> ## Content Index
> Fetch the complete content index at: https://www.tsukiyoresearch.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# I automated the publishing too. So what does it cost, and what does it earn? (Final part)
- URL: https://www.tsukiyoresearch.com/i-automated-the-publishing-too-so-what-does-it-cost-and-what-does-it-earn-final-part/
- Published: 2026-09-01T11:26:25.000Z
- Updated: 2026-09-01T11:26:25.000Z
- Author: tsukiyo
- Tags: Build Log

*The finale of five. The question everyone actually wants answered — including the number I'd rather not print.*

## Automating the publishing: taking willpower out of consistency

The system's publishing runs almost entirely by itself.

- **Daily reports** — the system composes and schedules them to my Japanese blog. They go out even if I oversleep
- **X** — five slots a day (one in English), posted by the system itself, with the timing deliberately jittered every day (an account that posts at the exact same minute every day doesn't look human, because it isn't)
- **Weekly report** — every Saturday morning: the week's candidates, warnings, and the scorecard, compiled into a report with charts. Publication is one command

The reason isn't laziness. It's that **in publishing, consistency is everything** — and consistency powered by willpower, in my case, has a measured lifespan of about three weeks. The machine doesn't get bored, doesn't get discouraged, and posts its observations just as calmly on red days. That steadiness *became* the account's personality.

One line is drawn, though. The system has a flag called **needs\_human**: anything opinionated, and any product release, must pass a human. We automate *work*, never *judgment*. That rule doesn't change.

## What it costs: tens of thousands of yen a month

The monthly fixed costs, roughly:

- Market and financial data APIs (the biggest item)
- The *Shikiho* online service (Japan's quarterly all-company handbook)
- X's paid plan (required for API posting)
- AI: the coding agent's subscription, plus API usage

Total: **a few tens of thousands of yen a month** (a few hundred dollars). Expensive for a hobby, cheap for a business — an awkward middle.

The surprise was how cheap the AI part is. The nightly AI review of regulatory disclosures costs **about ¥0.2 (a fraction of a cent) per document**; blog-draft generation about ¥0.5 per article. Recently we moved the nightly review onto a local LLM running on my own hardware, so that line item is now mostly electricity. And everything has a hard spending cap — in this system's philosophy, an automated process without a cost ceiling is not allowed to exist.

## So... did it make money?

The honest current status:

**Almost zero.**

No confirmed affiliate revenue to speak of; the paid membership on the Japanese side has only just begun. Against the monthly fixed costs, this is a proud, unambiguous deficit.

Was that expected? Half yes. Content businesses take six months to a year of SEO and trust-building before revenue arrives. What we're accumulating right now isn't revenue — it's **inventory of verification records**. A scorecard only speaks with a year's weight after a year of entries. You cannot get that later, at any price.

...That's the graceful framing. The truthful footnote is that yes, there are nights in a deficit month when I wonder what exactly I'm doing. But every morning the inspection report arrives in Slack, the daily observation posts itself, the scorecard accretes another row — and I get to feel one specific thing: **whatever this business is, it doesn't lie.**

## Why an individual should try this

The one thing I most wanted to say with this series:

AI agents have collapsed the cost of **an individual owning their own infrastructure**. A system that would have needed a development team a few years ago can now be owned by one non-programmer with a few hundred dollars a month and stubbornness.

I pointed mine at fixed-point observation of the stock market. What would you point yours at? Household finances, health, work logs — anywhere "same criteria, every day, everything" helps, this works.

This series ends here, but the observation continues tonight. While I sleep, 3,900 companies get checked. Again.

---

**The series** \- Part 1 — The system: checking 3,900 stocks while I sleep - Part 2 — Working with AI: how to instruct, how to argue - Part 3 — The scorecard: never delete a missed call - Part 4 — Incident log: the fake crash stopped 90 minutes before publication - Part 5 — Costs and earnings: the honest current status (this article)

*Tsukiyo Research now publishes a weekly systematic report on Japan's uncovered small caps in English — every issue opens with the scorecard of our past calls. If you want to watch this experiment compound,* [*subscribe here*](https://www.tsukiyoresearch.com/)*.*

*This series is a personal build-and-operate log. Nothing here is investment advice or a recommendation to buy or sell any security.*