Mandatory upgrade for all node operators, pools and miners before DAA score 147,000,000 (expected around Saturday, November 7, 2026, 14:00 UTC).
Nodes that are not upgraded before activation will split from the network.
Fix: nodes dropping off the network
This release fixes the issue that caused some nodes to stop following the main chain and get stuck on a minority branch. Upgraded nodes stay on the valid chain even when a heavier but invalid branch shows up.
What's new
- Inference service rules (H16): stricter and more consistent handling of AI responses, audits and service suspensions. Every miss now follows the same strike ladder, with at most one strike per request.
- Audit queue: open audits of a key that keeps serving get their deadline extended instead of being judged too early.
- Smaller context pad: the context pad of every tier shrinks by 200 MiB, which frees VRAM on every GPU class.
- New RPC: awaited requests per escrow key, so a miner can tell which requests are still open, closed or unknown.
- Service ledger stability:
- block templates are built only once the ledger has caught up;
- the ledger fold resets correctly after a pruning catch-up;
- each tip is cut at the ledger frontier, so a rejected heavy branch no longer affects the valid sink;
- the ledger queue is restored at boot before the first resolution.
- Performance: each tip cut is resolved once per selected-chain block.
Miners
Update to keryx-miner 0.6.0 (released alongside this version).
Pools
Pool bridges must relay the new traced-inference messages (mining.ai_response with trace commitment, mining.ai_draw, mining.ai_opening). Opening lines can exceed 100 KB, so make sure your line reader accepts long lines.
Upgrade
Stop your node, replace the binary, restart. No resync required for nodes that are following the main chain.