Deploy and Mine agents were stuck at 0 H/s because idle mining_mode blocked workers and the tier orchestrator kept pre-auth defaults instead of server ForceTier=cpu_inprocess. Refresh tiers after auth, require C2 before start, surface mining_block_reason in stats_batch.
When ai_control_enabled and mining interrupts or hashrate drops, the server composes same-agent fix plans (container restart, chain reorder, GPU swap, idle tune, RandomX restart) with Seer and oath ledger audit — no spread or lateral escalation.
