SivanCola
9799eae998
Interrupted provider streams can reconstruct positive usage without terminal reported counts. The shared quote owner forwarded that provenance but ignored it when certifying complete cost facts, allowing recovered usage to make a fully covered ledger. Use the existing incomplete-usage result while retaining fixed-point amounts and occurrence-time valuations. Reported usage, including rate-card estimates, keeps its completeness. Existing quote fields, reasons and ledger versions stay unchanged; no provider-visible request, permission or budget default changes. Cover shared event/config adapters, persisted ledger coverage and real boot SSE recovery with an actual file read. Add a Windows fourteen-event inventory in the existing job. This is a monetary-fact prerequisite of the full concurrency plan, not shared root-money admission or hard reservation. Verification: fourteen focused events and the real boot race pass; compiled preceding pricing fails twelve regression events with two controls passing. The prior reader accepts the incomplete quote fixture. Root vet, full lint and make check pass with the baseline unchanged. Eight Windows/Linux amd64 owner builds compile; full root inventory and complete owner race continue separately. Refs #14025 Refs #14133
5.1 KiB
owner: @SivanCola backup: @esengine status: active reviewed: 2026-10-26
Budget usage completeness / 预算用量完整性
Shared quote/ledger provenance is covered in Estimated usage and quote completeness. 共享报价及账本来源的后续边界见该双语说明。
RunBudgetTotals.Priced requires reported usage and a price table for every
observed round. A missing record or an existing Usage.Estimated record keeps
the turn and continuing task incomplete. Cost retains only their known priced
subtotal; unknown spend is not zero spend.
RunBudgetTotals.Priced 要求每个已观察轮次都有供应商用量及价格表。缺失记录或
已有的 Usage.Estimated 记录会让 turn 和继续执行的 task 保持成本不完整状态。
Cost 仅保留已知且已定价的部分；未知费用不等于零费用。
Producer and observer / 生产者与观察者
The shared HTTP-attempt helper preserves the exact request count when the
provider returns no usage, and marks that synthesized record with the existing
Estimated flag. It does not invent token counts or change a provider-reported
zero record. Existing estimated records remain estimated; inputs are copied.
共享 HTTP attempt helper 在供应商没有返回用量时保留准确的请求次数，并通过已有
Estimated 标记说明补出的记录缺少供应商计量。它不编造 token 数，也不改写供应商
明确报告的零用量。原有估算记录仍是估算；输入记录通过复制保留。
The Agent budget observer counts available requests and tokens as before, but
excludes estimated or unpriced costs from its known subtotal. Missing rounds
remain visible in Rounds. An aggregate marked estimated is excluded as a
whole; this layer cannot recover its exact partial charges.
A wholly unknown cost cannot trip the money gate. An already-known subtotal reaching the configured limit still triggers the existing soft landing even if other rounds are unknown.
Agent 预算观察者仍统计可用的请求次数及 token，但不会把估算或未定价费用加入已知
金额。缺失轮次仍计入 Rounds。标记为估算的聚合记录整体排除；这一层无法恢复其中
准确的部分费用。
完全未知的费用不会触发金额门禁；已知金额达到 显式阈值时，即使另有未知轮次，也仍会触发既有的软收尾。
Lifetime and compatibility / 寿命与兼容
The incomplete marker belongs to the existing turn/task budget. A continuation keeps task uncertainty while its own fully reported turn can be priced. A new task or explicit task-budget reset drops the old task reading. This does not repair historical totals or infer whether an unreported request was billed.
完整性标记归既有 turn/task 预算所有。继续执行会保留 task 的未知状态，但新 turn 若报告完整，仍可单独标记为已定价。新任务或显式重置预算会丢弃旧 task 读数。 这不会追溯修正历史统计，也不会推断未报告用量的请求是否产生了账单。
No field, serialized shape, configuration, tool schema or permission changes.
Existing readers already accept Estimated and Priced; old records missing
those flags retain their old zero values. Request-only records now say they
are incomplete, and estimated costs no longer certify the Agent money gate.
没有新增字段、序列化形状、配置、工具 schema 或权限。既有读取方已支持
Estimated 和 Priced；历史记录缺少这些标记时仍使用旧零值。仅含请求次数的
新记录现在明确其不完整性，估算金额也不再成为 Agent 金额门禁的确定依据。
Evidence and limits / 验证与边界
Real boot.Build tests use the OpenAI HTTP adapter, one actual fixture read and
two SSE responses, omitting usage from either response or neither. Agent tests
cover explicit zero usage, known-cost landing, unknown estimates and task lifetime.
The continuation fixture selects the existing host continuation flag; it is not a frontend interaction test.
真实 boot.Build 测试经过 OpenAI HTTP adapter，完成一次夹具读取及两次 SSE
响应，分别省略第一轮、末轮或不省略用量。Agent 测试覆盖明确零用量、已知费用
收尾、未知估算及任务寿命。continuation 夹具选择既有 host continuation 标记，
不冒充前端交互测试。
This is honest local accounting, not a root money reservation or hard cap. In-flight work and the existing finalization round can exceed a soft limit.
Root currency aggregation, auxiliary-call budget admission, durable accounting and physical process cancellation remain separate work. Provider-visible prompt prefixes and default limits are unchanged.
This author assessment is not an independent review.
这是本地计账完整性修复，不是根任务金额预留或硬上限。进行中的工作及既有收尾轮 可能超过软阈值。根任务货币汇总、辅助调用预算准入、持久计账及物理进程取消仍属 后续工作。供应商可见的提示前缀及默认阈值不变。作者评估不等于独立审查。