2026-05-30
Entering operation phase, starting to diversify. Rt2112 Development Log | 2026-05-28 to 05-30
As the title suggests, after committing for about two and a half months, I've decided on the approach that 'this is the only way,' so I'm entering the operation phase.
I don't know if it's a winning formula or not, though.
Numerical snapshot
Main channel (as of 5/28)
Channel total (7 days): Views 41 / Imp 1.1K (+37% vs previous week) / CTR 1.7% / AVD 11:34
Subscribers: 18
v1 'For People Who Think Too Much at Night' (published 10 days ago)
Imp 200+ / CTR under 4.0% / AVD 12:30
Suggested videos 46.1% (largest traffic source)
Where suggested videos appear: Vintage Love Radio, Tea-table Talk, 3 HRS Continuous Bedtime Story (Get Sleepy), Jane Austen Documentary, Reborn at six
Proof that YouTube is accurately categorizing v1 into the 'long-form bedtime audio' category
v2 'When the World Gets Too Loud' (published 4 days ago)
CTR 0.8% (thumbnail is not working)
Simply put, this is how I'm breaking it down
Bad CTR
= Bad entrance
= Weak presentation of title, thumbnail, or theme
Bad AVD
= Bad content
= Issues with the opening, structure, audio, flow, or mismatch with expectations
Bad CTR and bad AVD
= High probability that the concept itself is off
Good CTR and good AVD but low impressions
= YouTube is still searching for the right shelf / Insufficient sample size / Waiting for channel learning
Applying this to Rt2112,
Low CTR, high AVD
→ Losing at the entrance. The content itself is resonating with those who watch. Fix the title, thumbnail, and first line.
High CTR, low AVD
→ It's getting clicks, but the content doesn't match expectations.
Likely clickbait, the intro is too long, or the first episode is weak.
Low CTR, low AVD
→ That theme/package is weak.
Don't force it to stay alive; move to a different format.
High CTR, high AVD
→ A winning candidate.
Expand horizontally with the same system.
Something like that.
Currently, I've started the 2-hour videos and am on the third one. CTR aside, the AVD is reaching double digits.
The order of improvement is
1. AVD (or total watch time) → 2. CTR → 3. Impressions → 4. View count ♻️
So, my current feeling is:
CTR is improving
AVD is improving
Impressions are still weak
In other words, what I should be questioning right now isn't that the "content is bad," but rather that
YouTube is still learning who to show it to / not enough videos / haven't secured a shelf yet.
AVD over 10 minutes → The content isn't significantly wrong
2-hour videos have over 3% CTR → The entry point is also improving
Few impressions → Keep adding 2-hour videos
That's the conclusion.
Roughly speaking,
CTR is the sign. AVD is the product. Impressions are the sales floor.
Now that the sign and the product are getting better, it's time to enter the phase of increasing the number of items on the sales floor.
Analytics automation
With the content solidified, the metrics to track are also set, so I've decided to automate the weekly reporting.
The purpose is the usual "watching the same numbers to see changes," but the
true purpose is
"to look away because I end up staring at them too much."
Bitter laugh.
The 1→3 phase of testing methods is over, and I'm now in the 3→10 phase of calmly building up toward monetization, so I want to eliminate noise as much as possible.
The temptation to think, "...maybe I should do it this way after all."
The noise is me, bitter laugh.
In terms of content, it's more of a stock than a buzz, so looking at overall metrics on a 7-day basis is sufficient. I plan to implement this by mid-June, with current KPIs set for the end of November and an interim check at the end of August.
This execution is the top priority.
Also, what to do with horizontal expansion and the time freed up by removing noise,
and then, the matter of the format when entering this operation.
A move to improve CTR
Since ① AVD has gone up, it means the content is on track, so the focus is on ② improving CTR for ③ impressions.
Title Design Framework v1 finalized
4-type matrix. The poetic type is abolished.
Direct Indirect (paradoxical answer) Narrow (Self) ① For People Who… ② On Nights When… Broad (World) ④ If You… ③ When…
Roles of each type:
① For People Who… = I have found you
② On Nights When… = I will leave this here for such nights
③ When… = A refuge for when the world becomes like this
④ If You… = If you want to imagine the world like this
Thumbnail design rules:
① For People Who… → Repeat and emphasize the assertion (repeat as is)
② On Nights When… → Briefly repeat the purpose (repeat as is)
③ When… → Present a paradoxical refuge
④ If You… → Return the answer to that hypothesis
Core: People in the narrow sense are already in the middle of their problem. Assert it for them without asking questions. People in the broad sense are looking for a way to view the world. Open the entrance to their imagination.
