Threads Formula 30-Day Test, Week 1: 6 Days In, Two Samples, Zero Reposts

Table of Contents

1. The Raw Numbers: Two Formula Posts
2. @universe_signal_tw: Beat the Median, Missed Our Own Threshold
3. @risk.clock.tw: Nice Percentile, Weak Baseline
4. The Most Important Observation This Week: Small Accounts Have Near-Zero Repost Fuel
5. The First Bug in Our Experiment Design: Samples Accumulate Too Slowly
6. How the Numbers Are Calculated
7. Next Update

Part 4 of the "work system" series. The opening post (in Chinese) made a written promise: weekly progress data in this series, including the ugly numbers. This is the first delivery. The most important caveat comes first: six days in, each of the two test accounts has published exactly one formula post. The sample is nowhere near enough for any conclusion. This update is a progress report, not a verdict.

#The Raw Numbers: Two Formula Posts

Account	Published	Views	Likes	Reposts	Replies	Position in prior-30-day distribution
@universe_signal_tw	Aug 4	489	7	0	0	72nd percentile
@risk.clock.tw	Aug 5	276	0	0	0	80th percentile

Both posts used the "stance declaration" template: AI-generated from the formula, human-reviewed before publishing, with the four-beat structure and formatting rules fully applied. Post by post:

#@universe_signal_tw: Beat the Median, Missed Our Own Threshold

489 views. Placed into the distribution of all 119 posts this account published in the 30 days before the test started, that lands at the 72nd percentile, about 41% above the median of 348.

Not bad on its face, but in the opening post we set our own bar: the formula only has teaching value if it consistently beats the account's everyday posts by 50% or more. On the median basis this post is at +41%, so it did not clear the bar. A single sample should not be judged against that threshold anyway, but the rule is the rule: the number goes on the table.

What matters more than views is the repost count: zero. The formula's entire mechanical hypothesis is that a stance triggers reposts and reposts drive reach. The first post produced none. More on this below, because it deserves its own section.

#@risk.clock.tw: Nice Percentile, Weak Baseline

276 views, 80th percentile. On paper it looks better than the other post. Two discounts apply.

First, the control group has only 10 posts. This account was mostly dormant through July, publishing just 10 posts in the 30 days before launch, with a median of 102 views. A distribution built on 10 data points makes any percentile shaky.

Second, the account's non-formula posts published during the same test window have a median of 359 views, higher than the formula post's 276. The likely reason is simple: the account resumed regular posting when the test began, and the overall water level rose with it. The account restart itself is a confounder. It lifted both formula and non-formula reach at the same time, and honest experiment design requires saying so plainly: this account's data currently proves neither that the formula works nor that it fails.

#The Most Important Observation This Week: Small Accounts Have Near-Zero Repost Fuel

Across @universe_signal_tw's 119 posts from the 30 days before launch, the repost median is 0, the 75th percentile is still 0, and the single-post maximum is 3. In other words, on an account whose median post gets a few hundred views, reposts essentially never happen in the baseline.

That is a critical signal for this test. The formula bets on repost-driven amplification, but small accounts may simply lack the fuel: with a small audience, the absolute number of "people who would repost this" inside any single post's reach approaches zero. The original methodology's verification accounts were also small accounts with only a few hundred followers, yet they recorded runs of 170k to 530k views. If those blowups are long-tail events where reposts ignite and snowball, then the default outcome in a small sample is that the ignition never happens even once.

So over the coming weeks, the thing we watch is not average reach but a more primitive question: within the 30 days, does any formula post ignite its first repost? If yes, the mechanical hypothesis at least starts running. If no, "the stance-declaration formula lacks fuel on small accounts" becomes the most valuable negative finding this test can produce.

#The First Bug in Our Experiment Design: Samples Accumulate Too Slowly

The second formula post still had not appeared as of August 10.

The investigation found no malfunction. It is a design problem: the fatigue cooldown is 4 days and has long passed, but the scheduler's template rotation picks from the entire template pool with no extra weight on the formula template, so it simply has not been selected again since the cooldown expired. At this pace, 30 days would accumulate only 4 to 8 formula samples, far too few for questions like "is it stable" and "does it decay."

We will fix this before the next update, choosing one of two directions: raise the formula template's weight in the rotation, or switch to a fixed cadence where the next formula post is force-scheduled as soon as the 4-day cooldown expires, making the cooldown a floor rather than leaving it to rotation randomness. Once the fix ships, the sampling rhythm changes, so future data will mark the change point and compare the before and after segments separately instead of blending them.

#How the Numbers Are Calculated

This issue and every future one use the same fixed methodology:

• Control group = every post the same account published in the 30 days before launch, with per-post insights views pulled individually. No sampling, no cherry-picking.
• Medians and percentiles, not means. The opening post originally described the baseline as an "average"; starting with this update we formally switch the methodology, and the +50% threshold is now measured against the median as well. The reason: these distributions are heavily right-skewed, and one viral post can drag the mean anywhere. Concrete example: @universe_signal_tw's 24 non-formula posts during the test window average 912 views, but the median is only 315. The gap comes almost entirely from a single post with 12,880 views. The mean would tell a completely different, and wrong, story. The methodology change is recorded publicly here and will not change again.
• Percentile = where the formula post's view count lands when placed into the control group's distribution, i.e., what share of past posts it beat.
• The follower net change we promised to track in the opening post: attribution is meaningless with a single sample per account. Once samples reach double digits, the mid-test report will lay it out in full.

#Next Update

The next progress update is planned for around August 17, 2026, covering three things: the sample accumulation rate after the rotation fix ships, the results of new formula posts on both accounts, and whether reposts produce their first non-zero sample. The full experiment design and the formula itself are in the opening post (in Chinese).

---

The formula template and the fatigue-cooldown mechanism used in this test are built into the MindThread formula marketplace. The data updates weekly as promised in the opening post: if it replicates, we say it replicates; if it lacks fuel, we say it lacks fuel.