Record failed U3 A01 and pin usage accounting fix
Why: The authorized real-model sample failed without checkpoints or a final result, and failed-child Usage was underreported in the original candidate. Evidence: Independent A01 assessment 2ac373ed; Backend accounting review d53cc969; SELF combination receipt /tmp/fast-slow-u3-r04-20261014/usage-fix-final-dev-combo.json and Backend GATE run 3079. Impact: Keep eight draft PRs and the original sample intact; pin Backend 0637547 and Documents c53662a while marking stage 7, merge, and deployment not run. Refs: U3 r0.4 implementation plan; task feature-fast-slow-u2-u3-plan
This commit is contained in:
@@ -131,9 +131,11 @@ r0.3最终零付费组合:Dev `ef6f650` 固定同一五组件上 `make test &&
|
||||
|
||||
## 规划进展与下一步
|
||||
|
||||
r0.2唯一获批A01已执行,语义质量FAIL;r0.3唯一获批A01执行质量FAIL,两个结果原样保留。r0.4书面设计、详细实施计划及 Dev `aedcc7c` 零付费实施产物已获用户阶段6确认;独立代码复审五项问题闭合。主干 U7 账户事务及治理闭合随后纳入 Dev `ffb0ff5`,Backend `549798c`;独立影响复审、该HEAD三项平台CI、PG、浏览器及预检通过。用户又单独批准一次受控A01资源;Backend 新独立标签 `37e2cb7` 已零付费 RED→GREEN,需固定新Dev HEAD并重验CI后方可调用模型。本次不合入或部署。
|
||||
r0.2唯一获批A01已执行,语义质量FAIL;r0.3唯一获批A01执行质量FAIL,两个结果原样保留。r0.4书面设计、详细实施计划及 Dev `aedcc7c` 零付费实施产物已获用户阶段6确认;独立代码复审五项问题闭合。主干 U7 账户事务及治理闭合已纳入 Dev `ffb0ff5`,Backend `549798c`。资源准备后的固定候选 Dev `d38dbdc`/Backend `37e2cb7` 的准确HEAD平台CI:Dev run3075 三项作业、Backend run3074 成功(GATE);启动前零付费预检收据 `/tmp/fast-slow-u3-r04-model-20261014/preflight-receipt.json` exit0(SELF)。用户单独批准的这一次受控A01已经执行,授权不含合入或部署。
|
||||
|
||||
下一步固定 Backend `37e2cb7` 的新 Dev 指针,在新准确HEAD通过必需CI与启动前只读预检后执行唯一获准的A01,按逐调用Usage、运行中卡片、每条短引及最终正文判质量。任何FAIL停止并保留草稿PR,不自动重跑;PASS仍须人工阶段7验收及最新HEAD CI/Review,随后才可按计划合入。既有服务不变,U4与部署未启动。
|
||||
r0.4 A01 原始命令收据 `/tmp/fast-slow-u3-r04-model-20261014/a01-run-receipt.json` exit1(SELF);独立只读复审 `/tmp/fast-slow-u3-r04-20261014/A01-independent-assessment.md` SHA256 `2ac373ed66d47aaa9cdb699d3bf77eb8a515d46956ee0321b70cc0230d79b667`(REPLAY)确认仅2个付费run/3次调用/29,062返回Usage,后台工具流未观察到 `[DONE]` 而安全失败,0个持久检查点、0个发现、无最终正文;快通道还有无依据的“剩余4项含3项依赖接口”归属推断。**真实质量FAIL;阶段7验收、合入、部署均NOT_RUN。** 停止本任务模型实例且未自动复验;原始结果、逐调用记录和SQLite未改写,20份文件复制到受限归档 `/home/weichen/.local/share/agent-system-evidence/fast-slow-u3-r04-a01-20261014/`(manifest SHA256 `881991dcab17ad2539f8b24332f962e3505ab2f248662f58283fe08daa179077`),不含受保护路由或API key。详细分析见Documents验收记录。
|
||||
|
||||
原固定候选的持久Usage漏记失败子run已返回的7,148 tokens,结果JSON只报21,914而逐调用合计29,062。Backend `61625c6` 首次零付费修复后,独立审阅者发现“先成功、后失败且后轮缺total”仍漏记;永久SQLite反例 RED 收据 `usage-fix-mixed-turn-red-diagnostic.json` exit1,修订Backend `0637547` 的 GREEN `usage-fix-mixed-turn-green.json` exit0、完整Go vet/test及零付费live-tag `usage-fix-mixed-turn-full.json` exit0,均为SELF。独立复审 `/tmp/fast-slow-u3-r04-20261014/usage-fix-final-review.md` SHA256 `d53cc96942982bb0cc421928107cdad6af0f231f29c6ffd674b1eaaf9dfaab76`(REPLAY)重放混合回合与真实SQLite持久化,没有剩余Critical/Important;Backend准确HEAD `0637547` 的[平台run3079](https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Backend/actions/runs/3079) vet-and-test、受信home-runner全部步骤成功(GATE)。Dev暂存 Backend `0637547`、Documents `c53662a` 的 `make test && make build` 收据 `/tmp/fast-slow-u3-r04-20261014/usage-fix-final-dev-combo.json` exit0(SELF);首次组合因未暂存新gitlink被严格指针门禁拒绝,失败收据 `usage-fix-dev-combo.json` 原样保留。此修复晚于付费样本,不改变历史质量结论;Dev新提交的准确HEAD平台CI仍待推送后核验。下一步建议先评审流结束边界和快通道数字归属的书面设计修订,再决定后续零付费实现;新的付费样本需单独资源批准。既有服务不变,U4与部署未启动。
|
||||
|
||||
## 合入与清理
|
||||
|
||||
|
||||
+1
-1
Submodule modules/backend updated: 37e2cb7adb...0637547cc5
+1
-1
Submodule modules/documents updated: 8e66127a76...c53662a6d8
Reference in New Issue
Block a user