lili
|
455d45be88
|
docs(tier2): 收敛环终局落档——r4 4/5 收敛·decide_n5=KEEP_留观·fable 终判 go(留观=水果;Δ3 梯子=M3 保持中档默认);板对账+§13 三条蒸馏(双源默认影蔽/worktree 基线必验/pgrep 探针自匹配)
contract-gates / contract-gates (push) Has been cancelled
docs-gate / docs-gate (push) Has been cancelled
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
2026-07-04 06:52:39 -07:00 |
|
lili
|
dd2e60c4a6
|
docs(tier2): r3 双源默认值影蔽发现落档(writer 实仍 40/美食靠 step 顶让量/F-3 同病对照有效/糖水店 420s 仍超时记观测)
contract-gates / contract-gates (push) Has been cancelled
docs-gate / docs-gate (push) Has been cancelled
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
2026-07-04 03:27:38 -07:00 |
|
lili
|
5cdea9320c
|
docs(tier2): R3 结果与 fable 中期终审落档——产品口径 3/5(美食阈值修后走通)/咖啡 finished 口径拆解引出早收敛探测观测项/R2_补骨架出口 n=1 归因不采信/进 F-3 强档对照
contract-gates / contract-gates (push) Has been cancelled
docs-gate / docs-gate (push) Has been cancelled
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
2026-07-04 03:22:01 -07:00 |
|
lili
|
2e4dd56b76
|
docs(tier2): F-2 执行记录落档(R1 3/5 归因推翻分散判/修一 420s/R2 分层证据后奉令中止/修二阈值批/R3 在跑/五路工单并行)+ F-3 解锁口径(glm-5.2 在列·独立台账防轮污染)
contract-gates / contract-gates (push) Has been cancelled
docs-gate / docs-gate (push) Has been cancelled
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
2026-07-04 00:20:38 -07:00 |
|
lili
|
4d220e77da
|
docs(轨2+板): 质量模型与回流环两设计档(双评审已修·决策包⑦已批定稿)+ 工单板对账(阶段一②收口/决策②⑥⑦落账/R6+rubric落库前置工单/在飞板/06-29§7)
docs-gate / docs-gate (push) Has been cancelled
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
2026-07-02 19:31:53 -07:00 |
|
lili
|
bc1efb26d8
|
docs(tier2): 纠正 U5 退路层 —— win-balance 在 agent 层非工具层(创始人纠正)
经济/数值平衡是 agent 玩法职责,基础工具层做静态模板/补骨架治不了(静态验证调不动动态平衡 +
每游戏数值会改、预制浪费)。真杠杆 = verdict 反馈质量(让 M3 像 spike 自调平衡)+ 模型能力(换档)。
撤回工具层加固 + 精确率测量(率随 agent 演进会变)。go/no-go=conditional,卡 agent 层、随其演进改善。
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
2026-06-28 21:05:22 -07:00 |
|
lili
|
55a213e722
|
docs(tier2): 切片二 U5 收敛环 go/no-go 结论 = conditional go(留观+针对性加固)
真跑(M3-only,chrome 泄漏修后 iso 逐款):M3 写得出结构完整多系统富游戏(过大多数九门+三联动/latch),
但收敛 run-dependent(随机)、confounded pass 率 ~1-2/5;失败集中「进度/经济平衡门」
(H_progress + 经济门 profitPathWin)。退路 = win-balance 模板化/补骨架(非换档,M3 有能力)。
第二 confound(live worker:9501/@8200 争 4 核)未除,真率需暂停 live 干净跑测。
非 no-go(能过全门)、非干净 go(收敛太不稳)。
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
2026-06-28 20:10:53 -07:00 |
|
lili
|
8d4654ab63
|
fix(tier2): serve-and-play chrome cleanup 杀整树(治泄漏)+ 切片二 U5 执行发现
chrome 泄漏根因:cleanup() 只 kill $CHROME_PID(父)→ renderer/gpu 子进程 reparent 给 init 成孤儿、
跨款累积(实测 U5 Round1 撞 wall_timeout、杀批组仍残留)。修=cleanup 加 pkill -f "$USER_DATA_DIR"
杀本实例整棵 chrome 树(user-data-dir mktemp 唯一、不误伤别的 chrome/live worker)。
plan 执行发现:① env 闸缺口(fresh worktree 无 node_modules/esbuild → build 失败 → M3 假性不收敛,
已补 U1 S0 + Risks);② M3 已证能收敛(smoke 面包+水果 clean accept、九门全过);
③ U5 Round1 被 confound(chrome 泄漏 + live worker:9501/@8200 争 4 核)、需隔离干净重跑。
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
2026-06-28 16:54:17 -07:00 |
|
lili
|
4f444e0ca0
|
docs(tier2): 切片二 n=5 收敛环 go/no-go 细化执行 plan(Codex+Opus 双评审收口)
plan① 切片二的可跑 runbook 细化:U1 前置阻断(复现 smoke+反向 smoke+DockerWorkspace 小验)/
U2 批级+跨轮护栏 fail-closed / U3 退路树 n=5 适配 / U4 play-scene 门结构化 / U5 跑环+go/no-go。
执行决策(创始人):串行+M3-only;n=5 绝不 n≥30。§6.8 双评审 19 项发现全核验、全文档内修。
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
2026-06-28 08:59:29 -07:00 |
|