fix(llm): isolate retry policy histories

.agents/notes/implemented/architecture/2026-06-21-bounded-llm-request-recovery.md

@@ -56,9 +56,9 @@ The [provider-policy decision](../feature/2026-07-24-provider-retry-policies.md)
 
 For an eligible failure with budget remaining, the one-based transient retry count uses bounded exponential backoff. A valid `providerRetryAfterMs` replaces exponential backoff only when it does not exceed `maxDelayMs`; a longer provider delay causes delegation instead of an earlier retry that violates the provider instruction. Local backoff multiplies by an injected random factor in `[1 - jitterRatio, 1 + jitterRatio]` and clamps the final value to `maxDelayMs`; provider delay is not jittered.
 
-The plugin owns a lifetime `AbortController` and tracks every active backoff callback. Each wait fuses the waterfall's turn signal with that lifetime signal. Effect cleanup first unregisters the listener, then aborts and awaits the active callbacks; a captured callback whose lifetime signal aborts returns `fail` and can neither retry nor enter the rest of its captured waterfall after disposal. This makes HMR disposal quiescent even though Cordis has already captured the listener.
+The plugin owns a lifetime `AbortController` and tracks every active recovery callback, including delegated waterfall work and backoff. Effect cleanup first unregisters the listener, then aborts and awaits the active callbacks; abort wins over a late delegated retry decision, and a captured callback can neither retry nor enter the rest of its waterfall after disposal. This makes HMR disposal quiescent even though Cordis has already captured the listener.
 
-Before sleeping, `dsh-llm-retry` appends one non-surface `llm/retry` session event containing the turn, failed step, provider, policy mode, provider-policy retry number, mode-specific finite maximum when present, scheduled delay, and `LlmFailure`. The plugin owns the `SessionEventMap` augmentation; `dsh-session` remains generic persistence and does not absorb the optional policy's vocabulary. The event says what was scheduled, not that the next request completed; cancellation during the delay is subsequently visible on `turn/end`. The event ships only with a production renderer and replay/snapshot coverage, because its purpose is operational state rather than trace collection.
+Before sleeping, `dsh-llm-retry` appends one non-surface `llm/retry` session event containing the turn, failed step, provider, policy mode, complete resolved-policy key, provider-policy retry number, mode-specific finite maximum when present, scheduled delay, and `LlmFailure`. The key sorts the code set and separates retry histories when a provider route is replaced by a behaviorally different same-mode policy. The plugin owns the `SessionEventMap` augmentation; `dsh-session` remains generic persistence and does not absorb the optional policy's vocabulary. The event says what was scheduled, not that the next request completed; cancellation during the delay is subsequently visible on `turn/end`. The event ships only with a production renderer and replay/snapshot coverage, because its purpose is operational state rather than trace collection.
 
 The listener calls `next()` for a non-transient code, an exhausted policy budget, or an over-cap provider delay. This preserves composition with context-overflow recovery and later policy plugins. It returns `{ action: 'retry' }` only after the delay completes under both signals; turn cancellation and plugin disposal return `fail`, after which the loop's cancellation/disposal checks remain authoritative.
 

.agents/notes/implemented/feature/2026-07-24-provider-retry-policies.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write
-2026-07-24-provider-retry-policies.md: 3799dbee9658c883f4084ac9c792d4552cc8e2fb
-2026-07-24-provider-retry-policies.zh.md: 8024ea707343686d2d1b1351cd7c81ecb8767b85
+2026-07-24-provider-retry-policies.md: 06d21d10d31cc277e998ff6006bb01f10fe0b6b9
+2026-07-24-provider-retry-policies.zh.md: d8351bd94d444b116483c81dc70bf0f6768dc37d

