Threads Formula 30-Day Test, Week 3: Clean Samples Converge to −15%, Reposts Still Zero
Table of Contents
- The Raw Numbers: Two Clean New Samples
- The Promised Fix Shipped, and the First Interval Looks Right
- How to Read the Convergence: Both Rulers on the Table
- Reposts: Day 20, Eight Formula Posts, Zero Total
- Owning Up: risk Contributed Zero Samples This Week
- Account-Level Numbers
- How the Numbers Are Computed
- Next Update
Part 6 of the "work system" series. The opening post (in Chinese) promised weekly progress data, including the ugly numbers. This is the third delivery, and the headline comes first: the scheduler fix promised last week shipped on Aug 17, this week's two formula posts are the first clean samples under it, landing at −15% and −23% against the baseline median, a clear convergence from last week's −56% but still short of the median; and formula posts remain at zero reposts, all of them, on day 20 of the run. There is also one execution delay to own up to.
The Raw Numbers: Two Clean New Samples
| Account | Published | Views | Likes | Reposts | vs. baseline median | Percentile |
|---|---|---|---|---|---|---|
| @universe_signal_tw | Aug 19, 07:32 | 312 | 8 | 0 | −15% | 44th |
| @universe_signal_tw | Aug 23, 13:31 | 284 | 6 | 0 | −23% | 38th |
Baseline rules unchanged: every post from the same account in the 30 days before launch (121 posts), with this week's re-snapshot putting the median at 367 (366 last week; views accrue naturally, so the baseline drifts). The two posts land at the 44th and 38th percentiles, against the 18th for last week's only clean sample. A real step toward the median, but remember where the bar sits: the opening post demands a consistent +50% before the formula counts as worth teaching, and these have not yet cleared the median itself.
The Promised Fix Shipped, and the First Interval Looks Right
Last week's two stacked mechanisms (restock counting formula posts as stock, and the scheduler publishing the first item of an all-cooling queue anyway) were fixed in the early hours of Aug 17: restock no longer counts formula posts toward stock level, and a queue holding nothing but cooling formula posts now skips the slot. Last week we said the ship date would be noted in this update; noted.
The evidence so far is exactly one interval, but it is the right one: this week's two formula posts published on Aug 19 and Aug 23, exactly 4 days apart. Cooldown worked as designed, no back-to-back firing. These are the first fully normal-cadence samples since the fix, with no discount to apply, which is exactly why this week's convergence deserves to be taken seriously. The other half of the fix (skipping slots on a formula-only queue) was never triggered this week; its full verification comes next update.
How to Read the Convergence: Both Rulers on the Table
Comparing against the fixed baseline alone is not enough, because universe had a strong week overall: 26 everyday posts in the same window with a median of 437 (390 last week), a single post peaking at 7,148 views, and +25 followers in one week. When the whole account rises, formula posts float up with it.
So here is the second ruler, same-window everyday posts: the formula posts sit at −29% and −35% against this week's everyday median. Last week's clean sample measured −59% on the same ruler. In other words, the convergence holds on both rulers: from the 18th percentile up to the 44th and 38th against the fixed baseline, and from −59% to −29% and −35% against the same-window field.
But two samples cannot carry a trend claim. All this week earns is: the direction stopped getting worse for the first time, and the +50% teaching-value bar is still a river away.
Reposts: Day 20, Eight Formula Posts, Zero Total
This is the ugliest and most important number of the week. The 8 formula posts published to date still hold zero reposts, combined.
In the same window, universe's everyday science posts collected 13 reposts plus 5 quotes. The fuel is not just present, it was unusually abundant this week. The stance formula's core hypothesis is that a stance triggers reposts and reposts drive reach; on day 20 of the run, that mechanism has not fired once. Last week's line was "the fuel is there; the fire did not start." This week: more fuel, still no fire.
Owning Up: risk Contributed Zero Samples This Week
Last week we wrote, in black and white, that risk's formula posts were all spent and "a new batch will be queued before the next update, running through normal rotation." In reality, that restock happened today (Aug 24), after this week's data pull. No technical blocker; an execution delay in our own operations. The result: risk added zero samples this week, and with only 10 days left in the 30-day window, it can accumulate at most 2 or 3 more. Wrong is wrong; this one is on us.
The restock is done: 3 posts for risk, 3 for universe (whose 2 queued posts both rotated out this week), all through normal rotation.
One aside worth flagging: risk's everyday content is generated day by day, so its queue sits near empty, and this restock recreates the "nearly formula-only queue" composition, the exact trigger condition of last week's bug. This time the Aug 17 fix takes the hit: by design it should skip slots instead of firing back-to-back. Next update we verify its actual behavior.
Account-Level Numbers
Followers: risk 1,509 → 1,506 (−3), universe 282 → 307 (+25). The +25 is universe's largest single-week move across the three weeks of the test, which is also all the history there is to compare. Aug 17 and 18 happened to produce several high-view science posts and the timing lines up, but we have no day-by-day follower series, so that causality stays unclaimed. Same-window everyday medians: risk 218 (267 last week), a soft week; universe 437 (390 last week), clearly hot. Both accounts published every day on schedule; the pipeline is healthy.
How the Numbers Are Computed
Same rules as the previous two weeks: control group = every post from the same account in the 30 days before launch, insights pulled post by post; medians and percentiles, never means; the baseline re-snapshotted this week (universe 121 posts, median 367; risk 22 posts, median 121) from the version-controlled script. This week's window runs from the early hours of Aug 17 (last week's pull) to the Aug 24 pull.
Next Update
The next update is planned for around August 31. Three things to watch: the first samples from risk's new batch (through normal rotation); the fixed scheduler's actual behavior under the "formula-only queue" condition that triggered the original bug; and whether any formula post finally records its first nonzero repost.
One more thing that deserves an accounting: the opening post's third question (how AI-generated, human-reviewed formula posts compare against fully handwritten ones) has not made it into any weekly update, because this entire test runs on the AI-generated leg only, with no parallel handwritten control group. That question gets answered head-on in the wrap-up, including the "why was there no control group" part.
The 30-day window closes on Sep 3, with a full wrap-up to follow. The full experiment design and formula text are in the opening post.
The formula templates and fatigue-cooldown mechanics used in this test are built into the MindThread formula marketplace. Data updates weekly as promised in the opening post: if it replicates we say so, if the fuel is missing we say so, and if the delay is ours we say that too.