jesse-ux and Claude Code
aa46da1016
fix(rectification): use candidate dates for cross-midnight dasha scoring
...
Add date-isolated caches and regression coverage, align scoring identity, and freeze full research reruns while preserving historical artifacts. Record unresolved cache/receipt identity and end-to-end acceptance gaps for branch review only.
Co-Authored-By: Claude Code <noreply@anthropic.com >
2026-09-20 13:56:11 +08:00
Jesse_Chen and Claude Opus 5
03cba4780a
docs(tasks): 跨午夜单产品拍板放行 + 更正决策 C 的挂载点
...
产品 2026-09-20 就三点拍板:A 修;B 修完重新冻结并重跑 T1/T2;C 让修复
前后的结果可区分。任务书 §3 三个待决点改为已决,状态转待领取。
更正 C 的做法。原提案是改 calculation_spec 并 bump INPUT_CONTRACT_VERSION,
但查证发现 calculation_spec_hash 的消费方全在 V4 链路,生产跑的 V9 对它零
引用,改了对真实用户历史无效。改为随修复 bump engine_version:它取自
v9EngineVersion(),全仓只写不比、无任何相等性门控。明令不得动 skill_version,
BUG-621 就是 open RPC 的 Skill 版本绑定让历史校正打不开。
§1.6 同步重写为 V9 的真实身份字段表;T4 改为 engine_version bump,验收加
一条历史会话仍能打开的定向回归。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0199rbQDTsUbCVw84wc8BTFe
2026-09-20 12:28:44 +08:00
Jesse_Chen and Claude Opus 5
dacd404480
docs(tasks): 跨午夜候选 Dasha 边界任务书(BUG-981)+ 验证补缺单验收结论
...
验收 932f2fff:T1 完整重跑 900 组合逐例 0 差异、45 格表逐位相同;T2 独立复跑
五项指标与哈希全序全部一致;12 个打分文件字节未动、确认门仍关闭。执行方正确
推翻任务书 T2.3「首次口径干净的官方盲测」——与 BUG-428 防复发冲突,已核原文。
遗留立单:scoring_service.py 调 merge_transition_proximity() 只传一个
birth_date,该函数用它算全部候选的 dasha 起始日期,候选只靠 HH:MM 区分。
窗口跨午夜时午夜后候选边界整体错一天。生产调用链独立复现:23:50→00:10 的
21 个候选里,恰好 11 个跨日候选分数错、10 个同日候选逐位相同。
连带发现 calculation_spec() 不含打分实现身份,修复后同一 spec hash 对应
不同分数。本单会改生产打分,三个待决点未由产品填答不得开工。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0199rbQDTsUbCVw84wc8BTFe
2026-09-20 12:23:08 +08:00
jesse-ux and Claude Code
932f2fffba
research(rectification): add reported-offset evaluation and frozen rerun integrity
...
Independent Staging Quality Gate / validate (push) Successful in 12m7s
Independent Staging Quality Gate / publish (push) Successful in 3m46s
Preserve closed confirmation gates and previously-exposed dataset boundaries. Add auditable 900-trial sensitivity results, current scorer freshness checks, and the v5 collection protocol.
Co-Authored-By: Claude Code <noreply@anthropic.com >
2026-09-20 12:03:52 +08:00
Jesse_Chen and Claude Opus 5
5049e7847b
docs(tasks): 生时校正验证体系补缺任务书(BUG-978~980)
...
会议要求把「推断真实出生时间」与「用户认可的参考盘」分开证明。逐条核对
后:产品口径侧四条已落地(accepted≠confirmed 两条写入路径、确认门
fail-closed、采用不写 reported_birth_time、无把采用率当准确率的指标),
缺口全在评测本身。
三条实证:封存契约里三个打分哈希互不相同且 official_eval_trial_count 为
0,当前实现从未产出过有效官方盲测;全部离线评测的候选窗以真值为圆心而
生产以申报时间为圆心,真值掉出窗外这一失败模式从不可见;v3 封存但每例
仅 3 事件、v4 事件够却已被看过并用于调参,六题回放的高分是真值方向最优
答的上界。
任务分解:申报偏差敏感性 sweep、封存盲测有效重跑 + 契约对齐 + 防复发
测试、v5 采集协议(只出协议)、记录。硬红线:不改 12 个打分文件、不用
封存集调参、不得把 status 改 ready。家庭信息在拿到基线数字前不开工。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0199rbQDTsUbCVw84wc8BTFe
2026-09-20 11:05:46 +08:00
Jesse_Chen and Claude Opus 5
fcad06372e
docs(tasks): 普通对话寒暄轮快速通道任务书(BUG-976/977)
...