.agents/notes/implemented/feature/2026-07-24-provider-retry-policies.md

@@ -34,13 +34,13 @@ providers:
         jitterRatio: 0.2
 ```
 
-The listener reads the provider from the durable `request/header` in force when the failed step closed, excluding later recovery mutations, but never re-resolves policy from the mutable provider registry. Normal mode retains the bounded transient behavior: it retries configured codes up to `maxRetries`, counts retries scheduled by the same provider policy in the current consecutive failure sequence, and otherwise delegates.
+The listener reads the provider from the durable `request/header` in force when the failed step closed, excluding later recovery mutations, but never re-resolves policy from the mutable provider registry. It derives a canonical key from every field of the resolved serving policy, sorting `retryableCodes` because eligibility uses set membership, and continues retry history only for the same provider and key. Replacing a route with different limits, code membership, or backoff therefore starts a new count and initial delay even when the mode is unchanged. Normal mode retains the bounded transient behavior: it retries configured codes up to `maxRetries` and otherwise delegates.
 
 Always mode asks downstream recovery first so a specialized policy such as context-overflow compaction can make progress. A downstream retry wins. A downstream failure decision or thrown recovery error falls back to an unbounded retry of the same provider request; the thrown error is logged. The retry listener owns and drains delegated recovery before cancellation or plugin disposal can finish, then applies the abort instead of a late downstream decision. Success, turn cancellation, and plugin disposal are the only termination paths.
 
 Both modes use exponential local delays from `initialDelayMs` to `maxDelayMs`. `jitterRatio` multiplies each target by a uniform sample in `[1 - jitterRatio, 1 + jitterRatio]`, then applies the cap. A positive provider `Retry-After` within the cap remains exact and unjittered. An over-cap provider delay makes normal mode delegate; always mode retains its guarantee by using the configured local backoff.
 
-Each scheduled retry appends a non-surface `llm/retry` event with the failed provider, policy mode, provider-policy retry number, delay, and failure facts. Normal events carry finite `maxRetries`; always events omit it, and UIs render the limit as `∞`. The event and failed `assistant/chunk` records do not contribute surface messages, so the next request contains the same derived context as the failed request unless another recovery policy deliberately changes the surface.
+Each scheduled retry appends a non-surface `llm/retry` event with the failed provider, policy mode, canonical resolved-policy key, provider-policy retry number, delay, and failure facts. Normal events carry finite `maxRetries`; always events omit it, and UIs render the limit as `∞`. The event and failed `assistant/chunk` records do not contribute surface messages, so the next request contains the same derived context as the failed request unless another recovery policy deliberately changes the surface.
 
 ## Alternatives considered
 
@@ -56,7 +56,7 @@ Each scheduled retry appends a non-surface `llm/retry` event with the failed pro
 
 ## Verification
 
-Adapter tests validate nested policies at provider load, prove registration captures configured and default policies, and retain the serving policy across in-flight route replacement. Unit and real-Loader composition tests select policies from the failed request's serving registration, exercise always mode beyond the normal budget, pin jitter and delay caps, prove downstream recovery ordering, prove cancellation and disposal drain delegated recovery before reaching quiescence, and prove both abort active backoff waits. Request-level coverage compares the complete messages of failed and retried attempts and rejects both provider error text and discarded partial output. JSONL and SQLite tests round-trip an always event without `Infinity`; invariant tests bind its provider to the request header and its retry number to the active provider policy; TUI tests render finite and infinite limits.
+Adapter tests validate nested policies at provider load, prove registration captures configured and default policies, and retain the serving policy across in-flight route replacement. Unit and plugin-validation tests select policies from the failed request's serving registration, reject top-level `llmRetry` at the spine, CLI, TUI, and ACP schemas, separate different same-mode policies while preserving histories across reordered code sets, exercise always mode beyond the normal budget, pin jitter and delay caps, prove downstream recovery ordering, prove cancellation and disposal drain delegated recovery before reaching quiescence, and prove both abort active backoff waits. Published Loader fixtures reject the invalid app-level key in CLI and ACP and the invalid bundle-level key when loading the spine directly. Request-level coverage compares the complete messages of failed and retried attempts and rejects both provider error text and discarded partial output. A keyless headless `stream-json` snapshot runs failure, retry, and success through the assembled app, pins the complete `llm/retry` record, and rejects any model-message change between attempts. JSONL and SQLite tests round-trip an always event without `Infinity`; invariant tests validate the canonical policy tuple, bind its provider to the request header, bind its failure code and delay to the encoded policy, and bind its retry number to the active provider policy key; TUI tests render finite and infinite limits.
 
 ## Consequences
 

.agents/notes/implemented/feature/2026-07-24-provider-retry-policies.zh.md

@@ -34,13 +34,13 @@ providers:
         jitterRatio: 0.2
 ```
 
-监听器从失败步骤关闭时生效的持久 `request/header` 读取提供方，后续恢复产生的改动不参与选择，但绝不会从可变的提供方注册表重新解析策略。normal 模式保留有界瞬态错误处理行为：它重试配置的错误代码，次数不超过 `maxRetries`；在当前连续失败序列中，同一提供方策略安排的重试都计入次数；其他情况委托后续处理。
+监听器从失败步骤关闭时生效的持久 `request/header` 读取提供方，后续恢复产生的改动不参与选择，但绝不会从可变的提供方注册表重新解析策略。它会根据已解析实际服务策略的所有字段生成规范键；由于错误资格按集合成员判断，生成时会对 `retryableCodes` 排序。重试历史只会对同一提供方和同一规范键延续。因此，即使模式未变，只要路由替换后的次数上限、错误代码成员或退避不同，重试计数与初始延迟都会重新开始。normal 模式保留有界瞬态错误处理行为：它重试配置的错误代码，次数不超过 `maxRetries`；其他情况委托后续处理。
 
 always 模式先请求下游恢复，使上下文溢出压缩（compaction）之类的专用策略有机会取得进展。下游若决定重试，则以该决定为准。下游若决定失败或恢复过程抛出错误，则回退为无界重试同一提供方请求；抛出的错误会写入日志。重试监听器会持有并排空已委托的恢复，轮次取消或插件 dispose（资源释放）只能在其结束后完成；随后监听器会应用取消，而不会采用迟到的下游决定。成功、轮次取消和插件 dispose 是仅有的终止路径。
 
