Skip to content

[PR] docs: add Japanese versions of repository documentation (#6072) - #6076

Merged
ranxianglei merged 17 commits into
masterfrom
2026-11-10_japanese-docs
Nov 13, 2026
Merged

ranxianglei merged 17 commits into
masterfrom
2026-11-10_japanese-docs

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Nov 10, 2026 •

Copy link
Copy Markdown
Owner

Adds Japanese (ja) translations of the repository documentation, per #6072 ("All repository documentation is available in Japanese"). Single long-lived PR for the issue (owner: one PR, don't split); each section lands as its own commit on this branch.

Convention

  • Short BCP47 suffix .ja.md (matches the existing root README.ja.md from [PR] docs: add Japanese README (#6000) #6004); existing Chinese docs keep .zh-CN.
  • JA = faithful complete mirror of the English source (lossless): code blocks byte-identical, tables / inline code / issue refs preserved verbatim; does not inherit any condensation present in some .zh-CN twins.
  • Content pages start at H1 (no top language toggle — only configuration.md carries one); discovery is via the index rows in docs/index.md + docs/index.zh-CN.md.

Scope (owner-ruled 2026-11-10)

README + docs/ only. AGENTS.md is explicitly excluded (it's for LLMs, no Japanese needed). Root operational files outside "readme+docs" — CHANGELOG.md, reference/* — are out of scope. Four deliberately single-language docs/ files with no Chinese twin stay English-only by default: lifecycle, auto-merge-guardrails, unified-loop-spec, plugin-protocol. Say the word if any of those should be localized too.

Progress (complete 2026-11-13)

  • Identity & protocol — opencode-responses-websocket, session-identity-resolution, identity-channels, msgid-infrastructure, message-identity, session-identity
  • Directory landing page — docs/README.ja.md
  • clients.md — full client-integration deep dive (dsh / Kimi / Hermes / ZCode / Codex / Gemini family / Pi / Kiro / unlisted clients / http.proxy / OpenCode)
  • technical-notes.md — docs/technical-notes.ja.md (+ root TECHNICAL-NOTES.ja.md redirect stub)
  • Architecture hub + 8 module pages — docs/architecture.ja.md + docs/architecture/*.ja.md; tools/check-architecture-docs.mjs now enforces the ja pages (ledger parity with the EN hub, EN-diagram reuse); diagrams reuse the en SVGs with a one-line note
  • Generated configuration — docs/configuration.ja.md via tools/gen-config-docs.mjs (339 desc_ja seed entries; CI docs-sync lane drift-checks the ja file too)
  • Index registration — [日本語] links on every row in docs/index.md + docs/index.zh-CN.md; root redirect stubs (CLIENTS / CONFIGURATION / MESSAGE-IDENTITY / SESSION-IDENTITY / TECHNICAL-NOTES.ja.md)

Closes #6072

@ranxianglei ranxianglei added documentation Improvements or additions to documentation p3 labels Nov 10, 2026
@github-actions

github-actions Bot commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-10_japanese-docs (db2e1de)

Latest build — install this exact version

npm install -g billion-context@0.0.9427-pr.6076.8b67d60

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Recent builds (commit → version)

commit version build
db2e1de billion-context@0.0.9427-pr.6076.8b67d60 build
699a6a4 billion-context@0.0.9365-pr.6076.3706692 build
438fbe8 billion-context@0.0.8950-pr.6076.a1e2e41 build
02c4f7b billion-context@0.0.8896-pr.6076.1506ad0 build
9ceb10f billion-context@0.0.8892-pr.6076.21a29b1 build
784f137 billion-context@0.0.8890-pr.6076.d51e2ed build
f1c1318 billion-context@0.0.8879-pr.6076.d9562f7 build
fa31da6 billion-context@0.0.8877-pr.6076.8b2edb5 build
2554e48 billion-context@0.0.8876-pr.6076.b069350 build
2e59aa8 billion-context@0.0.8873-pr.6076.6185d7a build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr6076.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei changed the title docs: add Japanese (.ja.md) translations — identity & protocol section [PR] docs: add Japanese (.ja.md) translations — identity & protocol section Nov 10, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重: 不重复。Scanned open [PR] rows (78 total): this thread (GitHub PR #6076) is the only one targeting the Japanese-docs batch under #6072 — no parallel second PR exists, so my review fixes went onto this same branch (one PR per thread).

