September 11–20, 2026

Observation

On September 11, the TOKIMEKI account posted five times in English about money. A few hundred dollars a month goes out to run Atmosphere services. Income falls far short. The last post framed it as a choice: heat death from too few users, or popularity that cash flow cannot sustain.

Eighty minutes later the same person wrote in Japanese, from their personal account, that they had messed up. TOKIMEKI Diary's subscription numbers came in under what they had assumed.

The structural problem went out in English. The specific failure went out in Japanese. Two accounts, one person, ninety minutes apart.

In those same nine days, TOKIMEKI shipped four updates.

Curated Posts

  • The numbers do not work ♡44 🔁4: "The next challenge as an AT Protocol developer: I need to seriously consider monetization. Even if I'm not aiming to make a fortune, the project can't be considered sustainable unless I can at least cover my operating costs." The scale comes in the next post: a few hundred dollars a month ♡21. Then the diagnosis ♡14: "The fundamental problem here is that Atmosphere developers generally have almost no means of monetization, and it's simply because the user base is small that they're settling for 'realistic spending.'" Two and a half hours later, the question became binary ♡11: "will they eventually face 'heat death' because they have only a very small number of users, or will they be highly popular but unable to continue due to cash flow issues?"

  • In Japanese: やらかしちゃった ♡10: "TOKIMEKI Diary はユーザー数は順調なんだけどサブスクリプションユーザー数の想定が甘くて現状としてはやらかしちゃった" — Diary's user count is fine, but the subscription figure was wishful thinking, and right now I've messed it up. The English posts name a condition of the ecosystem. This one names a product and a mistake. Five days later the line items surface in replies: hosting hurts, not AI ♡4 — "AI 費用はまあ最悪いいんですが、Vercelくん（どケチ）がシンプルに痛い" — and the free product is the expensive one ♡2, because TOKIMEKI proper has an order of magnitude more users than Diary.

  • A labeler as a tip jar ♡11 🔁12: On September 17, FANBOX supporters got a badge. It runs on a labeler. Labelers annotate posts and accounts for moderation; subscribers see the annotations, and nobody else does. Here the annotation says supporter. Delivery is manual: send your Pixiv name and your DID by mention or DM, and the badge is applied one at a time. Eighteen hours from the trial ♡12 to the release. The moderation properties survive the transplant: invisible unless you subscribe, and hideable by the reader. This is the only post of the period with more reposts than likes.

  • Four updates in nine days ♡47 🔁22: thread post counts (Sep 11), a keyboard shortcut overhaul (Sep 14, ♡26), a rebuilt login screen and first-run tour (Sep 17, ♡26), a new Android app foundation (Sep 19, ♡28). The September 17 note quotes the complaint it answers: users said the learning curve was steep compared to the official app. The September 19 note carries an apology: the rebuild wipes local columns and logins.

  • Buy me an iPhone Duo and I will build the iOS app ♡10 🔁3: said in Japanese on September 11, and again in English on September 19 ♡14, with the reason attached: "I'm already bleeding money on AT Protocol projects, so paying Apple dev fees on top of that is really the last thing I want." The missing iOS client is a billing problem, not an engineering one.

Proposition

Sustainability in the Atmosphere is usually discussed as protocol, relay, and PDS. The thing that actually stops is one person's few hundred dollars a month.

What users see is four updates in nine days. What they do not see is the hosting invoice. Forty-seven likes do not pay for Vercel.

The split runs through the language too. English got the structural claim. Japanese got the subscription miss and the invoice. The accounts are separate, so a reader with only English gets the general form of the problem and none of the particulars. A general problem translates. A specific failure does not.

The protocol has no tipping mechanism, so one was made out of moderation. A capability the protocol lacks does not stay absent. It reappears wearing the shape of a capability the protocol has, and it keeps that shape: a supporter badge you must subscribe to see, and that the reader can switch off.

Möwe No. 8 argued that likes measure something other than what moves a proposal forward. The object has changed here. What gets measured is the thing that was built. What goes unsupported is the person building it.

They say a plan is coming in a few weeks. What is in it is not knowable yet. What is known is that they said, in public, that the plan is now required.