A Function That Was True Only on Mondays
The engine that auto-publishes my social threads has a check called is_devlog_day. It decides whether today is a day to introduce a dev blog post as a thread. The condition was one line: true on Monday, false otherwise.
On its own that looked fine. Once a week, bundle the blog intro chain with one recent post. A clean rhythm.
The problem was on the inventory side. There were 33 posts waiting to be introduced, and the recent-post slot held one per day. At one exposure a week, cycling through 33 posts takes 33 weeks. Well over half a year.
In your own publishing pipeline, have you ever computed how many weeks it takes for the inventory to cycle once?
The Bottleneck Was the Calendar, Not the Logic
The rotation itself was healthy. The recent post is picked by a date-seeded rotation, so even on Mondays a different post comes up each week. There was no bug in the logic. That is why it took longer to notice.
The bottleneck was how often the function returned true. If it is true one day a week, the exposure rate is pinned to that one day no matter how well the rotation shuffles. The more inventory grows, the longer the tail, and nothing anywhere reported that the tail was growing. Every post I wrote went into inventory, and inventory drained at one post per week.
What Would You Do?
Would you make it true every day, or add just one more weekday?
Monday and Thursday, With Different Roles
I chose to add one day. I changed is_devlog_day to be true on Monday and Thursday.
But I did not run the same format twice. Monday keeps the intro chain plus a recent post. Thursday posts a single recent post only. Repeating the intro chain twice a week would mean readers see the same explanation twice, so Thursday became the single-item day.
The date-seeded rotation helped here. Monday and Thursday have different seeds, so the two days surface different posts. No per-weekday rotation was needed.
The 33-week tail was cut in half. The change touched two files, 18 lines added and 4 removed. I added tests for the day check and the Thursday single-post branch.
Self-Check
- Have you divided your publishing inventory by exposures per week?
- If you add more days where the day check returns true, does any day end up repeating the same format?
- If your rotation is date-seeded, is there a test confirming the added day picks a different item than the existing day?
The Honest Part
I do not know whether twice a week is the right cadence. Making it true every day was an option, and I did not compare the two. I have no data on whether a Thursday single post performs better or worse than the Monday bundle. Once the 33 posts cycle through in half the time, the rotation starts over, and the number changes again depending on how much inventory exists by then. The only thing I am sure of is the arithmetic: the tail is now half as long.