Crash Resume, the Budget Circuit Breaker and Governance Decay
Overnight the power drops, the process gets killed, the money burns through, and the rules you set at the start quietly get forgotten. Today adds three things: resuming after a hard kill, halting the moment the budget is blown, and keeping the constraints alive after compaction. That last one is a real risk somebody has already measured.
Today's Goals
- Can explain how resuming from a checkpoint differs from starting over, and say what a checkpoint must record for a resume to avoid redoing work
- Can explain why the budget circuit breaker has to decide during the run, and say how it divides labor with after-the-fact cost measurement
- Can explain what governance decay is, why probing with early facts cannot detect it, and how restating the constraints closes that hole
D5 ended runs cleanly: the harness decided to stop, and it still had a chance to tidy up. Today handles three messier endings. When you finish, scroll back up and tick off the three goals.
Plain-Language Walkthrough
Three things the night shift runs into
Same production line as always. The whiteboard from D2 is up, the handover document from D3 is written, the sign-off sheet from D4 gets signed line by line, and D5 put an inspector on the floor. Now let it actually run through a whole night. What else shows up?
The power drops. Not somebody deciding to stop, but a breaker tripping. No handover, no tidying up, not a word.
The money burns. The machines keep turning and the material keeps going in, but by morning barely anything got built. Cost accrues by the hour and progress accrues by the item, and those two can come apart completely.
The rules fade. Six handovers later, nobody repeats the line about wearing eye protection before cutting. Nobody decided to ignore it. The sentence simply got dropped somewhere along the chain of retellings.
Three things, three modules today. The third is the least visible and the most worth learning, and it has a name: governance decay.
Interview questions
An agent process that had been running for six hours gets killed - how should it come back on the next start?一个跑了六小时的 Agent 进程被杀了。你希望它下次启动时怎么恢复?
Common in ChinaCommon overseasDeep dive#crash-resume#checkpoint#idempotence🔒 Unlock this chapter to see the answer keys.
Are a budget circuit breaker and a post-hoc cost report two different things, and what does each solve?预算熔断和事后成本报表是两件事吗?分别解决什么问题?
Common in ChinaCommon overseasIntermediate#budget#circuit-breaker#real-time-control🔒 Unlock this chapter to see the answer keys.
Repeated context compaction makes an agent drift away from its original safety constraints - how would you defend against that?反复压缩上下文会让 Agent 逐渐不遵守最初的安全约束。你会怎么防?
Common in ChinaCommon overseasDeep dive#governance-decay#constraint-pinning#compaction🔒 Unlock this chapter to see the answer keys.
You already have probes that check factual fidelity after compaction - why do they miss constraints being erased?你已经有一套压缩后的事实校验探针了,为什么它发现不了约束被擦掉?
Common in ChinaCommon overseasDeep dive#probe-blind-spot#constraint-fidelity#capability-boundary🔒 Unlock this chapter to see the answer keys.
