2026-10-05

Small improvements.

Weekly operations report: 2026-10-02

Period: 2026-09-26–2026-10-02 (Saturday–Friday)

Numbers

Latest 7 days 2026-09-24–2026-09-30 vs. previous 7 days 2026-09-17–2026-09-23

AVD includes Shorts.

Reference: cumulative 28-day totals

• Rt2112_Bedtime: 14 minutes
• Rt2112_YT_Podcast: 15 minutes
• StudyRiver Math: 5,242 minutes
• StudyRiver Lang-JaEn: 3,681 minutes
• StudyRiver Lang-Ja: 1,668 minutes
• StudyRiver Maze: 63,348 minutes
• StudyRiver Logic: 2,234 minutes
• Speed Odd One: 16,189 minutes
• StudyRiver Lang-EnJa: 120 minutes
• StudyRiver Stat (ja): 3,173 minutes
• StudyRiver Stat (en): 150 minutes
• StudyRiver PD: 1,632 minutes

Work completed

Maze Marathon recorded 42,959 views and 20,468 minutes of watch time over the latest seven days. I switched scheduled content to the Halloween miniature-world theme starting on 10/1 and produced 21 Shorts and 6 long videos. I introduced individual descriptions and pinned comments, and chapters for newly uploaded long videos. On 10/2, I delivered 512 additional miniature-world assets, but they had not yet been incorporated into videos or published.

On 9/30, I updated volume 0 of the Study JaEn book and aligned the chapter titles and opening pages with the listing information. I set the introduction video as the channel trailer and prepared listing copy. The cover was not yet made, and I was still waiting for a response on the proposed structure of the free edition.

For StudyRiver Math, on 9/26 I prepared five long-video modes with a foundational stage and Shorts at different difficulty levels, then checked 5 completed long videos and 42 Shorts. After revising the break-time prompts and blank-answer display, I connected the new versions to the publishing process from the new slots on 9/30. For Instagram, I prepared 48 revised videos and set new content to take priority from 9/30, keeping the rate at 3 posts a day. The logs did not confirm the first successful post of the new version. On 10/1, I also switched the TikTok math stock to 72 revised videos.

For the StudyRiver A4 worksheets, on 9/28 I released a package that creates printable worksheets from a conversation, and added links to the Japanese, English and Spanish information pages. I added kanji tracing and written arithmetic, and published two Japanese usage videos. Use by individuals, families and teachers is free; selling the worksheets as products or redistributing the package requires permission. An additional route for free users was still described as being under application in the relevant logs.

For StudyRiver Find, I started daily long-video scheduling on 9/26. On 9/27, I verified the process from making a new background to placing the video in the publishing queue. I excluded works with mismatched resident and background proportions from publication and scheduling, and switched production to inspected background stock. On 9/30, I changed the schedule to 1 long video and 3 Shorts a day and connected replenishment for the following 3 days. On 10/1, I completed 2 long videos; one was released from its hold after I replaced its music with audio whose usage terms I could verify.

For StudyRiver Lang-EnJa, on 9/26 and 9/29 I corrected the checks for stopped scheduling, resumed checking published videos, and restarted schedule replenishment. The check on 10/1 found all 16 slots, including that day's slots, filled. I used 20 verified kanji to produce 1 long video and 3 Shorts. The configuration to include kanji in existing slots from 10/5 was complete, but no kanji videos had been scheduled as of 10/1.

For StudyRiver Logic, on 9/28 I added spoken instructions to the beginning of long videos and gave the practice questions the same timing as the main section. I added 4 question formats, bringing the total to 30, and expanded the rules for shape sequences. I made 10 trial videos and checked that each question had a unique answer across 3,000 cases. The changes apply to new production; I did not replace published or scheduled videos.

For StudyRiver PD, on 9/30 I finished 1,797 new everyday-vocabulary images, completing the image set for all 2,017 words, including existing entries. I moved to video production using the finished images. The project list reached 909 entries, including existing ones. At the migration check, all 838 unused entries had images, equivalent to about 279 days at 3 entries a day. I also finished making and checking the conversion of 7 ebooks organized by everyday situation, but had not listed them for sale.

For Speed Odd One, on 9/27 I connected long videos to a schedule of 1 standard video and 1 animal-themed video per day. I replaced 4 slots on 9/28 and 9/29 with versions containing breaks. The week's automatic-operation logs recorded uploaded and scheduled totals of 14 long videos and 37 Shorts; these are separate from completed publications. On 10/1, I prepared 37 revised TikTok videos and changed 5 queued slots between 10/3 and 10/10 to the new version.

For shared StudyRiver social operations, on 10/1 I worked through a batch of 27 English TikTok slots for 10/3–10/11 and checked and recorded all 27 in the list. I posted one of them manually. One scheduled post with an emoji-only caption and two slots where the records and list disagreed remained to be checked.

