neuron-technologies/neuron
2
0
Fork 0
Code Issues 22 Pull Requests 5 Actions Packages Projects Releases Wiki Activity

Graph API returns nodes with corrupted ids/labels (mojibake) — blocks safe cleanup of #71's fabricated-memory pollution; consistent with the arena use-after-free findings #72

New Issue
Open
opened 2026-07-04 15:34:29 +00:00 by tim.lingo · 0 comments
tim.lingo commented 2026-07-04 15:34:29 +00:00
Member
Copy Link
Copy Source

Follow-on evidence for #71, observed 2026-07-04 ~10:40 CT on prod soul :7770.

While attempting to locate (and, after backup, remove) the fabricated-run memory pollution from #71, GET /api/graph/nodes?limit=999999 returned 361 nodes of which 16 match the fabrication fingerprints — but the matching nodes' id and label fields contain corrupted bytes (repeating RÃô-pattern mojibake, garbage ids like qU1RÃôE{, labels bleeding fragments of other fields such as model names and system-prompt text). Cleanup was ABORTED: issuing deletes against unreliable identifiers on the live engram is not safe, so the pollution remains in place.

  • Backup of the 16 raw nodes as returned: saved locally (available on request — /tmp/poisoned-memory-backup.json on Tim's machine).
  • This looks consistent with the request-arena use-after-free documented in docs PR #4 (June corruption root cause) — string fields appear to alias freed/overwritten memory.
  • Practical consequence: #71's memory-reinforced confabulation cannot be safely cleaned client-side; the fabricated 'sandbox container' story keeps feeding new agentic runs.

Asks: (1) confirm whether the arena patches in docs PR #4 cover this read path; (2) a safe soul-side cleanup route (or guidance) for excising the #71 pollution once ids are trustworthy.

🤖 Generated with Claude Code

Follow-on evidence for #71, observed 2026-07-04 ~10:40 CT on prod soul :7770. While attempting to locate (and, after backup, remove) the fabricated-run memory pollution from #71, GET /api/graph/nodes?limit=999999 returned 361 nodes of which 16 match the fabrication fingerprints — but the matching nodes' `id` and `label` fields contain corrupted bytes (repeating `RÃô`-pattern mojibake, garbage ids like `qU1RÃôE{`, labels bleeding fragments of other fields such as model names and system-prompt text). Cleanup was ABORTED: issuing deletes against unreliable identifiers on the live engram is not safe, so the pollution remains in place. - Backup of the 16 raw nodes as returned: saved locally (available on request — /tmp/poisoned-memory-backup.json on Tim's machine). - This looks consistent with the request-arena use-after-free documented in docs PR #4 (June corruption root cause) — string fields appear to alias freed/overwritten memory. - Practical consequence: #71's memory-reinforced confabulation cannot be safely cleaned client-side; the fabricated 'sandbox container' story keeps feeding new agentic runs. Asks: (1) confirm whether the arena patches in docs PR #4 cover this read path; (2) a safe soul-side cleanup route (or guidance) for excising the #71 pollution once ids are trustworthy. 🤖 Generated with Claude Code
will.anderson was assigned by tim.lingo 2026-07-04 15:34:30 +00:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
No items
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
tim.lingo (Tim Lingo) will.anderson (Will Anderson)
No Assignees will.anderson
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: neuron-technologies/neuron#72
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?