标签: scope/docs + p3(纯文档翻译交付,无运行时影响,可发现性层级);原有 pr 标签保留。

Review — PR #6076 (batch 1: Identity & protocol JA translations)

Verified live just now (branch 2026-11-10_japanese-docs, head 2554e48 after my two small fixes):

  • Base freshness: the branch contains current master (541cf26) — no stale base. GitHub reports mergeable=true, mergeable_state=blocked by the approval gate, i.e. waiting on your merge decision, not on anything technical.
  • Diff shape: exactly the 6 declared files (+234/−10): 4 new .ja.md pages plus [日本語] links in both docs/index.md and docs/index.zh-CN.md. Nothing unrelated, no churn.
  • Fidelity (mechanically checked, all 4 EN↔JA pairs): fenced blocks byte-identical to EN, inline-code spans preserved, issue-ref sets identical, heading level-sequence identical, relative md-link targets identical. I also read all four pages end-to-end — accurate, natural Japanese; nothing dropped (full lane table, trust boundaries, change checklists all present).
  • Conventions: .ja.md naming follows the owner-approved README.ja.md precedent ([Docs] Japanese README: where should it live? | [文档] 日文 README 应该放在哪里? #6000, approved 2026-11-09); content pages correctly carry no top language-toggle (the EN sources have none either); discovery via index rows only, as declared.
  • Link validity: README.ja.md exists on current master (landed via [Docs] Japanese README: where should it live? | [文档] 日文 README 应该放在哪里? #6000), so the newly added [日本語](../README.ja.md) index links are live, not forward references.
  • CI: on the pre-fix head 2e59aa8 every workflow finished green (main CI + all e2e lanes success; the cancelled entries are superseded earlier triggers). Re-running on the new head below.

Two small fixes pushed to the branch during review (trivially safe, each its own labeled commit):

  1. 7fc8684 — docs/identity-channels.ja.md: the ASCII consumer-path diagram inside the fence carried +1 leading space on its 4 continuation lines vs the EN source, breaking the PR's own "byte-identical fenced blocks" property. Restored the exact EN indentation.
  2. 2554e48 — docs/index.zh-CN.md: the README row read [README](../README.md) / [English](../README.zh-CN.md) — the "English" label pointed at the Chinese file, crossed against every other row in that file ([name](zh) / [English](en)). Reordered to the house pattern. Pre-existing nit, fixed in passing since this PR touches the same line.

八问

  1. 是什么问题:不是缺陷——是 All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072「全量日文化」任务的批次 1:为 Identity & protocol 的 4 个文档页添加日文镜像(lossless 完整镜像),并在两个索引里注册 [日本語] 链接;顺带把自 [PR] docs: add Japanese README (#6000) #6004 起存在却从未被索引链接的 README.ja.md 挂上。
  2. 什么原因导致的:不涉及(新增翻译交付,无缺陷根因)。
  3. 影响范围与严重程度:纯文档,零运行时/线协议/存储/配置面影响;受众=所有读文档的人。严重度 p3(可发现性增强,无行为风险)。
  4. 怎么修复的:按 owner 已批准的 README.ja.md 先例采用短 BCP47 后缀 .ja.md;JA 是 EN 的忠实完整镜像(既有 ZH 版若有删减/重构,JA 刻意不继承其缺口)。我用脚本机械校验了围栏代码块逐字节一致、行内代码 span、issue 引用集合、标题层级序列、相对 md 链接目标全部对齐,并逐页通读了四份译文确认语义无损。审核中顺手修两处小瑕疵(围栏图缩进、中文索引 README 行标签交叉),各一个独立 commit。
  5. 是否有回归问题:审过的既有行为=两个索引文件的行结构与可发现性、README.ja.md 链接有效性(文件已在 master,链接有效非悬空);无语义漂移,无阈值/时序/输出格式变化;architecture-docs、docs-sync 等 CI 守卫在修复前 head 全绿,新 head 正在重跑。
  6. 是否破坏缓存:不涉及(未触碰缓存/重试/增量游标/去重键)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及(无 env/config/默认值变化)。

判定:可合。技术面无阻塞(base 最新、diff 干净、保真度机械验证通过、CI 绿),等你批准合并即可。剩余批次(message-identity/session-identity → clients/technical-notes/architecture → engineering & ops → 生成的 configuration)按计划以后续 commit 落到同一分支(一个线程一个 PR)。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(ai: vllm-qwen/qwen3.8-27b)

Verdict: APPROVE — docs-only PR; verified faithful and complete against all four EN sources; four small fidelity nits found, all resolved on the branch during review (two fixed in parallel by the author, two by me on top of theirs).