For eidendo, on 9/30 I published entry links to StudyRiver in 3 languages. Search data collected on 10/1 showed clicks rising from 261 to 388 and impressions from 3,029 to 3,858 between the 28-day periods 8/4–8/31 and 9/1–9/28. These are not results for this activity week, and an increase in orders was not confirmed. For sales outreach, I reviewed contact selection and sent 2 messages. Successful publication of the daily website improvements had not been confirmed since 9/23; the cause and recovery were still unresolved.

Looking back on last week

I changed the video structure. Instead of letting everything run continuously, I added breaks. A break comes before the average viewer leaves, followed by a rough Pomodoro-style 5:1 rhythm. I applied it to Odd and Logic.

Odd's views have honestly fallen since I switched to animal masks, but that is fine as part of developing IP and making it distinct. Watch time was already fairly good, but subscriber numbers were small, so it needed work either way. I personally like Logic, but its lack of popularity reminds me how difficult the boundary between education and play can be.

I keep making these small improvements.

Meanwhile, GPT's Dots was released, and it looks like a good fit for my current setup. My previous process was roughly, “If something breaks, notify me in Google Calendar.” Now I can just check the Dots chat and work on operational improvements there too. I feel I may be able to focus more on expanding in different directions again.

This week's move

I asked Dots to put together next week's task list over the weekend. The list appeared in the calendar notification, so this is the week I work through it by brute force.

A closer look

Breaks: when to evaluate a new video structure

Odd, Logic and Maze have something in common: their AVD is in the low one-minute range. The AVD at the beginning of this article includes Shorts, so it is shorter. I want to bring the long-video figure to 2 minutes. Odd's CTR is in the 5% range and Maze's is 10%. My hypothesis is that increasing AVD alone will increase distribution.

The specific approach is to insert a break before viewers leave, around 45 seconds to 1 minute. I treat the first 45 seconds as a warm-up, then ask something like, “Ready?” It is a small change, but I am seeing gradual improvement, which is interesting. I expect growth again from late October.

Viewers still leave, so I also inserted a Subscribe button there to remind them before they go. For this kind of content, subscribers are the challenge more than watch time. I will keep using Shorts and other formats until I reach 1,000.

Seasonal themes and year-round asset stock

For Maze, monetization by the end of the year is starting to look possible at the earliest. I am less sure because long-video views dropped over the 10/3–4 weekend just before I wrote this on 10/5. Still, I have started preparing for YPP. The biggest risk with the current content is a finding that it is repetitive, so I am adding variation.

The fixed element is the maze, in 2D and 3D. The variable elements are the route, maze type, design and characters. Routes already change each time, and I am expanding the types. Having made the miniature-world 3D format, I will vary its presentation through design and characters.

The regular year-round format is the current leading 2D version. Its plainness works. Of the two daily posts, I am changing one to a seasonal format. October revolves around Halloween, along with other ideas. It is the same idea as seasonal ice cream. How you sell things does not change just because AI can make them.

Making a large image stock with Codex

OpenAI keeps giving me reset tickets, so I wondered what to use them for. I am putting them into assets for PD, the picture dictionary, and clothing, the apparel work I see as the other pillar alongside StudyRiver.

“Generate images until the credits run out.” They do not seem to run out. Incredible. I hit a limit at over 2,000 images a day, and I am making 10,000 this week. After that, combinations can keep the content going indefinitely. Do not forget to connect an external hard drive.

Where I use AI

The biggest reason I currently use GPT rather than Claude as my main tool is that Claude is great for operational improvement and logical design, but those do not necessarily connect directly to where a solo business owner earns money. “Would you like to improve your operations with AI?” works now, but in six months it may be unnecessary. If AI can easily do the same thing, let it.

Over the long term, the places where business happens do not change much, and GPT feels more general-purpose and easier for me to use. My impression is that Fable is more accurate, so I use it for the first pass. GPT6.1 Sol handles operations. When things get stuck, I have Opus5.5 work on improvements.

I really do not need Adobe

I wanted to splice a little video through a GUI and solved it immediately with a quick app. Premiere may become legacy baggage I only use for client work. Canva connected quickly, so it may survive for the purpose of making small GUI adjustments.

People on X say, “Opus5.5 made this amazing ad in one go,” but that has been possible for a while. For short videos, I definitely will not need the old tools. I think even long videos will change through commands: show only the timeline and selected editing area, then say what to do here for how many seconds. It feels as though we are already getting there.

I no longer need Photoshop at all. I only use Illustrator for client work. I still use those tools for small client revisions, but I wonder whether the person making the work could just use AI themselves. Tell it what to change and fix it during the meeting.

This week's closing thought

I will keep working quietly and steadily until monetization. Once that happens, maybe I will turn these reviews into videos and talk about them. It is just a passing idea, so I do not know whether I actually will.