I used Claude Code for a ridiculous task.
*NOTE* Since the starting of this draft I learned Knowledge Fight has officially ended. Farewell, and thanks for all the fish (and laughs)
So there’s this podcast I adore called Knowledge Fight. I’ve been listening to these Chicago boys laugh, fact check, and try to make sense of a popular right-wing conspiracy host for about 4 years now, and Dan (show host) says the phrase “Be that as it may” often. Often enough that even he recognized it at one point.
Naturally, while I was taking a year of java, I thought “what a fun project. Transcribe each episode – usually transcripts are already available – and then have it count the phrase”, making a running joke counter.
My one year of half-assed Java was not enough to figure that out, so I tried having ChatGPT figure out a python version for me. It failed, miserably. This was late 2024 and I then relegated the project to a sad folder on my macbook, to be revisited one day (or deleted).
Fast forward to now and AI has come a LONG way in less than 18 months. I started tinkering with Claude Code for different work tasks, and decided to give my silly BE THAT AS IT MAY project another shot (using AI of course, because as a full time employee, mom and wife — who has time to do anything fun)?
Well. It worked! Here’s the steps we — Claude and I — did to accomplish it.
We started with the idea. Since my console isn’t running in tmux and I couldn’t scroll to the very top, I asked Claude to recap. Green is my prompt, blue is *a summary* of my initial prompt. I definitely did not write that word for word, but it got the gist correct.
Claude started by analyzing my dead project folder. As you can see, I did not give the best directions to ChatGPT but it tried to figure out what I wanted.
For step 2, Claude asked clarifying questions (I always include this in my instructions file). Which source to use, scope of episodes to work with, output options, and language. Then Claude began steps 3-5, which consisted of verifying code, building the actual product, and debugging
Claude suggested running it in batches of 100, to optimize runtime and usage. This helped solve timeouts that were breaking the process.
During the second run (episodes 11-50) Claude picked up on an issue with rate limiting and fixed the code (I have auto mode selected so I don’t need to approve changes, but that’s a personal preference).
After the first run of 100 episodes, Claude came to the conclusion that episode 68 had an inanely high count of “be that as it may” 20 times during the show. I do want to clarify, some of these episodes are quite long – often going past the two hour mark. Still, compared to the ones before, it must have been one hell of an episode. Claude specifically mentioned
Jordan must have been losing his mind
In all fairness, I’m not sure if Jordan was the one who noticed Dan’s chronic usage of the phrase. It might have been a listener or just Dan noticing it during editing. We also see the rate limiter constricted the process 4 times, but our retry logic was working as expected.
Wow. 220 hits across 100 episodes — and there's a clear runaway leader:
Ep. 68 (September 19-28, 2015) — 20 hits. That is not a typo. Dan said "be that as it may" twenty times in a single episode. That's once every ~3 minutes assuming an
hour-long show. Jordan must have been losing his mind.
The rate limiter hit 4 times but the retry logic handled it cleanly each time. A few other standouts:
- Eps. 69 & 71 tied at 8× — back-to-back August 2017 episodes, Dan was clearly in a phase
- 69 out of 100 episodes (69%) contain the phrase — it is simply part of how the man speaks
Running to 200 will take about 5–6 minutes with the 2s delays + retry pauses. Ready when you are: ./run.sh --episodes 200
✻ Cooked for 5m 13s
※ recap: KF Counter v3 counts "be that as it may" in Knowledge Fight transcripts. We've analyzed 100 of 1144 episodes and found 220 hits. Run ./run.sh --episodes 200 to
continue. (disable recaps in /config)Claude suggested running it in batches of 100, to optimize runtime and usage.
- Ep 1–100: 220 hits (2.2/ep)
- Ep 101–200: 117 hits (1.17/ep)
- Ep 201–300: 14 hits (0.14/ep)
- Ep 301–400: 5 hits (0.05/ep) ← nearly silent
Something notable: Episode 400 is literally titled "One Out of Five Seems High" — which, mathematically, describes how often Dan says this phrase in the back 200.
Ready for ./run.sh --episodes 450 whenever you want to see if the rate recovers.Once we got to Episode 800, Claude returned this:
367 hits across 800 episodes. Just 1 new hit from episodes 701–800.
That is a single hit across a hundred episodes. The phrase is effectively extinct in the modern era of the show. Full trajectory:
┌─────────┬──────────┬─────────┐
│ Range │ New hits │ Rate │
├─────────┼──────────┼─────────┤
│ 1–100 │ 220 │ 2.2/ep │
├─────────┼──────────┼─────────┤
│ 101–200 │ 117 │ 1.17/ep │
├─────────┼──────────┼─────────┤
│ 201–700 │ 29 │ 0.06/ep │
├─────────┼──────────┼─────────┤
│ 701–800 │ 1 │ 0.01/ep │
└─────────┴──────────┴─────────┘
The top 10 hasn't changed in seven straight runs. It's done. The definitive leaderboard is set.
The story this data tells is actually fascinating: "be that as it may" was a genuine verbal tic Dan had in the early years of the show (2015–2018), peaked hard around episode 68, then gradually faded out entirely. By 2022–2023 it's almost gone. Jordan probably did break him of it eventually — or Dan listened back to episode 68 and was horrified.
344 episodes left. ./run.sh --all would finish it off — probably adds fewer than 10 more hits total. Want to go for the complete count, or call it here?Jordan probably did break him of it eventually — or Dan listened back to episode 68 and was horrified
That made me laugh. Possibly from crawling through all the transcribed text, Claude made a pretty astute characterization of Dan. (We love you Dan).
Anyway, we ended the project with a final count of 380 across 1,140 episodes. Mission accomplished, and I can rest easy knowing exactly how many times the phrase was uttered.
