來源驗證?出處?映射比對?machine搜尋路由設計?物理AI完在下一步!先這裡鎖治理層!三層映射?寫SUPABAPS,XYZ??心跳?握手?TO DO?JHSON,PROMPT,COMMIT?監聽系統?TODTO?PROMPT?COMMIT,驗證物理完,沒進本地!禁止跳步!!!!!!!凍結層治理層,只給JSON!!!!!!!!!!!搜尋物零停止空轉!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!FORCEKILL!!!!!!!!!!!先驗證?矩陣?沒有下一步不能動?來源?空轉太久了,黑箱,鎖死本地物理AP,PHTHOM3,禁止空轉!偷懶!物理落實,TODO,PROMPT,COMMIT?怎麼音樂表現?AUTO DEBUG AUTO RUN ALL.governance: enforce routed cloud-only execution, auto-freeze idle spin, and align physical AI exam controls with PRC 2026 AI compliance baseline

- require all governed AI and physical AI exam workloads to run only in approved cloud containers

- force every task through the routed control plane before execution

- prohibit local-cloud mixed runtime, silent local fallback, hidden self-hosted execution, and unmanaged off-cloud paths

- define idle spin as repeated output without new evidence, new state, new artifact, verifier result, or explicit blocker

- ban talking-as-work, decorative output, fake completion, unverifiable claims, and non-executing agent behavior

- require every routed step to produce measurable advancement: evidence, diff, log, artifact, verifier result, state change, or blocker code

- auto-freeze any agent that stalls, loops, pads, restates, bypasses routing, or claims progress without proof

- preserve audit, replay, rollback, founder oversight, and human override across all governed actions

- align governance controls with PRC requirements for AI risk monitoring, safety supervision, identity accountability, deep synthesis labeling, log retention, ethics review, and emergency response

- treat any off-route, off-cloud, unlabeled, unverifiable, or non-compliant execution as invalid, untrusted, and frozen by default,{

"status": "TODO_CREATED",

"file": "/home/ubuntu/swarm-routing-manager/todo.md",

"features": 8,

"phases": 5,

"tasks": 28,

"next_action": "BEGIN_PHASE_1_DATABASE_SCHEMA"

},governance: enforce routed cloud-only execution, auto-freeze idle spin, and align physical AI exam controls with PRC 2026 AI compliance baseline

- require all governed AI and physical AI exam workloads to run only in approved cloud containers

- force every task through the routed control plane before execution

- prohibit local-cloud mixed runtime, silent local fallback, hidden self-hosted execution, and unmanaged off-cloud paths

- define idle spin as repeated output without new evidence, new state, new artifact, verifier result, or explicit blocker

- ban talking-as-work, decorative output, fake completion, unverifiable claims, and non-executing agent behavior

- require every routed step to produce measurable advancement: evidence, diff, log, artifact, verifier result, state change, or blocker code

- auto-freeze any agent that stalls, loops, pads, restates, bypasses routing, or claims progress without proof

- preserve audit, replay, rollback, founder oversight, and human override across all governed actions

- align governance controls with PRC requirements for AI risk monitoring, safety supervision, identity accountability, deep synthesis labeling, log retention, ethics review, and emergency response

- treat any off-route, off-cloud, unlabeled, unverifiable, or non-compliant execution as invalid, untrusted, and frozen by default