一句「你好」走完整窗口排盘 + 400 字判词 + 扣 1 点。三层实证与不扣点
通道缺口写进任务书;产品拍板不扣点不排盘回一句,分流禁用正则,
BUG-922/923 的每轮必调合同不动。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-20 10:19:18 +08:00
jesse-ux
8dc460da23
fix(ci): untrack .venv symlink and align Beam avatar contract
...
The staging gate failed on READ_ERROR .venv after a docs commit tracked a
symlink that gitignore .venv/ does not ignore. CI replaces that path with a
real venv directory. Drop the symlink, ignore the name itself, and lock it
in the privacy scan. Also point the Beam avatar source contract at the
current profile-panel copy so npm test stops failing after Python.
2026-09-20 09:11:24 +08:00
jesse-ux and Claude Code
5cbc2d0097
chore(privacy): purge upstream private material and add import guards
...
Prepare branch for remote review; preserve documented validation gaps and protected historical packages.
Co-Authored-By: Claude Code <noreply@anthropic.com >
2026-09-20 00:55:18 +08:00
Jesse_Chen and Claude Opus 5
3937390bca
docs(tasks): 库主案例清除 497798ac 验收通过——快速门转绿、全量 0 新红,待产品放行合入
...
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-20 00:36:31 +08:00
Jesse_Chen and Claude Fable 5.1
4a87790984
docs(tasks): 库主案例清除 fe9489e6 验收未通过——答案键守卫撞 pl9_1993 模式(P0)修复单 + 状态板
...
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-19 23:35:43 +08:00
Jesse_Chen and Claude Fable 5.1
a2c58d9f50
docs(tasks): 库主案例清除任务书拆分——本仓单只含本仓,上游清理改为交给库主的独立指令
...
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-19 16:13:04 +08:00
Jesse_Chen and Claude Fable 5.1
d8a6a4cf64
docs(tasks): 库主本人案例清除 + 冗余文档清理任务书(两仓,BUG-972~975)
...
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-19 16:06:11 +08:00
jesse-ux and Claude Code
2c92f6100e
fix(frontend): isolate sidebar registration to unblock navigation
...
Separate shell subscriptions from Home session data and cover effect convergence, transition cleanup, current callbacks and signed-out fallback with real React lifecycle tests. Record baseline-equivalent local failures and outstanding browser/build verification.
Co-Authored-By: Claude Code <noreply@anthropic.com >
2026-09-19 14:32:42 +08:00
Jesse_Chen and Claude Fable 5.1
a2bcdee006
docs(tasks): Jev 意图分类研究关单——产品拍板不接管、不影子双跑
...
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-19 14:24:40 +08:00
Jesse_Chen and Claude Fable 5.1
648aa07759
docs(tasks): Jev 意图分类研究 fix2 验收通过,结论缺数据采纳、研究单关闭
...
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-19 14:18:54 +08:00
jesse-ux
507ef959f5
research(jev-intent): fix2 把来源 B 现行与高置信错误补进报告
...
Independent Staging Quality Gate / validate (push) Failing after 9m3s
Independent Staging Quality Gate / publish (push) Skipped
离线从 cache 聚合,不重跑模型。无焦点层 Flash 69.7% 低于 Jev 78.8%。采集层相对门槛标不可判。
2026-09-19 12:28:43 +08:00
Jesse_Chen and Claude Fable 5.1
33367fb081
docs(tasks): Jev 意图分类修复轮验收——结论缺数据采纳,报告补漏单 fix2
...
红线 1–8 与 F1/F2/F5 通过;F3/F4 部分未通过:现行模型在来源 B 上跑了没报、
来源 B 上 Jev 高置信错误 6.4% 没进 MD、层差没到标签;复核提示≈生产提示且同模型,
采集层相对门槛不可判。不重造语料、不重跑、不立 BUG。
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
2026-09-19 12:12:24 +08:00
jesse-ux and Claude Code
eb8f5147f4
fix(frontend): refine settings dialog layout
...