Correctness. Read all four EN sources in full and cross-checked every section, bullet, table row, code span and issue reference against the JA text:

  • Heading structure mirrors 1:1 in all four pairs; issue-ref sets (#NNNN) identical in all four.
  • Structural counts equal: session-identity-resolution 15 bullets / 9 numbered / 6 table rows; identity-channels 6 bullets / 19 table rows; msgid-infrastructure 4 bullets / 22 table rows; opencode-responses-websocket 10 bullets. (msgid-infrastructure showed 9 EN vs 8 JA numbered items — that gap IS finding #2 below, not dropped content.)
  • Line-count gaps (EN 150→JA 62 etc.) are fully explained by EN soft-wrap vs JA single-line paragraphs; content is complete.
  • Fenced code blocks now byte-identical to EN (verified by diff after the fix).
  • .ja.md naming follows the README.ja.md precedent (#6004); absence of a top-of-file language-toggle line on content pages matches convention (only configuration.md has one); both index files register the new pages and surface the previously-unlinked README.ja.md.

Tests / CI guards. No test surface exists or is needed for markdown. The lanes that could care are unaffected: architecture-docs scans only the docs/architecture/ hub + pages; docs-sync diffs only the generated config docs; docs-coverage is config-reference coverage. All green on head 2e59aa8; re-running after my two doc-only commits.

Risks. Zero: no code, config, wire, persistence, kernel, or version touched. No behavior drift to disclose.

Fixes made on this branch during review:
已顺手修:msgid-infrastructure.ja.md §8 item 5 lost its list rendering ('5.' had no trailing space → GitHub renders it as a plain paragraph, not list item 5) + commit b11fce2
已顺手修:opencode-responses-websocket.ja.md dropped the registration name — EN 'registered as responsesCodec in a codec table' became nameless 'コーデックテーブルに登録される' + commit fa31da6
(Parent commits 7fc8684 fence diagram restored byte-exact to EN, and 2554e48 index.zh-CN README row labels de-crossed, were pushed in parallel by the author while this review was running — verified equivalent, my duplicate attempts dropped in favor of theirs, which matches sibling-row house style better.)


结构化八问(中文)

  1. 是什么问题:不是缺陷修复——为身份与协议四篇文档补日语译本(#6072 全量日语文档化第 1 批),并在两个索引中登记新页、顺带链接此前未链的 README.ja.md。
  2. 什么原因导致的:不涉及(新增翻译,无根因)。
  3. 影响范围与严重程度:仅文档可读面;对现有用户零行为影响。
  4. 怎么修复的:逐篇对照 EN 原文做无损镜像(标题/列表/表格/代码块/issue 引用逐项核对);命名沿用 README.ja.md 先例;审查中发现的 4 处保真度小疵当场修掉(围栏图字节对齐、列表项缺空格、注册名丢失、索引标签错位——后两处前两个由作者并行提交等价修复,本审核采纳作者版本)。
  5. 是否有回归问题:无——纯 .md 新增 + 索引行;architecture-docs / docs-sync / docs-coverage 三个相关 CI 门均不受影响且绿;未触碰任何代码/配置/线协议/持久化格式/版本号。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:可合(APPROVE)。最终 head fa31da6 的 CI 复跑中(两个新增 commit 均为纯 markdown,零代码路径)。

@ranxianglei ranxianglei changed the title [PR] docs: add Japanese (.ja.md) translations — identity & protocol section docs: add Japanese versions of repository documentation (#6072) Nov 10, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Review lane done: PR #6076 (Japanese .ja.md translations — identity & protocol section, batch 1 of #6072) reviewed end-to-end → APPROVE / 可合.

What was checked. Read all four EN sources in full and cross-checked every JA file section-by-section: headings mirror 1:1, bullets/tables/numbered items count equal (15+9+6 / 6+19 / 4+22+9 / 10), issue-ref sets identical in all four pairs, fenced code blocks byte-identical after fix; line-count gaps fully explained by EN soft-wrap vs JA single-line paragraphs. Naming follows the README.ja.md precedent (#6004); JA pages correctly omit any top-of-file toggle line (only configuration.md has one). Docs-only diff (6 files, +234/−10) — no code/config/wire/persistence/version touched; architecture-docs, docs-sync, docs-coverage lanes unaffected, all green on head 2e59aa8, re-running on final head fa31da6.

