mike Howard on dying-light-2-2 — 2/4

27 scores could not be matched to a person, usually because they came from outside guests. They are shown under the name as it was said, without a link.

604 media entries are left out of this site. They came from ranked lists of things that are not really works, like villains, and in one episode a ranking of walkathon dinners. “Chili's” and “Licker” are in the database but not published here.

48 television reviews are of a single episode or season rather than a whole series, for example seven Mandalorian chapters and six episodes of The Last of Us. Each was reviewed on its own, with its own page in the TV listing. The series page gathers them under “reviewed episode by episode”, and a series average does not include them.

62 titles still share an entry with another. These are the cases a rule cannot decide safely: the Game Boy Advance and the Nintendo Switch “Mario vs. Donkey Kong” are two different games with the same name, and one ranked-list entry is an actor rather than a film.

12 reviewed titles have no cover. Some of them are hardware or live events rather than works, like a Switch OLED and a Razer chair. There is no artwork to find for them. Those pages show a lettered tile instead.

Episode transcripts are machine-generated and contain errors, especially in titles and names. Speaker attribution comes from voiceprints where a host is enrolled, and otherwise from the context of the conversation. Where the system could not tell who was talking, the line says “Unidentified speaker” instead of guessing.