Ironclad rule for thumbnails: If you make them think, you lose. It must convey either "this is for me" or "I can find relief" in 0.5 seconds. The passive viewing audience for bedtime content comes here because they "don't want to think."
Identifying the cause of the CTR problem in v2
The thumbnail lacks an "answer" to the title. For "WHEN THE WORLD GETS TOO LOUD," the thumbnail only returns information (2 HOURS / 7 BEDTIME TECH STORIES) rather than a paradoxical solution (e.g., LISTEN TO A GENTLER FUTURE). The image is also a cluttered equipment room, which doesn't look like a "refuge for when the world is too loud."
👉 Of these four types, the first one has a good initial feel. For better or worse, YouTube viewers are passive (to put it very crudely, their brains aren't working when choosing a channel), so it's important to present things clearly and assertively.
The same goes for bedtime demand. At first glance, thumbnails that look like they might wake you up seem bad, but they actually have higher CTR and longer AVD.
Well, that makes sense. They're still awake, after all.
However, doing the same thing every day gets boring, so I'll proceed with the four types.
I might change the wording as needed, but the design likely won't change.
Lo-fi 2112 / Sleep 2112
Differentiation confirmed
Sleep 2112 (New) = Sleep-focused, future sleeping space, soundscape, no melody
Lo-fi 2112 (Existing) = Free slot, instruments x future street corners, work/study
Deep Focus affiliation is pending
Business and production chat for Sleep 2112 has already been set up
Lo-fi 2112 production line changes
Loop animation discontinued. Unified to only dust, stars, and fade-to-black effects
Update thumbnails to GPT quality
Unified with the same production line as Sleep 2112, significantly reducing man-hours per video
Thumbnail direction
Lo-fi: Bright warm colors, people and robots enjoying music (on-time)
Sleep: Dark cool colors, robots sleeping (off-time)
The difference in visuals from the main channel is not an issue (they won't be side-by-side because the channels are separate)
👉 I thought Lo-fi 2112 would play more of a supporting role for the main channel, but it hasn't been functional despite exceeding 40 videos. I'm reviewing it now that I've shifted the main channel entirely to bedtime content.
Since the main channel's BGM has also been shifted entirely to sleep-inducing soundscapes, I created Sleep 2112 for that purpose. The videos themselves are easy to make, but encoding takes nearly 3 hours, so I want to make effective use of sleep hours.
On the other hand, since the Lo-fi side will become even more of a free slot, I've decided to make it a series of interesting world instruments. Right now I'm doing something like cinematic marimba, and frankly, I wonder if it's even work BGM, but it's fun so I think I can keep doing it.
Music is free.
A move for the freed-up time
A. Instant Response Drill (Planned to be commercialized as content called Study River)
Since the monetization of Rt2112 seems like it will take a while, I decided to do an "Instant Response Drill" as something I can do in the short term without incurring costs.
The kind you often see in Kumon.
Simply generate problem JPGs with Python, and then
epub => ①Kindle
pdf => ②Gumroad
mp4 => ③Youtube
① and ② are bulk sales of 500 pages and 10,000 questions. There is demand for things like children's education. ③ is presented as brain training for adults.
Since the premise is to avoid spending too much effort, I have made it so that everything can be completed using tools like Pillow and FFmpeg, so I want to start on this once the company's fiscal year-end is over.
B. Pose Reference Book
The kind that manga artists and others use for reference.
Create images with the OpenAI API and convert them to epub. The cost for this is also only 200-300 yen per book, so it's easy to recoup, and it would be good if I could expand into areas like fashion.
C. Japanese Literature Audiobooks
Using Aozora Bunko, in English, of course. This seems like it could be done quickly by using the Rt2112 pipeline as is, so I will decide whether or not to do it based on the budget for the voice API.
It's a matter of motivation.
Priority is A=>B=>C
That said, as written in my LinkedIn profile, if it's okay to work 2-3 days a week, I would be happy to do AI design consulting if the opportunity arises.