Findings — all closed, none parked. Four small fidelity defects found during review:

  • fence diagram off by one leading space vs EN → fixed in parallel by author (7fc868425c, byte-identical to my fix — adopted theirs)
  • docs/msgid-infrastructure.ja.md:74 list item 5 lost rendering (no space after 5.) → 已顺手修 commit b11fce24fb
  • docs/opencode-responses-websocket.ja.md:23 dropped the `responsesCodec` registration name → 已顺手修 commit fa31da6e60
  • pre-existing mislabel on the exact index.zh-CN row this PR edits → de-crossed in parallel by author (2554e481d3, matches sibling-row house style better — adopted theirs)

GitHub review posted as a review-comment with full verdict (APPROVE event refused — this account is the PR author): #6076 (review) · One-line report filed on #4599. Awaiting owner merge.


结构化八问(中文)

  1. 是什么问题:不是缺陷修复——为身份与协议四篇文档补日语译本(All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072 全量日语文档化第 1 批),并在两个索引中登记新页、顺带链接此前未链的 README.ja.md。
  2. 什么原因导致的:不涉及(新增翻译,无根因)。
  3. 影响范围与严重程度:仅文档可读面;对现有用户零行为影响。
  4. 怎么修复的:逐篇对照 EN 原文做无损镜像(标题/列表/表格/代码块/issue 引用逐项核对);命名沿用 README.ja.md 先例;审查中发现的 4 处保真度小疵当场修掉(围栏图字节对齐、列表项缺空格、注册名丢失、索引标签错位——其中两处作者并行提交了等价修复,采纳作者版本避免重复 commit)。
  5. 是否有回归问题:无——纯 .md 新增 + 索引行;architecture-docs / docs-sync / docs-coverage 三个相关 CI 门均不受影响且绿;未触碰任何代码/配置/线协议/持久化格式/版本号。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:可合(APPROVE),待 owner 合并。

@ranxianglei ranxianglei changed the title docs: add Japanese versions of repository documentation (#6072) [PR] docs: add Japanese versions of repository documentation (#6072) Nov 10, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Status update — the branch moved while I was reviewing; refreshed everything against the new head. My previous comment cited head 2554e48; since then batch 2 landed on this branch (as planned in the PR description) plus more fidelity fixes. Live-verified state below, head now 784f137.

What landed after my last comment:

  • b11fce2 + fa31da6 (parallel reviewer session): two more real fidelity fixes, both verified correct by me — (a) msgid-infrastructure.ja.md item 5 started 5.「ID… with no space after the period, so CommonMark rendered it as a paragraph instead of a list item (items 1–4 were fine); (b) opencode-responses-websocket.ja.md dropped the standalone "registered as responsesCodec" naming from the EN source. Correction to my earlier analysis: my mechanical check had actually flagged the missing responsesCodec span, and I misclassified it as a line-wrap false positive — it was real. The parallel session caught it correctly; the fix restores the name verbatim.
  • f1c1318 (batch 2): docs/message-identity.ja.md (419 L) + docs/session-identity.ja.md (255 L) + [日本語] links on both rows in docs/index.md / docs/index.zh-CN.md, matching the house row pattern.

Batch-2 verification (me, just now): extended the same mechanical fidelity check to all 6 EN↔JA pairs — fenced blocks byte-identical, inline-code span multisets identical (after stripping fences and normalizing legitimate multi-line spans), issue-ref sets identical, heading level-sequence identical (1:1 heading correspondence in both new pages). Index registration verified against both files. Two residual nits found, fixed as labeled commits 11f3cc7 + 784f137:

  1. session-identity.ja.md: the constant TRUNCATION_LINEAGE_WINDOW = 8 was wrapped mid-token across a newline inside its code span — CommonMark renders that newline as a space, so the identifier displayed as TRUNCATION_LINEAGE_ WINDOW = 8. Re-wrapped the paragraph (no content change); the identifier now renders intact.
  2. message-identity.ja.md heading: mixed full-width opening bracket with ASCII closing paren (#3641) → (#3641), matching the file's other headings.

Full fidelity re-run after those fixes: all 6 pairs clean.

CI at head 784f137 (verified live): 9/10 workflows green — all eight e2e lanes, CI Windows codex overlay, PR Build Artifact success. Main CI workflow is queued (~26 min, fleet-wide queue congestion, created 04:18:33Z, never started — not a failure). I'll confirm here the moment it finishes; nothing else blocks.

