Introduction
In January 2026, an anonymous account called "Aozora News" collected massive reposts on Bluesky, drawing attention as a "diffusion power test" in the Japanese-speaking community. As of January 25, 2026, the post has reached approximately 7,700 reposts. Yet this number has a critical gap: it contains no structural information about who reposted to whom, or how information propagated through the network.
7,700 shows quantity. It does not show shape.
This experiment aims to fill that gap. By adopting a path-tracking design, we visualize the structure of information diffusion in Japanese Bluesky. This serves as a sequel to "Network Perception in ATmosphere" (https://plurality.leaflet.pub/3mbdfcihvhs2u), providing empirical data to support the theoretical framework.
The question is simple: How does information spread on Bluesky?
Methods
On January 23, 2026, at 19:00 JST, a post was published on Bluesky with the following instruction: "If you repost this, reply with the name of the person whose repost you saw, followed by 'Udon Computer Legend.'"
The phrase "Udon Computer Legend" served as a tracking keyword and carried no meaning. Its meaninglessness was intentional—to eliminate search noise and collect only replies from experiment participants.
The observation period was 48 hours (January 23, 19:00 – January 25, 19:00 JST). Replies were collected using Bluesky API's app.bsky.feed.getPostThread, yielding 143 replies.
From the collected replies, source attributions were extracted and structured as a directed graph. Nodes represent participants; edges represent the direction of information flow (parent → child). 132 edges were successfully parsed; 10 replies lacked source attribution (e.g., posting only "Udon Computer Legend").
At the start of the experiment, the posting account (Nighthaven) had approximately 2,700 followers. No additional reposts were made to resurface the original post during the observation period. We judged that observing natural decay without intervention had value.
Note: The original post mistakenly stated "January 24" instead of "January 23," corrected immediately after posting. This correction post is included in the 143 replies.
Results
Basic Numbers
Total replies: 143
Parseable edges: 132
Maximum chain depth: 7
Temporal Distribution
Diffusion concentrated in the initial phase.
Day 1 (Jan 23): 117 replies (82%)
Day 2 (Jan 24): 25 replies (17%)
Day 3 (Jan 25): 1 reply (1%)
The first reply arrived approximately 3 minutes after posting (09:57:17 UTC); the last reply came about 5 hours before the deadline (05:26:43 UTC). Roughly 50% of all replies arrived within the first 2 hours, followed by a clear decay curve after 24 hours.
Network Structure
The structure was tree-shaped (directed arborescence). No exponential explosion occurred. The pattern was radial (star-shaped) with localized linear chains.
Top Hubs by Out-Degree:
Direct diffusion from Nighthaven accounted for 63 edges (approximately 48% of total). Chains through secondary hubs were limited.
Special Pathways
Several special pathways were observed:
For You feed entry: Multiple participants reported discovering the post through the For You feed. Algorithmic recommendation contributed to diffusion.
Original post visibility via reply: One participant (banachan) reported: "Shirokuma's reply surfaced Nighthaven's post—I saw the original post, not a repost." This suggests replies can increase the visibility of the original post.
Hub formation after visualization: tomo-x posted a network visualization mid-experiment, then propagated to 4 additional participants. Meta-observation activity accelerated diffusion.
Discussion
Diffusion Characteristics of Bluesky
The results indicate that diffusion in Japanese Bluesky is not "viral."
On X, algorithms create "buzz." A single post reaching tens of thousands of reposts is common. On Bluesky, however, diffusion power depends heavily on the originating account's follower count. Users form "islands," and cross-island diffusion is difficult.
In this experiment, starting from Nighthaven's 2,700 followers, chains extended up to 7 levels, but total participants remained at 143. Compared to Aozora News's 7,700 reposts, the scale is small—but qualitatively different in possessing structural information.
This structure resembles the blogosphere of the late 2000s or mixi. Bluesky should be viewed not as an X replacement, but as a different ecosystem.
Diffusion and Conversation
Bluesky advocates for "better conversations, not louder ones" (October 31, 2025 official blog). This experiment forced "conversation" through a reply-required design.
As a result, the participation threshold rose, and the sample was filtered to "those who actively cooperate." As Sugisaki pointed out, this represents diffusion through "goodwill cooperation," which differs in motivation structure from natural "what's this, interesting" diffusion.
However, this design enabled path information capture. The "shape" invisible in repost counts alone became visible. Conversation carries psychological cost, but that cost can also guarantee information quality.
A Structure Resistant to Chain Explosions
Chain explosions are unlikely on Bluesky. Whether this is good or bad depends on what you seek.
It's unsuitable for affiliate marketing. Monetizing buzz is difficult on Bluesky. Conversely, it's suitable for community formation. Information flow based on trust within islands creates a low-noise environment.
Bluesky is "an SNS that doesn't go viral"—and this is likely by design. The mission to "reclaim conversation" includes the choice to sacrifice diffusion power.
Limitations
This experiment has several limitations:
Sample size: 143 is statistically small; generalization requires caution.
Sampling bias: The reply requirement raised the participation threshold, making it harder for those avoiding interaction outside their follow/follower circle to participate.
Motivation structure difference: This was "goodwill cooperation" diffusion, qualitatively different from natural diffusion.
Origin-dependent: Results depend on Nighthaven's follower scale (2,700) and cluster characteristics. Different origins may yield different structures.
Date error: The original post mistakenly stated "1/24," later corrected. This confusion may have affected participation rates.
Conclusion
Diffusion in Japanese Bluesky has a tree structure dependent on a small number of hubs. No exponential explosion occurs. The first 2 hours determine most of the structure.
Small-scale data with path information carries more structural information than large-scale data without it. 7,700 reposts is an impressive number, but a directed graph with 132 edges tells us more about the "shape" of diffusion.
Bluesky is "an SNS that doesn't go viral." And that is probably intentional.
Acknowledgments
Thanks to the 143 participants in this experiment.
Special thanks to Sugisaki for the sharp observation on motivation structure differences, tomo-x for implementing API-based visualization and publishing on GitHub (https://github.com/tomo-x7/bsky-repost-network), and l-tan and Afternoon Curry for technical discussions.
References
Nighthaven (2025). Network Perception in ATmosphere. Leaflet. https://plurality.leaflet.pub/3mbdf4fzkac2u
Bluesky (2025). Building Bluesky: Toward Better Conversations. https://bsky.social/about/blog