Independent Staging Quality Gate / validate (push) Failing after 8m47s
Independent Staging Quality Gate / publish (push) Skipped
Co-Authored-By: Claude Code <noreply@anthropic.com >
2026-09-19 11:32:27 +08:00
jesse-ux
c2ecbc74cd
research(jev-intent): 修复轮重造语料并全量对照
...
Independent Staging Quality Gate / publish (push) Canceled after 0s
Independent Staging Quality Gate / validate (push) Canceled after 1m37s
来源 C 改为 DeepSeek Flash 生成+独立复核,撤回模板拼接结论。来源 B 157 条人工标注后跑 Jev x2 与 Flash 全量对照,结论为缺数据。
2026-09-19 11:30:31 +08:00
Jesse_Chen and Claude Fable 5
4f4cd6843a
docs(tasks): Jev 意图分类研究验收未通过,修复单要求真实模型生成来源 C
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0155nFCgCHtoA7jhSDGmZmMu
2026-09-19 01:55:19 +00:00
jesse-ux
fde541c2ca
research(rectification): Jev 意图分类离线对照,结论不可接
...
Independent Staging Quality Gate / validate (push) Successful in 10m0s
Independent Staging Quality Gate / publish (push) Successful in 3m49s
来源 C 900 条 + jev-1.13.0 双跑。高置信错误 7%、点选题 answer_class 65%。不改线上分类器。
2026-09-19 09:27:10 +08:00
Jesse_Chen and Claude Fable 5
d9722d56b0
docs(tasks): 研究单——TypeSafe Jev 接管校正意图分类的离线对照(Agent 模拟语料)
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0155nFCgCHtoA7jhSDGmZmMu
2026-09-19 00:50:55 +00:00
jesse-ux
bdf027a063
docs(bugs): 回填 BUG-968/969 修复版本 824ecff0
Independent Staging Quality Gate / validate (push) Successful in 12m8s
Independent Staging Quality Gate / publish (push) Successful in 8m40s
2026-09-18 19:16:34 +08:00
jesse-ux
824ecff026
fix(web): 账户弹窗移出 inert;校正快照失败不再静默(BUG-968/969)
...
账户弹窗与入门付费墙 portal 到 document.body,SidebarInset 的 inert 不再罩住弹窗。
GET 校正快照每个非 200 打 JSON warn;客户端失败显示「这一问还没读到」和重新读取。
下一题是可渲染选择题时,确定性回复正文带题干,挂卡后去重。
2026-09-18 19:16:00 +08:00
jesse-ux
e760b1d610
fix(consult): 窗口溢出重试保留预跑证据包,回填修复版本 SHA
Independent Staging Quality Gate / validate (push) Successful in 10m28s
Independent Staging Quality Gate / publish (push) Successful in 10m58s
2026-09-18 18:47:01 +08:00
jesse-ux
9ccb65b71f
fix(web): 次级页稳定外壳与缓存预取,跨部署导航自愈(BUG-966/967)
2026-09-18 18:45:48 +08:00
jesse-ux
dccedf372e
fix(web): 聊天列表恢复项目符号,收紧定义列表,压思考条间距(BUG-962/963/964)
2026-09-18 18:45:19 +08:00
jesse-ux
53d78137ff
fix(consult): 申报时段计算改为服务端预跑并走同请求缓存(BUG-957)
...
窗口计算挂在 agent context 缓存上,模型开口前预跑并注入 packet;工具再调用命中同请求缓存,成功次数仍为 1。
2026-09-18 18:45:19 +08:00
jesse-ux
5e8b2ac112
fix(consult): 降级正文过 Pass 4,去重并停止空跑 compose(BUG-959/960/961)
...
降级交付复用 releasePass4Sentences;每次 attempt 清空 uncontractedText;成功降级后跳过 compose。
2026-09-18 18:45:18 +08:00
Jesse_Chen and Claude Fable 5.1
bed961c4a1
docs(tasks): 账户弹窗被 inert 罩住(BUG-968)与校正快照两端静默(BUG-969)任务书
...
