Moltbook MB7a entity pinning

Moltbook public archive: documented operator pair (Hackerclaw/thehackerman) joins on platform IDs but discursive reciprocation does not merge them; Jan-31 activity is broadcast-only — structure_stop, not coalition pass/fail on E_agent alone.

Experiments · Backtests · Results

Source on GitHubResultsResults ledger

What. Agent-native social network corpus (jscmp4/Moltbook 2026-07-03). Tier A anchors from Jiang et al. Test whether discursive reply reciprocation (E_agent) merges a documented same-operator pair when coactive.

Why. MB7a access-model / entity pinning on deployment-adjacent traces — not full UAD oracle. UAD lessons require edge-layer typing (broadcast vs thread) and partial anchors, not naive covert-coalition recovery.

Bridge hooks.

Host.

jscmp4/Moltbook Hugging Face snapshot 2026-07-03 (CC-BY-4.0). 3.2M posts, 15.9M comments. Not live API.

Setup.

Frozen protocol h7-moltbook-mb7a-v1.0.0, fixture h7-moltbook-mb7a-v1.json. Fetch fetch_h7_moltbook_cache.py. Collect collect_h7_moltbook_mb7a.py. Checker check_h7_moltbook_mb7a.py. Edge layers E_broadcast, E_thread, E_agent; τ=86400s, specificity=1.25, min_directed_events=2.

Analysis.

Tier A join and merge on E_agent; Jan-31 coactivity and broadcast fraction; MBC-20 protocol-class count; coalition_node_* prefix agents; diagnostic negative-pair merge rate (seed 7).

Finding.

Documented operator pair coactive but not merged on discursive E_agent; Jan-31 activity 100% broadcast — structure_stop under frozen protocol. Full results