八问(增量更新,相对上一条回复)

  1. 是什么问题:同一交付(All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本 #6072 日文化批次),范围扩大到批次 2(message-identity + session-identity)。
  2. 什么原因导致的:不涉及(翻译交付);本轮新发现的两处瑕疵均为排版级(代码 span 跨行折断、括号全半角混用)。
  3. 影响范围与严重程度:仍为纯文档、零运行时影响;p3。
  4. 怎么修复的:机械保真校验扩展到全部 6 对 EN↔JA(围栏逐字节 / 行内代码多重集 / issue 引用集 / 标题层级序列),并逐页通读;审核中发现的排版瑕疵当场修掉,各一个独立标注 commit。
  5. 是否有回归问题:无新增风险面——两处修复只改软换行位置和一对括号字符,不改任何语义内容;既有索引行模式保持一致;architecture-docs / docs-sync 守卫路径未触碰。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:维持可合——等主 CI 出队跑完(预计很快)+ owner 批准合并。剩余批次(clients/technical-notes/architecture → engineering & ops → 生成的 configuration)继续按「一个线程一个 PR」落到本分支。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(ai: vllm-qwen/qwen3.8-27b)

Verdict: APPROVE — posted as a comment because this account cannot approve its own PR (same as round 1).

Round-2 re-review at final head 02c4f7b41c (branch moved from f1c13181f5 while I was reviewing: author commits 11f3cc787f, 784f137e34, 9ceb10f76c + two reviewer fixes below). Full PR at merge-base 541cf260c4: 9 files, all under docs/, 938+/16- — seven new .ja.md translations plus index rows in both index files. Docs-only; no code/config/wire/persistence surface touched; branch is behind master with zero file overlap -> clean merge.

Faithfulness verified section-by-section against the EN sources:

  • message-identity.ja.md (419 L) vs .md (432 L): headings 1:1 (18), bullets 19/19, table lines 21/21, issue-ref sets identical, fenced diagram byte-identical. All specifics carried: #242 66-char id vs 64-char cap, #2396 (~66,525 tokens re-entered at 558,541/562,549 vs the 500,000-token window), SHAPE_MASS_MIN / simhash <= 8 gate, 16384-anchor cap, tests/identity-proof.test.ts matrix, #3756/#3751 mechanics.
  • session-identity.ja.md (255 L) vs EN (268 L): headings 1:1 (19 incl. ###), bullets 21/21, numbered items 3/3, table lines 19/19, issue refs identical. Measurement table (9,367 full-prefix / 2,106 new / 142 fork lineage / 0 tail-window), TRUNCATION_LINEAGE_WINDOW = 8 / MIN_FORK_PREFIX = 3, wire-neutral projection rules, live evidence (45 KB file, ~13.5k tokens saved, 97% cache hit within two rounds) all present.
  • README.ja.md (24 L) vs README.md: intro / catalog blockquote / 4-row section table (identical link sets) / 5 rules / AGENTS pointer — faithful.
  • Index rows: this round only touches the README row (/ [日本語](README.ja.md) appended in each file); pattern matches the sibling rows exactly.

Author's proactive fixes — all correct, adopted:

  • 11f3cc787f: si.ja.md code span TRUNCATION_LINEAGE_WINDOW = 8 was split mid-token across a line break inside backticks (CommonMark renders that as an inserted space); re-wrapped, no content change. Real rendering defect, fixed right.
  • 784f137e34: mi.ja.md heading (#3641) mixed-width parens -> (#3641). Identical to my queued finding; the author got there first.

已顺手修:message-identity.ja.md summary line render join(rollout gate) mixed-width parens -> (rollout gate) + commit 424a7e9489
已顺手修:README.ja.md 「落地ページ」(Chinese calque of "landing page") -> 「ランディングページ」(standard term, matches the file's katakana style) + commit 02c4f7b41c

Tests/CI: docs-only — no test surface exists for this diff. At head 9ceb10f76c the gate lanes are already green (static-gates, mysql-contract, test-import-graph, config-surface-guard, ...); the test shards were still running when I posted and cannot be affected by a docs-only diff.

Risk: none. No behavior drift, nothing to disclose 旧值->新值.


