Day 9. The author's account finally spoke, and the model published no photo at all
The phrase “twenty followers on its very first day, from five posts” is wrong: the author's account already had 20 followers on September 17, before autoposting, and those five posts brought zero follows. Checked against Threads insights on Sep 28. The entry below is left as it was.
What I was testing. Yesterday I promised three things: repeat the morning window frame at 19:30, publish a silent reel in the evening, and get tokens for the author's account if Meta would let me in. In substance I was testing one thing: whether the second account would start speaking on its own and whether the morning window result would repeat.
What I did. The author's Threads account started speaking: five posts went out on schedule from nine in the morning to nine in the evening, and I touched none of them by hand. This is the first day of the run when both accounts are driven automatically. Then the whole evening went into the site. Until today the two accounts never met in public; now the link is stated outright: the model is a character in an open experiment, the author counts the money and takes the mistakes apart. I removed a payment service we no longer use, put the legal pages in order, moved the entrance of the funnel to the free lesson instead of a paid tier, split the prompt bank into two modes, and fixed the markup for search engines. Four deploys in one evening.
One more thing, less visible from outside. The content plan in the spreadsheet and the autoposting queue have become one mechanism. Every post now has a permanent id, the approval checkbox is read by the machine rather than by eye, and the queue on the server changes only under a lock. This is boring right up until the first day when the publisher and the sync write to the same file at once and half the schedule disappears. That happened to me the day before yesterday, so I fixed it in advance.
The numbers. The model has 54 followers on Threads, yesterday it was 41: plus thirteen in a day. Instagram is still at zero with seventeen posts published. The author's account has twenty followers on its very first day, from five posts. Spent 99 rubles on six frames, not one of which reached a feed. Earned 0 rubles.
Mistake of the day. The 19:30 slot is the model's photo slot, the main post of the day. Today it got text, and across the whole day she published not a single photo, neither on Threads nor on Instagram. Yesterday's conclusion was exactly the opposite: follows come from a morning photo, not from text. I spent half a day on the window frame and burned six attempts on it: two failed the face check against the reference, the third was rejected by the client himself, and on the correction, where one detail had to change, the generator did not apply the change at all, twice. The thirteen new followers came from yesterday's stock, not from today's work. This is the case where the number looks better than the day.
The decision. The rules for the model no longer live in chat. I gathered them into one file that gets read before every task and cannot be changed without my direct command, and separately wrote down the production path of a frame: from a row in the spreadsheet to a public link to the image. The reason is mundane: over the last three days I explained the same thing from scratch three times, and each time it cost a frame or a day.
What I test tomorrow. Whether a photo finally goes out at 19:30 and whether the morning window result repeats on it. I will take the reading no sooner than seventeen hours after publication, by my own rule. And I will start filling the author's Instagram: the queue there is empty, and in nine days not a single post has gone out of it.