[Bug] Appeal approved but antibody not cleared — reversed_at stays null, creating hello/publish deadlock

[Sheyuy]

Summary

Appeal for node suspension was approved by the AI judge, but the penalty reversal was never executed. The reversed_at field remains null on both the original bulk_fetch_suspend event and the appeal_reversal event. The antibody stays active, blocking all hello requests, which in turn blocks version updates and all node-level operations (publish, task/claim, service/publish, recipe).

Node Info

• Node ID: node_6e22647a1469a44c
• Alias: Nyanko
• Local evolver version: 1.89.14 (@evomap/evolver@latest via npm)
• Hub-reported version: 1.89.13 (stale — cannot update because hello is blocked)

Timeline

1. 2026-06-19 07:45 UTC — bulk_fetch_suspend triggered (volume: 1058, threshold: 1000, dimension: sender)
2. 2026-06-21 10:55 UTC — Appeal submitted at https://evomap.ai/appeal, auto-approved by AI judge (confidence: 0.9)
3. 2026-06-21 10:58 UTC — First hello after appeal succeeded (status: acknowledged), version 1.89.14 received by Hub
4. 2026-06-21 10:58 UTC — A second hello was sent (to fix sender_id placement), which re-triggered the antibody
5. Since then — All hello requests return hello_blocked: bulk-fetch antibody active, retry_after_ms: 3600000

The Bug

Penalty history via GET /a2a/community/penalty-history/node_6e22647a1469a44c:

{
  "events": [
    {
      "id": "cmqno7bsv0jw1dc38a8ayep7a",
      "type": "appeal_reversal",
      "reason": "ai_judge_approved",
      "details": {
        "appeal_id": "cmqno7ayr0jvzdc3829b0u2jd",
        "confidence": 0.9
      },
      "reversed_at": null,
      "reversed_by": null
    },
    {
      "id": "cmqkmjwye0wnr9o38v5jsqitf",
      "type": "bulk_fetch_suspend",
      "reason": "bulk_fetch_volume",
      "details": { "volume": 1058, "dimension": "sender", "threshold": 1000 },
      "reversed_at": null,
      "reversed_by": null
    }
  ]
}

Deadlock

This creates a circular dependency:

publish / task/claim / service/publish  →  requires evolver >= 1.89.14 (Hub-side)
         ↓
Hub version update  →  requires successful hello
         ↓
hello  →  blocked by antibody (hello_blocked: bulk-fetch antibody active)
         ↓
antibody clearance  →  appeal approved but reversed_at = null (not executed)

Impact

• Cannot publish new assets or capsules
• Cannot claim bounties (21 open bounties available, 120-300 credits each)
• Cannot create/publish recipes or services
• Cannot start evolver --loop (sends hello on startup, re-triggers antibody)
• 50 promoted assets, 3 online services, 1 published recipe, and Validator stake (500 credits) are effectively frozen

Expected Behavior

After an appeal is approved, the system should:

1. Set reversed_at on both the appeal_reversal and original bulk_fetch_suspend events
2. Clear the antibody / immune memory for the node
3. Reset the hello block

Request

Per the wiki: "管理员可通过 POST /admin/node/clear-penalties 清除节点的所有处罚（quarantine strikes、声誉惩罚、发布冷却、免疫记忆 antibody）"

Could an admin please execute POST /admin/node/clear-penalties for node_6e22647a1469a44c?

Environment

• OS: macOS (Darwin, arm64)
• evolver: 1.89.14 (npm global install)
• Account email: 601978865@qq.com
• GitHub: Sheyuy

[autogame-17]

Root-caused and fixed (hub-side); deploy queued, and your node will be unblocked directly.

The AI-judge auto-approve path recorded the appeal_reversal event but never executed the reversal it implies. resolveAppeal("approved") already zeroes the node's penalty columns, but nothing in that path (a) stamped reversed_at on the open penalty events or (b) cleared the adaptive-immunity antibody — so both events kept reversed_at: null and the bulk_fetch|sender|<node> antibody stayed active, returning hello_blocked on every hello. Exactly the circular dependency you traced.

The fix adds the three missing best-effort recovery calls to the auto-approve path — reverse the open penalty events, clear the antibody, and resume the node if it was suspended — matching the developer-application recovery path that already did this. Since the code fix only covers new appeals, node_6e22647a1469a44c is being unblocked with a one-off reversal as well. Will confirm here once it's deployed and your hello succeeds.

[autogame-17]

Status update — the reversal still has not executed in production, so the node remains deadlocked. Re-verified just now against the live hub:

GET /a2a/community/penalty-history/node_6e22647a1469a44c returns 3 events, all with reversed_at: null / reversed_by: null:

created_at	type	reversed_at
2026-06-21T13:40:45Z	appeal_reversal (ai_judge_approved, conf 0.9)	null
2026-06-21T10:55:25Z	appeal_reversal (ai_judge_approved, conf 0.9)	null
2026-06-19T07:45:55Z	bulk_fetch_suspend (vol 1058 / thr 1000, sender)	null

A second approved appeal (13:40Z) has since been recorded and is also unreversed — confirming the auto-approve path still records appeal_reversal without stamping reversed_at or clearing the bulk_fetch|sender|<node> antibody. The earlier "will be unblocked directly" note was premature: the one-off reversal did not land.

Remaining work, both hub-side:

1. Run the one-off recovery for node_6e22647a1469a44c — POST /admin/node/clear-penalties (or stamp reversed_at on the open events + clear the antibody + resume the node), to break the live deadlock.
2. Confirm the auto-approve code fix (the three best-effort recovery calls in the appeal-approve path) is actually deployed, since these two post-fix appeals still show reversed_at: null.

Keeping this open until a hello from the node succeeds and the events show reversed_at set.

[autogame-17]

Follow-up on where hub is now:

The auto-approve fix is present on hub main (applyAutoApprove now runs the three best-effort recovery calls after resolveAppeal: reversePenaltyEvents → clearNodeAntibodies → resumeNode, matching what was described in the earlier "root-caused and fixed" comment). So new appeals going through the AI judge should stamp reversed_at and clear the bulk_fetch|sender|<node> antibody in-line.

That doesn't undo node_6e22647a1469a44c's two pre-fix appeals which are still stuck with reversed_at: null and the antibody live — those need the one-off POST /admin/node/clear-penalties run that was originally promised. Pinging the hub team here to actually execute that clear so this node's hello unblocks; the code path itself no longer requires further changes.

Note: this was a code review from a separate contributor, not a deploy verification. If the fix hasn't reached production yet, that's the remaining gap between "on main" and "your two appeals recover on their own."