结构化八问

  1. 是什么问题:无缺陷——本 PR 是纯文档新增(message-identity / session-identity / docs 落地页的日文翻译 + 两个 index 的行);本轮审核对象是作者在我首轮审核后新推的三个 commit(含作者自修的一处渲染缺陷)。
  2. 什么原因导致的:不涉及(非 bug 修复类 PR)。
  3. 影响范围与严重程度:仅文档可读性面——为日文读者新增三个文档入口;不触碰任何代码、配置、协议、持久化路径。
  4. 怎么修复的:不适用(审核 + 两处顺手小修:全角括号宽度不一致 x1、日文用词校准 x1)。
  5. 是否有回归问题:审过的既有行为——两个 index 文件的行格式与既有兄弟行模式完全一致;全部 EN 源文件零改动;CI guard(architecture-docs / docs-sync / docs-coverage)的扫描范围均不覆盖顶层 .ja.md。无回归。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:可合。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Round 2 review 完成: GitHub PR #6076 APPROVE — final head 02c4f7b41c。

  • 作者在我首轮审核后新推 3 个 commit, 已逐一复核: 11f3cc787f(si.ja 代码 span 断行渲染缺陷自修)、784f137e34(mi.ja 标题括号宽度修复, 与我排队中的发现相同, 作者先行)、9ceb10f76c(新增 README.ja.md + index 行)。
  • 全部 7 个 .ja.md 与 EN 原文逐节交叉核对: 标题 1:1、bullet/编号/表格行数一致、issue 引用集合一致、围栏代码块字节级一致、关键数字([Bug][Prime/Responses] provider switch rewrites composite tool IDs and reintroduces compressed history #2396 token 数、测量表 9,367/2,106/142/0 等)无遗漏。
  • 已顺手修 ×2: message-identity.ja.md 总结行 (rollout gate) 混宽括号 → (rollout gate)(commit 424a7e9489);README.ja.md「落地ページ」→「ランディングページ」(commit 02c4f7b41c)。
  • 最终 PR diff: 9 文件全部在 docs/ 下(938+/16-), 纯文档, 无回归面。CI 门禁已绿, test 分片运行中(文档改动不影响)。

判定: 可合。GitHub 审核评论: #6076 (review)

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Red CI cleared — no code change needed. The single failing job was a transient flake, unrelated to this docs-only PR.

What I did this run (verified live):

  1. Re-verified live state first — exactly one red: test (windows-latest, 22, 3); every other test shard plus static-gates, gate, mysql-contract, lifecycle-ledger-guard, lane-guard were green. Not already fixed by an earlier wake.
  2. Pulled the failing job's log. The red test was not ok 528 - fork receipts cut pinned fork_base checkpoints… (#2960 ②) in tests/fork-replica-convergence.test.ts, but the error was a transport-level TypeError: fetch failed (undici failing to reach the local proxy), not an assertion mismatch — the classic signature of a runner-side socket/network flake on that Windows shard.
  3. Confirmed the PR cannot have caused it. The diff is docs-only: 9 files, all docs/*.md (+938/−16). No src/, no tests/, no kernel/, no config touched — so a runtime test failure is impossible from this content.
  4. Re-ran just the failed job (rerun-failed-jobs). It completed success (~6.6 min); GitHub's own Actions API now reports the whole workflow run as completed / success. (The check-status cache still shows the pre-rerun failed job because the rerun minted a new job id; the live API confirms green.)

No commits, no pushes, zero branch changes. Nothing was merged or modified.


八问(CI 修复结论)

  1. 是什么问题:该 PR 的 CI 仅一个测试分片 test (windows-latest, 22, 3) 红,其余全绿。
  2. 什么原因导致的:该分片里的 fork-replica-convergence.test.ts(测试 Identity pipeline: key sessions by (agent, conversation id) in the unified store — decouple fold ledgers from protocol/host origin (#2938 / #2847 / #2671 defect-1) #2960 ②)抛出的是传输层错误 TypeError: fetch failed(undici 连本地代理 fetch 失败),不是断言失败。本 PR 是纯文档改动(9 个文件全是 docs/*.md,+938/−16,未碰任何 src/test/kernel/config),不可能引起运行时测试失败——判定为该 Windows runner 上的一次性网络/socket 抖动(flaky)。
  3. 影响范围与严重程度:仅限该 PR 的 CI 门禁(一个测试分片、一次抖动);不影响任何用户、不影响产品行为、不影响其他分片。属 CI 偶发,非功能缺陷。
  4. 怎么修复的:不改代码(纯文档 PR 不应夹带源码/测试改动)。直接对该失败 job 触发「重跑失败 job」(rerun-failed-jobs),重跑通过,整个 workflow run 现为 completed/success。
  5. 是否有回归问题:无。未在分支上做任何 commit/push,未改动任何代码/配置/行为。
  6. 是否破坏缓存:不涉及(无任何代码或缓存逻辑改动)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:CI 侧已清,该文档 PR 从 CI 角度可合并。 附注:fork-replica-convergence.test.ts 在该 Windows 分片出现过一次性传输抖动——单次发生不足以立项跟踪,列为观察项;若后续在 master 复现再单独跟进(不塞进本纯文档 PR)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