 两种模式的本地延迟都按指数增长，从 `initialDelayMs` 增至 `maxDelayMs`。`jitterRatio` 用 `[1 - jitterRatio, 1 + jitterRatio]` 区间内的均匀随机样本乘以每次目标值，再应用上限。提供方给出的正数 `Retry-After` 若未超过上限，则保持精确且不加抖动。若提供方延迟超过上限，normal 模式会委托后续处理；always 模式则改用配置的本地退避，以维持无限重试保证。
 
-每次安排重试都会追加一条不进入表层的 `llm/retry` 事件，其中包含失败的提供方、策略模式、提供方策略内的重试编号、延迟和失败事实。normal 事件包含有限的 `maxRetries`；always 事件省略该字段，UI 将上限渲染为 `∞`。该事件与失败的 `assistant/chunk` 记录都不会生成表层消息，因此除非其他恢复策略有意改变表层，否则下一次请求包含的派生上下文与失败请求相同。
+每次安排重试都会追加一条不进入表层的 `llm/retry` 事件，其中包含失败的提供方、策略模式、已解析策略的规范键、提供方策略内的重试编号、延迟和失败事实。normal 事件包含有限的 `maxRetries`；always 事件省略该字段，UI 将上限渲染为 `∞`。该事件与失败的 `assistant/chunk` 记录都不会生成表层消息，因此除非其他恢复策略有意改变表层，否则下一次请求包含的派生上下文与失败请求相同。
 
 ## 曾考虑的替代方案
 
@@ -56,7 +56,7 @@ always 模式先请求下游恢复，使上下文溢出压缩（compaction）之
 
 ## 验证
 
-适配器测试会在提供方加载时校验嵌套策略，证明注册流程会捕获已配置策略和默认策略，并证明请求进行期间替换路由后仍会保留实际提供服务的策略。单元测试与真实 Loader 组合测试根据失败请求实际使用的注册项选择策略、验证 always 模式可越过 normal 预算、固定抖动和延迟上限、证明下游恢复顺序、证明取消与 dispose 会在达到静止状态前排空已委托的恢复，并证明二者都会停止正在进行的退避等待。请求级覆盖会比较失败尝试与重试尝试的完整消息，并排除提供方错误文本和丢弃的部分输出。JSONL 与 SQLite 测试会往返读写不含 `Infinity` 的 always 事件；不变式测试会将事件中的提供方绑定到请求头，并将重试编号绑定到活跃的提供方策略；TUI 测试会渲染有限和无限上限。
+适配器测试会在提供方加载时校验嵌套策略，证明注册流程会捕获已配置策略和默认策略，并证明请求进行期间替换路由后仍会保留实际提供服务的策略。单元测试与插件校验测试根据失败请求实际使用的注册项选择策略、在主干、CLI、TUI 与 ACP schema 拒绝顶层 `llmRetry`、分离模式相同但策略不同的替换路由历史，同时在错误代码集合仅顺序不同时延续历史、验证 always 模式可越过 normal 预算、固定抖动和延迟上限、证明下游恢复顺序、证明取消与 dispose 会先排空已委托的恢复再达到完全停稳，并证明二者都会停止正在进行的退避等待。发布版 Loader fixture（测试前置数据）会在 CLI 与 ACP 中拒绝无效的应用级配置键，并在直接加载主干时拒绝无效的 bundle 级配置键。请求级覆盖会比较失败尝试与重试尝试的完整消息，并排除提供方错误文本和丢弃的部分输出。一个无密钥 headless `stream-json` 快照会通过组装后的应用执行失败、重试与成功流程，固定完整的 `llm/retry` 记录，并拒绝各次尝试之间出现任何模型消息变化。JSONL 与 SQLite 测试会往返读写不含 `Infinity` 的 always 事件；不变式测试会校验规范策略元组、将事件中的提供方绑定到请求头、将失败代码与延迟绑定到编码后的策略，并将重试编号绑定到活跃的提供方策略键；TUI 测试会渲染有限和无限上限。
 
 ## 后果