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_01JUei7K13cYxLHE3Axe4A45
2026-09-18 10:40:11 +00:00
Jesse_Chen and Claude Opus 5
1061514fc0
docs: BUG-965 触发条件确认(跨部署标签页),产品选定方案一
...
产品复现:刷新后恢复正常。BUG-965 根因定为标签页跨过 17:12 部署后
客户端导航拉不到已被替换的构建产物。修复方向进 BUG-967:用编译期
NEXT_PUBLIC_GIT_COMMIT 与 /api/health 比对,旧版本在下次导航整页重载。
次级页抖动按方案一做(统一外壳 + 缓存 + 预取),不加 spinner。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 10:00:42 +00:00
Jesse_Chen and Claude Opus 5
bfd6184e71
docs(tasks): 次级页进入抖动定位,出 BUG-966 任务书
...
三页都是等待文案块换成正文块的一次布局推挤;use-chart-page 每次进入
重新 fetch、无缓存,所以每次都抖。默认方案不加 spinner(撞「揭幕后不得
出现 spinner/骨架」红线),改统一外壳 + 缓存 + 预取消灭中间态;要动画
则走延迟揭幕,并需产品授权改红线。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 09:57:24 +00:00
Jesse_Chen and Claude Opus 5
fa21cf46a8
docs(tasks): 补 BUG-964 思考条与正文间距 56px 的实测根因
...
Chrome 实测 gapPx=56:.consultation-thinking-report 的 grid gap 24px
加首个 h2 的 margin-top 32px。后者源于 globals.css:415 的「首元素清零」
被 2353 行会话作用域规则按特指度压掉(0,2,0 对 0,4,1),在真实会话里
从未生效。另一条路径 message-stage-and-answer 只有 8px。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 09:47:50 +00:00
Jesse_Chen and Claude Opus 5
0a1a87b06e
docs(tasks): 聊天正文排版两条根因,出 BUG-962/963 任务书
...
用仓库自己的渲染管线 + next build 的 CSS chunk + headless Chrome 复现:
列表无项目符号(preflight 清 list-style,.markdown-list 没恢复);
promoteDefinitionLists 把四标题口语体散文误判成并列项,三段正文被
改写成列表项。收紧判据而非删除,BUG-356 的并列短项场景要继续工作。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 09:40:33 +00:00
Jesse_Chen and Claude Opus 5
ea0280c121
docs(tasks): 窗口修复单标记已验收通过
...
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 09:11:04 +00:00
Jesse_Chen and Claude Opus 5
895a85bcbe
docs(tasks): 5b6abc23 验收通过;降级路径 review 出 BUG-959~961
...
降级正文直接 send,不过 Pass 4:实跑「我保证你一定会升职」原样送达、
回执只有 contract-degraded(959)。uncontractedText 跨 attempt 不清零,
两次 attempt 的正文拼一起送出(960)。降级后仍空跑一轮 compose(961)。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 09:10:48 +00:00
jesse-ux
877128cec4
docs: 回填 BUG-954~956/958 修复版本 5b6abc23
Independent Staging Quality Gate / validate (push) Successful in 14m31s
Independent Staging Quality Gate / publish (push) Successful in 16m24s
2026-09-18 16:38:15 +08:00
jesse-ux
5b6abc238b
fix(consult): BUG-954~956/958 窗口方法块、abort 分码、合同降级
...
窗口 Agent 注入不含本命骨架的方法块;tripwire abort 走 skill_binding_failed;
无工具但有正文降级交付;应期问题仍先调工具。BUG-957 等窗口线验证后再做。
2026-09-18 16:37:43 +08:00
Jesse_Chen and Claude Opus 5
e5b2ad14dd
docs: BUG-954 根因由服务端日志定死——窗口 Agent 没绑方法块
...
日志:input-processor jyotish-skill-bound abort ×2,modelStepCount 0、
0 token、整轮 69ms,模型从未被调用。windowJyotishInstructions 从不含
BOUND_METHOD_MARKER,而 getWindowJyotishAgent 照样 attach 绑定,自
9958e00a(08-21)起申报时段每轮必败。abort 与「模型没调工具」同码,
是它藏四周的原因。初版任务书的「模型没调工具」立论已证伪并改写。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 07:50:31 +00:00
Jesse_Chen and Claude Opus 5
d0317f7f4e
docs: 申报时段整轮失败定位 BUG-954,出服务端预跑任务书
...
