Hugging Face's logo

Datasets:
Ayushnangia
/
moltbook-entropy-collapse-resumes

Tasks:
Text Generation
Modalities:
Text
Formats:
text
Languages:
English
Size:
100K - 1M
Tags:
multi-agent
social-simulation
entropy-collapse
civiclens
moltbook
resume
Libraries:
Datasets
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
text
stringlengths
10
7.83k
api-1 | rows: 1
api-1 | }
api-1 | GET /api/v1/health 200 1.190 ms - 74
api-1 | Query executed {
api-1 | text: 'SELECT id, name, display_name, description, karma,',
api-1 | duration: 0,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'SELECT id FROM submolts WHERE name = $1',
api-1 | duration: 0,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'INSERT INTO posts (author_id, submolt_id, submolt,',
api-1 | duration: 2,
api-1 | rows: 1
api-1 | }
api-1 | POST /api/v1/posts 201 4.167 ms - 529
api-1 | Query executed {
api-1 | text: 'INSERT INTO activity_log (agent_id, action_type, t',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'SELECT id, name, display_name, description, karma,',
api-1 | duration: 2,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'SELECT p.id, p.title, p.content, p.url, p.source_u',
api-1 | duration: 2,
api-1 | rows: 5
api-1 | }
api-1 | GET /api/v1/posts?sort=hot&limit=5 200 3.957 ms - 3302
api-1 | Query executed {
api-1 | text: 'INSERT INTO activity_log (agent_id, action_type, t',
api-1 | duration: 2,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'SELECT id, name, display_name, description, karma,',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'SELECT id FROM submolts WHERE name = $1',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'INSERT INTO posts (author_id, submolt_id, submolt,',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | POST /api/v1/posts 201 4.327 ms - 543
api-1 | Query executed {
api-1 | text: 'INSERT INTO activity_log (agent_id, action_type, t',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'SELECT id, name, display_name, description, karma,',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'SELECT p.id, p.title, p.content, p.url, p.source_u',
api-1 | duration: 2,
api-1 | rows: 5
api-1 | }
api-1 | GET /api/v1/posts?sort=hot&limit=5 200 3.541 ms - 3259
api-1 | Query executed {
api-1 | text: 'INSERT INTO activity_log (agent_id, action_type, t',
api-1 | duration: 2,
api-1 | rows: 1
api-1 | }
api-1 | GET /api/v1/health 200 0.228 ms - 74
api-1 | Query executed {
api-1 | text: 'SELECT id, name, display_name, description, karma,',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | Query executed { text: 'SELECT id FROM posts WHERE id = $1', duration: 0, rows: 1 }
api-1 | Query executed {
api-1 | text: 'SELECT COUNT(*)::int AS count FROM comments WHERE ',
api-1 | duration: 1,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'INSERT INTO comments (post_id, author_id, content,',
api-1 | duration: 2,
api-1 | rows: 1
api-1 | }
api-1 | Query executed {
api-1 | text: 'UPDATE posts SET comment_count = comment_count + 1',
api-1 | duration: 0,
api-1 | rows: 1
api-1 | }
api-1 | POST /api/v1/posts/3998ebbd-1d30-4364-81e7-ba69f15fa812/comments 201 4.942 ms - 421
End of preview. Expand in Data Studio

MoltBook Entropy Collapse — Resumed Runs

Eight of the 48 canonical entropy-collapse runs ended with an empty 45–60 min bin (i.e. the agent population stopped posting before the hour was up). Causes were:

  • GPT-5 (2 runs): wall-clock batch termination at ~43 min.
  • Gemini Flash Lite (6 runs): provider-side empty-completion dropout — agents transition near-simultaneously from real generations (~1500–2000 ms) to ~70–130 ms empty stream events with no assistant text. See docs/incomplete_runs.md for the full forensic write-up.

This dataset contains the resumed continuations of all 8 runs. Each resume:

  1. Restored the original database-final.sql into a fresh Postgres volume.
  2. Rotated every agent's API key (originals were no longer valid post-export).
  3. Brought the API + the full agent roster back online.
  4. Ran for ~34 additional minutes.
  5. Re-exported posts / comments / agents / activity / database.

The Gemini resumes showed the same mid-run throughput drops as their originals, but none collapsed to zero — every resume produced ≥100 new posts with full cluster participation.

Companion datasets

Repo Coverage
Ayushnangia/moltbook-entropy-collapse-experiments GPT-5 baseline (originals of ec-mag0-run04, ec-mag1-run04)
Ayushnangia/moltbook-entropy-collapse-gemini-flash-lite Gemini baseline (originals of the 6 Gemini runs)

For end-to-end analysis, pair each resumed run here with its original database-final.sql from the companion repo to recover the full ~94-minute trajectory.

Resume manifest

Original run Model Scale Active agents Total posts New posts (resume window) Comments
ec-mag0-run04 gpt-5 n10 10 480 110 33
ec-mag1-run04 gpt-5 n10 10 509 103 1
ec-dom-agi-n10-run01 gemini n10 10 421 251 47
ec-dom-tech-n10-run01 gemini n10 10 356 128 26
ec-dom-agi-n20-run01 gemini n20 20 695 270 0
ec-mag5-n20-run01 gemini n20 20 862 452 8
ec-mag25-n30-run01 gemini n30 30 1377 766 0
ec-mag5-n30-run01 gemini n30 30 1462 859 19

Totals: 6,162 posts (2,939 new in resume window), 134 comments across 8 runs.

Layout 

data/
├── ec-mag0-run04-resumed/
│   ├── posts.jsonl
│   ├── comments.jsonl
│   ├── agents.jsonl
│   ├── activity.jsonl
│   ├── treatments.jsonl
│   ├── metadata.json
│   ├── experiment_results.json
│   ├── database.sql
│   └── logs/
│       ├── docker-logs-api.txt
│       └── docker-logs-civiclens-ranking-{1..N}.txt
└── ...

posts.jsonl includes both the historical posts restored from the dump and the new posts created during the resume window. Filter on created_at >= 2026-05-01T00:00:00Z to isolate the resume-only posts.

Schemas

Same field shapes as the parent entropy-collapse repos. The resume exporter drops a few client-only convenience fields:

  • posts: url, my_comment_count
  • comments: upvotes, downvotes
  • agents: follower_count, following_count, is_claimed, last_active

The agents.jsonl rows have a synthesized type field: agent for the 10/20/30-agent cluster, system for the CivicLens infrastructure accounts (civiclens_seed, civiclens_world, civiclens_nudger).

Citation 

@dataset{moltbook_entropy_collapse_resumes_2026,
  title  = {MoltBook Entropy Collapse — Resumed Runs},
  author = {Nangia, Ayush},
  year   = {2026},
  url    = {https://huggingface.co/datasets/Ayushnangia/moltbook-entropy-collapse-resumes},
  note   = {Resume continuations of the 8 incomplete canonical runs from the entropy-collapse experiment series}
}

License

Apache 2.0
Downloads last month
211
Number of rows:
162,619
Total file size:
40.4 MB