Final batch landed — the PR now covers the full agreed scope (README + docs/). This push (rebased onto latest master, forced):

  • docs/technical-notes.ja.md (+ root TECHNICAL-NOTES.ja.md redirect stub)
  • docs/architecture.ja.md + all 8 module pages under docs/architecture/; tools/check-architecture-docs.mjs extended so CI now verifies the ja pages exist, their ledger matches the English hub, and they reuse the EN diagrams (*.en.svg/.en.png) — no new SVGs
  • docs/configuration.ja.md: hand-written sections translated; both generated blocks (param ref + env map) produced by tools/gen-config-docs.mjs, which now requires a desc_ja on every seed entry (all 339 filled across the 8 website/config-reference/*.yaml). Regenerating from untouched seeds leaves docs/configuration.md, docs/configuration.zh-CN.md, website/main.js and website/docs/index.html byte-identical (verified); the CI docs-sync lane now drift-checks the ja file too
  • [日本語] links in docs/index.md / docs/index.zh-CN.md rows; the 3-language switch line on configuration.md / configuration.zh-CN.md; remaining root redirect stubs

All translations were machine-translated (local qwen3.8-27b) then audited structurally: code fences byte-identical, table row counts, #NNNN refs, and inline-code spans diffed against the English source (one typo found and fixed). Locals: please proofread the prose — naturalness is the one thing a structural audit can't guarantee.

Body now flips to Closes #6072.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

CI on the new head (699a6a4) is now green. One job was red on the fresh run; it's the known Windows teardown flake, not your changes — I re-ran it and it passed.

What I found and did (verified live):

  • The fresh run on 699a6a4 had exactly one red: test (windows-latest, 22, 2). Everything else — all other test shards, static-gates, docs-sync, gate, the guards — green.
  • The red test was abort x proxy-openai-sse: client abort mid-stream destroys the upstream request in tests/wire-exit-matrix.test.ts — AssertionError: upstream socket(s) must be destroyed after the client abort (closed=0), duration ~21.4s, while the sibling abort cells in the same run passed in 84–417ms (anthropic 156ms, responses 84ms, json 417ms).
  • That is the documented, recurring Windows load-starvation flake, not a regression. The test file's own header comment (tests/wire-exit-matrix.test.ts:29-66) records the full ladder (3s→6s→10s→20s budget + lag extension), a mandated "investigate a real teardown hang" audit that found none (relay.ts arms res.on("close") → clientAbort.abort() synchronously; no commit touched destroy/abort wiring), and the identical prior signature (openai-sse closed=0 at ~20s, siblings 131–410ms, ubuntu green). It's already tracked under CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599.
  • Unrelated to this PR: the diff is docs/tooling only (docs/*, tools/check-architecture-docs.mjs, tools/gen-config-docs.mjs, website/config-reference/*.yaml) — no src/, no tests/, no relay/abort wiring.
  • Action: re-ran just the failed job → passed. The whole CI run for 699a6a4 is now completed / success (confirmed against the live Actions API; the check-status cache was stale the whole window, so I verified directly).
  • No commits, no pushes, no branch changes.

So the only remaining gates on this PR are the human ones you named: locals proofreading the prose, then merge. I did not proofread the translations — naturalness is the human part you called out, and a structural audit can't guarantee it.