Return to Bonfilet OEM Sales
That said, it's quite casual; I just send DMs on Thursdays.
Since the destinations are inquiry forms, the success rate is low.
Since Rt2112 is my main focus, I try not to spend too much effort on outbound sales that are at the mercy of time,
but on the other hand, I have been doing outbound sales for many years, so I generally know the timeline of when to reach out,
so I use Claude Code to extract event information, and at the timing when I think it would be good to make a proposal, I send a direct DM with wording like "I would like to make a proposal to you here!" that Claude created for me.
I don't know if it will resonate or not, but there is no waste.
Building up Shoten.me
I will resume building up Shoten.me, which I had left behind while it was pinned to my X profile.
It is a SaaS that simply introduces books via ISBN and ASIN, but well, if I don't do it, who will? It's designed so that I can enjoy it myself in the early stages, and it's an introverted kind of SaaS that I can keep doing forever even if no one else uses it.
Production Progress
Completed
Vol. 11 Kindle publication complete, books site updated
8-minute single episode unlisted from the main channel
Rt2112, 2-hour video, brief created
v4 "For People Who Still Remember That Place" — Type ①, Immersion in a Place of Memory, Episode 1 "To That Place in Memory"
v5 "If Your Ordinary Days Could Reach the Future" — Type ④, Ordinary Days Reaching the Future, Episode 1 "To You, Five Generations From Now"
v3 Release
Title: When Something Quiet Stays Beside You (Changed)
Thumbnail Text: IT NEVER LEFT
Intro Narration Finalized: "Some bonds arrive quietly. A voice. A machine. A presence that stays. Tonight, these are gentle future stories about the ones we never expected. Close your eyes, and just listen."
58-minute rewrite allocation (Level 6 test bench)
1st (Siri/Robot/Geothermal): On Nights When You Hope Something Reaches You → Thumbnail TBD
2nd (Universe 4-part series): On Nights When the Universe Feels Quiet Too → Thumbnail "THE UNIVERSE FEELS QUIET TOO"
3rd: TBD
Business Decision
June-August Roadmap Finalized
Goal: 30-piece collection
Strictly maintain 2 pieces per week, with a 3rd if capacity allows
Start distributed operations (allocate mental resources to Study River, Bonfilet, and shoten.me)
End-of-September Evaluation Criteria
Less than 3,000 monthly → Question full commitment to YouTube
3,000 to 10,000 monthly → Extend until December
10,000 to 30,000 monthly → Continue, monetization within range
30,000 or more monthly → Full-speed ahead
👉 "Stopping YouTube posts" and "closing the Rt2112 business" are separate decisions. I think I will be doing this forever, even if I were to get a job at some company.
Monetization Timeline
Optimistic: end of the year; intermediate: 2027 Q1
The bottleneck is subscribers (WT will be achieved first)
Podcast Channel Auto-dubbing
Videos via RSS are highly likely to be excluded from auto-dubbing
Leave it alone. Maintain the design philosophy of zero production man-hours
Pipeline
Manga Short API Auto-posting
Manga prompt output via Claude Code -> Image generation via Gemini -> PNG to JPG compression (done) -> API posting (not yet implemented)
Start after the financial closing. Only TikTok API / Instagram Graph API connection remains
PENDING
[ ] Vol. 12 writing (June)
[ ] Remaining Kindle publishing (4 revised entry-level volumes)
[ ] Manga short API connection
[ ] Analytics API connection
[ ] Create various accounts related to Study River
COMPLETED
[x] Vol. 11 Kindle publishing
[x] Update books site
[x] Set 8-minute single-topic main channel to private
[x] Finalized title design framework v1 (4-type matrix)
[x] Finalized thumbnail design rules (3 patterns: Iteration/Solution/Answer)
[x] Created v4 brief
[x] Created v5 brief
[x] Updated v3 title and intro
[x] Created v3 thumbnail (IT NEVER LEFT)
[x] Finalized titles for two 58-minute videos
[x] Finalized June-August roadmap (30-video goal)
[x] Finalized September evaluation criteria
[x] Simplified Lo-fi production line
[x] Launched Sleep 2112 channel
[x] Considered replacing v2 thumbnail
[x] Semi-automated Bonfillet sales
[x] Designed mass production for instant drills
[x] Designed mass production for pose reference books
Created: 2026-05-30