2026-07-13

Looks like it's going to be a week of shifting to GPT5.6 Sol

fable is prone to usage limits anyway.
In a way, I can take that as a forced break.

Weekly Operational Report 2026-W29

Target Period: 2026-07-06 to 2026-07-12

Metrics

Comparison between the last 7 days and the 7 days prior.

Channel views Total watch time AVD
Rt2112 4 (▲33%) 4 min (±0%) 1:14
StudyRiver 10,859 (▼8%) 1,038 min (▼10%) 0:16
StudyGym 6,099 (▲1,678%) 814 min (▲825%) 0:32

The 28-day cumulative total watch time is 512 minutes for Rt2112, 2,905 minutes for StudyRiver, and 903 minutes for StudyGym.

What I did this week

• StudyJaEn: Established a new English writing drill for Japanese speakers. Generated a total of 4,000 sentences, 1,000 each for A1 to B2 levels, and passed 4,000 machine verification checks. There were zero duplicates. I corrected about 10 items through random spot checks. Next steps are PDF conversion and listing.

• StudyRiver: Regarding additional distribution channels for short-form videos, I chose the official TikTok review route. I changed the operation to test multiple genres in the same experimental slot and only spin off those that show a reaction. In the first week, calculation and mental math-related content showed a relatively strong reaction. => TikTok has been renamed from River to Gym. Operating as Gym.

• StudyGym: Views increased by 1,678% and total watch time by 825% compared to the previous week. This is the first clear numerical reaction as a newly established lane.

• System Trading: Running in the background. Expanded the verification infrastructure with fable. I retracted good results that appeared once by confirming nearby conditions and raised the reporting standards. I also checked signals in other directions, but currently, they cannot exceed costs, so I have not moved to actual operation.

• Nenne Bunko: Now in the cultural category. Changed the policy to refine modern translations as readable material while maintaining the facts of the original text. I changed the order to complete the book first and then expand into videos. The plan is to complete the translation of the first installment in two volumes.

• Eidendo Make: Decided on a policy to move existing services to the operational lane and shift my own time toward the sales side. Rather than increasing new services, I will organize the flow to existing receptacles.

• Shoten.me: Resumed low-speed operation. Proceeded with bug fixes, publishing 2 new shelves, organizing the procurement process, and designing the automatic posting lane.

Verification

• For StudyGym, not only views but also total watch time and AVD increased simultaneously. This is the strongest new signal in this week's numbers.

• StudyRiver maintained the 10,000 views level, but both views and total watch time decreased by about 10%. As an existing lane, it is being maintained, while new growth is appearing on the StudyGym side.

• StudyJaEn kept the correction rate to about 0.3% while generating 4,000 sentences in one day by separating machine verification from visual confirmation.

• In system trading, I was able to retract good results from a single condition instead of adopting them as is. Improvements to the judgment criteria remained more significant than the results.

• Across multiple businesses, I am shifting from "creating new things" to an operation of "observing reactions and spinning off only the hits."

Hypothesis

Gym is originally a red ocean area, which conversely means it is easy for algorithms to run. If autonomous operation is the premise, you can pack it with an overwhelming amount of data, so in a sense, it is a blue ocean.

Next week's move

Since GPT 5.6 Sol is powerful, the main focus will be on operational improvements while shifting towards Codex. I will create a two-person (Fable and Sol) system.
Storage is particularly tight, so I am debating whether to move video generation to a sub-desktop (it's an Intel Mac, so Codex won't run, haha) to separate it, or to buy an external SSD (which is expensive).
It looks like it will be a week of trial and error, pushing things I was running locally to GitHub and so on. The time to cancel Adobe might finally be here. The issue of it eating up too much storage. Under consideration.
In short, I want money.

This week's focus

As I wrote in my hypothesis, 'stepping into a red ocean zone' might be the right move on YouTube.
The reason is that YouTube itself understands the algorithm.
BGM channels also had a fast initial response, so things that are understood are likely fast.
Conversely, distribution for things like Rt2112, which make people go 'What is this?', is cautious.
It really feels like a platform run by highly educated, stuck-up people, but well, it can't be helped.

With that said, I've decided to use TikTok for high-speed POC.
In the end, even if you think 'this might be good,' whether it hits or not is a different story, so I'll just make simple content and push it to TikTok. I want the facts verified through field testing.

The first numbers are in for StudyGym

I simply put the 'brain training common tropes' category onto an automated lane. However, things that hit, hit.
The maze content seems like it will grow, so I'll probably run it for another two weeks, and if the results improve, a dedicated channel will be created.

I am a lifelong learning manufacturer.

About the English learning channel

About Study JA/EN. This is the third horizontal expansion, coming after River=>Gym. Language-based content is hard to expand globally, so it's different from my usual non-language automated lanes, but
I realized that YouTube's initial response always starts from Japan, the destination of the IP address. That hasn't changed, so I'm going to try to turn that to my advantage.

While YouTube's auto-dubbing translates into over 20 languages for English content, it only goes to English from Japanese.
Therefore, I'm sticking to the axis of using only English.
On the other hand, if Japanese viewers don't have the settings disabled, it will speak in Japanese automatically, so it won't function as an English listening channel. That's why I've decided to specialize in a writing channel.
Originally, River was for calculation drills and used Gumroad as the conversion destination, so English will also specialize in writing from Japanese to English.
As a byproduct, there is English audio, and since subtitles, audio speed, and audio can be changed automatically, it's designed to connect not just to English study, but also to language groups around the world studying Japanese.
Because of that, I don't want to spend money on audio, so I'll adopt open-source TTS. It's actually a great match for something like English learning.
Since auto-dubbing machine audio is also possible, for example, if a Hindi speaker sets it to Hindi audio and subtitles, it becomes a way to study Japanese as well.

From an English-speaking perspective, it's like looking at Japanese and translating it into English.
It's like they are doing translation study, the way we usually do 'I have a pen => 私はペンを持っています'.
Or rather, Japanese people can do English listening anywhere, but it's said that they lack speaking and writing, so I'm imagining capturing the needs for that.

In any case, this is content 'for myself' (English is a bottleneck, haha).

Businesses to build, and businesses to fold

I will continue to manufacture 'lifelong learning' content that can be put on an automated lane.
What hits on TikTok will get a dedicated lane, and what doesn't will be discarded.
The reason I'm adopting TikTok is that the initial response is fast, there are many young and bright users with relatively good reactions (in terms of sentiment), and it doesn't have the distribution filtering on the platform side like YouTube does. Also, TikTok itself is a platform where you can earn money. (Instagram is low priority).

Rt2112 isn't producing results. I'm not folding it, but I'm thinking of running it on a smaller scale as a project I want to do.
If anything, I'm thinking of it as routine work for my own creator career, so I'll stick to that and do it calmly.

Summary

Since my commitment to YouTube operations for Rt2112 in early spring, I've broken down the methods and expanded them to other businesses, so although it's a detour, I'm moving straight ahead, the evolution of AI is fitting in well, and it's now in a state of autonomous operation until results appear.
In particular, Fable and Sol are almost entirely on autopilot.
When it gets to this point, I've started to think on my own that the 'lifelong learning' genre might actually grow in the future.
It's a senior's nature to want to keep their brain from declining even when they have nothing left to do, and I wonder if that will eventually reach the working generation as well.
Originally, senior content for lifelong learning is no different in method from educational content for toddlers, and I wonder if it's a future that will gradually narrow down.

Either way, it looks like I'm going to shift to Sol operations. Japanese is the best, so I hope Claude does his best.