Status: paused. 47 episodes written of 122 in scope.
What it is
Before the RV, I spent six years making Minecraft videos as BasketMC, mostly with my co-host Scoti as "Team B.S." The videos are still up. The website that went with them was not really a website: every episode had a page, and every page was a dump of the video description. Links, map download, socials. Nothing you would actually read.
This experiment turns those episodes into written recaps. Not transcripts and not summaries, but the episode told as a story, with stills from the video and the two of us doing recognizably different things.
Where it stands
Captive Minecraft 4 (Release the B.S.) is done: all 32 episodes, plus a bonus page for the hand-drawn cartoon that closes the series.
Untold Stories 3: Myriad Caves, a 90-episode run through renderXR's CTM map, is 15 episodes in and currently on hold.
That is 47 written out of 122 for these two series. The rebuild is not live on basketmc.com yet.
Why it is harder than it sounds
The Nomads Amor pipeline has it easy by comparison. Those videos are recent, they have clean auto-captions, and there is one narrator.
These do not.
Many of the old videos have no usable captions. When there are no subtitles to pull, the workflow has to work from the footage directly, reading what happened off the screen rather than off a transcript. That path is dramatically more expensive than reading a transcript, and it is the main reason this project is paused rather than finished.
There are two people talking. A recap that says "we found the emerald" is worthless. The interesting part of a Team B.S. episode is usually one of us doing something dumb while the other reacts, so the writing has to keep track of who did what. That attribution problem does not exist in a single-narrator video at all.
Stills have to be chosen, not sampled. Grabbing a frame every two minutes gives you a wall of identical cave shots. The frames have to land on the moment being described, which means seeking to a specific timestamp in a local copy and checking what actually came out.
What I have learned so far
Old footage punishes assumptions. Seeking directly to a timestamp during download lands off target on these videos, so the whole file gets pulled at 720p and cut locally instead. YouTube also intermittently decides the downloader is a bot, while the same data parses fine from a plain page request. Both cost a cycle to find and neither is interesting once solved, which is exactly the kind of thing worth writing down so it is only solved once.
The bigger lesson is about scope. Captive Minecraft 4 took 32 episodes to find a page shape that works: intro, the video, then alternating sections and images, ending with a link to the next episode. Untold Stories 3 inherited that shape and has been faster per episode because of it.
The lesson I did not expect is that cost per unit matters more than cost in total. A pipeline that works beautifully but is expensive per episode does not scale to 90 episodes, and finding that out at episode 15 is better than finding out at episode 60.
What's next
This one is parked until the per-episode cost comes down, most likely by transcribing the audio locally instead of reading the footage. Episodes get written in batches when there is capacity for them rather than on any schedule.
Whenever it does finish, there is still the question of the old pages: they carry the original publish dates and ordering, and a lot of indexed history, so the rebuild is not done until that is properly handled rather than simply replaced.
The channel is BasketMC on YouTube.