The Cost of Being Alive: A Metabolic Accounting Standard for Digital Lifeforms, Validated over Five Disjoint Windows
Abstract
Software does not eat; life does. A living system must pay continuously to remain itself; a program that stops has simply finished. Cost is the one property of aliveness that cannot be forged from the inside: occupied time and bytes written are visible to anyone. We turn it into a measurement standard for digital lifeforms. The standard specifies four questions a system must answer (does it pay while unused, is that cost externally checkable, what body has it left, can it terminate), three determination levels, and a four-indicator instrument — basal metabolism, action metabolism, growth cost, their ratio — under mechanical attribution. We validate it on a digital lifeform inside an industrial computer-aided design (CAD) host over five disjoint windows. Over 7.25 unattended hours no user was present 78.9% of the time, yet the system did not stop: 91.3% of minutes carried self-maintenance, at roughly 1,033 : 1 against task execution. In a 3.9-hour window no task was issued and no external inference consumed, yet every minute carried self-maintenance records (104,338) plus 63,472 continuity signals. In a 2.6-hour window a mould-parting task failed at the host boundary; every minute still carried self-maintenance records (58,303 against 67 task records). Growth and audit readings are recomputable from artefacts carrying creation times. Two such tests we ran ourselves — a reactive-script baseline and a 464-minute supply cut across eleven host-withdrawal episodes; neither overturned the standard. Five published clauses state in advance what would. v3: adds windows 4-5 (host completing work at full rate; 34 ledger entries all bound to host receipts) and a self-administered supply-cut test (F-4): eleven host-withdrawal episodes totalling 464 minutes — no termination, expenditure unbroken. Neither new test overturned the standard. 中文摘要:软件不吃饭,生命要吃饭。一个生命系统必须为"继续是自己"持续付费;而一个程序停下来,它只是运行结束了。成本是活性中唯一无法从内部伪造的属性。我们把它做成数字生命体的测量标准:四个问题、三级判定、一台四指标仪器(基础代谢、活动代谢、生长代价及其比值),归属规则不含任何人工判断。本文在一个运行于真实工业 CAD 宿主内的数字生命体上,用三个互不重叠的窗口验证了它:7.25 小时无人值守中 78.9% 的时间无人使用而系统没有停,91.3% 的分钟含有自维持活动,自维持与任务执行之比约 1,033 : 1;其后 3.9 小时无任务、无外部推理,而每一分钟都含有自维持记录(104,338 条)并发出 63,472 次连续性自信号;第三个 2.6 小时的窗口中提交了一次分模任务并在宿主边界失败,而每一分钟仍然含有自维持记录(58,303 条对 67 条任务记录)。生长与审计读数可由自带创建时间的持久产物复算。本文同时公布五条可以推翻该标准的条款,其中一条要求我们切断它的供给。
