How we score - and why we ignore downloads.

For each show we sample recent episodes, transcribe them, and have a large language model score the transcript against a fixed five-dimension rubric. Each dimension is scored 0-20 and the five sum to a 0-100 episode score. A show's substance score is the average of its five most recently scored episodes. We keep a history of every period so ranks can rise and fall over time.

The model must return a score and cited evidence - verbatim quotes from the transcript - for every dimension. We never accept a score without its evidence.

1. 01

   Insight Density

   0-20

   Novel, non-obvious claims per minute vs. filler.

   Rewards: Ideas a smart operator hadn't already heard, packed tightly.

   Penalises: Padding, throat-clearing, and obvious advice.

2. 02

   Originality

   0-20

   Fresh thinking vs. recycled takes.

   Rewards: Contrarian, first-principles, or counterintuitive arguments.

   Penalises: The same frameworks and quotes that circulate everywhere.

3. 03

   Guest Caliber

   0-20

   Operators and practitioners over career guests.

   Rewards: People who have actually done the thing at scale.

   Penalises: Career podcast guests and thinly-relevant names.

4. 04

   Specificity & Evidence

   0-20

   Named examples and real numbers vs. vagueness.

   Rewards: Specific companies, metrics, timelines, and dollar figures.

   Penalises: Hand-waving and abstraction.

5. 05

   Conversational Craft

   0-20

   Sharp questions and real follow-ups vs. softball chats.

   Rewards: Genuine follow-ups and productive disagreement.

   Penalises: Softball PR chats and unchallenged claims.

• We pull a show's recent episodes from its public RSS feed and use the most recent five scored episodes for the substance score, so the ranking reflects a show's current form, not its back catalogue.
• Transcripts come from the publisher's own transcript when one is published in the feed, otherwise from machine transcription of the audio. We never republish full transcripts - quotes are used only as short cited evidence.
• Trailers, ads-only entries, and sub-five-minute clips are excluded.

• We don't use downloads, so buying ads, swapping promos, or inflating subscriber counts does nothing to your rank.
• Scores are evidence-bound: the model must quote the transcript, so a keyword-stuffed description or a slick intro can't move the number - only the substance of the conversation can.
• We average multiple episodes, so a single unusually good (or sponsored) episode can't carry a show.
• The rubric and model version are recorded with every score, and we calibrate the model against human-rated episodes before trusting its output.

We only ever publish a positive, ranked list - we never publish a "worst" list, and scores are framed as relative quality, never as attacks. If you host or produce a show and believe a score is wrong, want a fresh episode sampled, or would like your show removed from the Index, email hello@fame.so and we'll respond.