Jenn’s Reading Room · Method
How the tracker works.
The Checks and Balance tracker is only as trustworthy as the steps behind it. Here is where the numbers come from, how I decide who won a round, and what I still don’t trust.
I’d rather show the seams than hide them. A chart on these pages is only worth putting up if I can say exactly how the number under it was made.
Only one show, Checks and Balance, is coded round by round. The rest of the family is catalogued, not scored. So everything below is really about how one weekly quiz becomes a row I’m willing to stand behind.
Sources
Where the numbers come from.
Episode titles, dates, and descriptions come from each show’s public RSS feed: the same feed any podcast app reads. The quiz itself isn’t in that text, so for Checks and Balance I take it from the closing minutes of the episode audio.
The only sign-in involved is my own podcast subscription, used to reach the audio. Nothing here touches my email, my calendar, or anything private.
Coding a round
From audio to a row.
Transcription and the first draft are automated; the judgement is not. Each step leaves something I can check later.
- 01
Clip the closing minutes
The quiz runs at the end of the episode, so I take just that tail of the audio rather than the whole show.
- 02
Transcribe the clip
Automatic transcription turns those minutes into text I can read and search.
- 03
Draft the row
A language-model pass turns the transcript into a structured row: the question, the official answer, each host's guess, and who got it right.
- 04
Check it by hand
Nothing counts until I've read it back against the published episode page and the audio. The draft is a starting point, not the record.
Scoring
How a winner is decided.
The winner is read from who answered correctly, not from a result I remembered. Three outcomes cover every round.
Winner
One host answered correctly.
Tie
More than one host got it right.
Stumper
Nobody got it right.
Verification
What counts as verified.
A model’s confidence is only a hint about review order. A row is verified when I can trace it back to the episode and recover why it says what it says.
Every verified row clears these
- 01The episode date and title match, even when a same-day bonus episode exists.
- 02A transcript excerpt and the audio source are saved with the row, not just the final answer.
- 03Any guest or stand-in host is named from the episode line-up, not guessed.
- 04The question, the answer, and each guess can be recovered from that saved evidence.
- 05The winner, tie, or stumper comes from who actually answered correctly, never from a label I typed.
Honest gaps
What I don’t trust yet.
The tracker is in a good place, but a few things still need work before I’d call it finished.
Episode status isn't fully tracked
I'd like every episode marked as tracked, confirmed no-quiz, still to do, or a blocked source. Today some of that lives in my notes rather than the data.
Not every row carries its receipt
Newer rounds save a transcript snippet and a timestamp. A few older ones were coded before I made that the rule, so they are thinner than I'd like.
Hosts aren't first-class yet
Who said what is read from the answer slots, not stored as real people with their own histories. Giving each host a proper record is the next thing I'd build.