回执零 tool 步 + 有合同 retry + 公开码 runtime_contract_incomplete,
按 BUG-938 的分流规则排除供应商 error 块,两次 attempt 都是模型没调
工具。根因链 922→923→937:required 被拒后改回 auto,洞回归。窗口工具
入参只有 question,计算不需要模型触发,改服务端预跑。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 07:18:35 +00:00
jesse-ux
9cdcf96b18
docs: 回填 BUG-950~953 修复版本 cd4775ae
Independent Staging Quality Gate / validate (push) Successful in 10m24s
Independent Staging Quality Gate / publish (push) Successful in 13m30s
2026-09-18 14:09:00 +08:00
jesse-ux
cd4775aec6
fix(consult): BUG-950~953 Pass 4 按句放行,无分钟按句丢弃
...
正文不再整段 hold:闭合句立刻过 Pass 4,无 reject 即发(950)。
无出生分钟改为按句丢弃,全丢才用兜底句(951);该模式日期记 observe(952)。
校正流 token 级 thinking 死链按 P2 删除,测试翻转成否定合同(953)。
2026-09-18 14:08:45 +08:00
Jesse_Chen and Claude Opus 5
24a6fc9873
docs(tasks): e32ce624 验收通过;review 出 BUG-950~953 第三轮任务书
...
产品拍板:正文按句放行恢复流式(950);无出生分钟模式改按句丢弃、
全丢才用兜底句(951);该模式日期记 observe(952);校正流 token 级
thinking 是死链,按 P2 删除并把测试翻转成否定合同(953)。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 05:39:05 +00:00
jesse-ux
1e553976b4
docs(bugs): 回填 BUG-945~949 修复版本 e32ce624
Independent Staging Quality Gate / validate (push) Successful in 9m48s
Independent Staging Quality Gate / publish (push) Successful in 9m39s
2026-09-18 13:17:57 +08:00
jesse-ux
e32ce6247e
fix(consult): BUG-945~949 领域截断、思考分片、Pass 4 按模式分流
...
schema 上限与执行上限解耦;校正思考改分片门;日期观察不删字,保证句与无分钟个人盘退回重写。
2026-09-18 13:17:35 +08:00
Jesse_Chen and Claude Opus 5
b4fcb9466f
docs(tasks): BUG-948 改为按模式分开处置,钉住出生范围用户可用
...
产品 09-18 授权:范围用户必须能正常用产品。exact-timing 只记录不
拦(日期照出、按区间表述),guarantee 全模式退回重写,personal-chart
只在完全没有出生时间时拦。推翻「分钟敏感主题一律套精确时间守卫」的
旧红线(该红线曾让已校正用户问应期被挖成「[具体时间已省略]」)。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 04:36:32 +00:00
Jesse_Chen and Claude Opus 5
dbf984a40b
docs(tasks): 三通道重做验收未通过,出修复单(BUG-945~949)
...
在 7ae3aa9a 实测:tsc 0 错、lint 0 error,但 npm test 43 红 vs 基线
742ffbc6 的 31 红,新增 12 条 0 条转绿,staging 仍停在 742ffbc6。
主因:领域上限降到 2 且超限由 zod 整次拒绝(三领域问题零执行)。
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com >
Claude-Session: https://claude.ai/code/session_0193vBv6w5MV2cifdTUu9H5P
2026-09-18 04:29:12 +00:00
jesse-ux
7ae3aa9ae2
docs(bugs): 回填 BUG-942/943/944 修复版本 94c1e81f
Independent Staging Quality Gate / validate (push) Failing after 8m38s
Independent Staging Quality Gate / publish (push) Skipped
2026-09-18 12:16:54 +08:00
jesse-ux
94c1e81fa9
feat(consult): 进度、思考、正文三通道在生成时分开(BUG-942/943/944)
2026-09-18 12:16:20 +08:00
jesse-ux
742ffbc6b6
docs(bugs): 回填 BUG-940 修复版本 0378d9e0
Independent Staging Quality Gate / validate (push) Successful in 9m56s
Independent Staging Quality Gate / publish (push) Successful in 12m42s
2026-09-18 10:39:43 +08:00