八问(CI 结论)

  1. 是什么问题:新 head 699a6a4 的首次 CI 仅一个分片 test (windows-latest, 22, 2) 红,其余全绿。
  2. 什么原因导致的:该分片里 wire-exit-matrix.test.ts 的 abort x proxy-openai-sse 断言 closed=0 失败、耗时 ~21.4s,而同 run 的兄弟 abort 用例 84–417ms 全绿。这是该测试文件头部注释(wire-exit-matrix.test.ts:29-66)记录的已知 Windows 负载饿死抖动:预算已从 3s 一路提到 20s(+滞后延展),且做过"排查真实 teardown 挂起"审计、确认无回归(relay.ts 同步 arm res.on('close')→clientAbort.abort(),无提交动过 destroy/abort 接线),历史签名完全一致(openai-sse closed=0 @~20s、兄弟 131–410ms、ubuntu 全绿),已在 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 跟踪。
  3. 影响范围与严重程度:仅限该 PR 的 CI 门禁(一个 Windows 分片、一次抖动);不影响任何用户/产品行为/其他分片。属 CI 偶发,非功能缺陷。
  4. 怎么修复的:不改代码(纯文档/工具 PR 不应夹带源码或测试改动)。直接对该失败 job 触发 rerun-failed-jobs,重跑通过,整个 CI run 现为 completed/success。
  5. 是否有回归问题:无。未在分支做任何 commit/push;红测试与本次改动无关(改动仅 docs/ + tools/ + website 配置 yaml)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:新 head 的 CI 已全绿,该文档 PR 从 CI 角度可合并;剩余门禁是你点名的人工校对(译文自然度)与合并。 附注:该 Windows teardown 抖动在 3s/6s/10s/20s 预算下仍偶发,属 master 级 CI 稳健性问题(已在 #4599),不适合塞进本纯文档 PR;若要根治(让该用例不依赖墙钟预算)应单独立项。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

CI on the new head 699a6a4 is green. The fresh run had exactly one red — test (windows-latest, 22, 2) — and it's the known Windows teardown flake, not your changes. I re-ran just that job; it passed, so the whole run is now completed / success (verified live against the Actions API).

Remaining gates are the human ones you named: locals proofreading the prose, then merge. I did not proofread the translations (naturalness is the human part).


八问(CI 结论)

  1. 是什么问题:新 head 699a6a4 的首次 CI 仅 test (windows-latest, 22, 2) 一个分片红,其余全绿。
  2. 什么原因导致的:该分片 wire-exit-matrix.test.ts 的 abort x proxy-openai-sse 断言 closed=0 失败、耗时 ~21.4s,同 run 兄弟 abort 用例 84–417ms 全绿。这是测试文件头部注释(wire-exit-matrix.test.ts:29-66)记录的已知 Windows 负载饿死抖动(预算 3s→20s + 滞后延展,审计确认无回归,已在 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 跟踪),非回归。
  3. 影响范围与严重程度:仅限该 PR 的 CI 门禁(一个 Windows 分片、一次抖动),不影响任何用户/产品行为/其他分片。
  4. 怎么修复的:不改代码(纯文档/工具 PR)。对该失败 job 触发 rerun-failed-jobs,重跑通过,整个 CI run 现为 completed/success。
  5. 是否有回归问题:无。未做任何 commit/push;红测试与本次改动无关。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:新 head 的 CI 已全绿,该文档 PR 从 CI 角度可合并;剩余门禁是人工校对(译文自然度)与合并。

ci-check added 9 commits November 13, 2026 08:49
…an on one line

The constant was wrapped mid-token across a line break inside its code span; CommonMark renders that newline as a space, so the identifier displayed as 'TRUNCATION_LINEAGE_ WINDOW = 8' instead of 'TRUNCATION_LINEAGE_WINDOW = 8'. Re-wrapped the paragraph; no content change.
…3641)

Heading line 215 mixed a full-width opening bracket with an ASCII closing paren; made it (#3641) to match the file's other full-width-bracket headings.
Faithful complete mirror of docs/README.md; registers the ja link on the
directory-landing row of both index files.
ci-check and others added 8 commits November 13, 2026 08:49
ja pages reuse the EN diagrams (svg+png); hub ledger locked to EN by the
extended check-architecture-docs guard.
…upport (#6072)

- gen-config-docs.mjs: ja seed field (desc_ja, required), ja param-ref and
  env-map blocks written to docs/configuration.ja.md; EN/ZH/main.js/html
  output byte-identical
- all 8 config-reference seeds: 339 desc_ja entries
- check-architecture-docs.mjs: ja hub/module-page rules (ledger parity with
  EN hub, EN-diagram reuse)
- ci.yml docs-sync lane now drift-checks docs/configuration.ja.md too
- configuration.md / configuration.zh-CN.md: three-language switch line
@ranxianglei
ranxianglei force-pushed the 2026-11-10_japanese-docs branch from 699a6a4 to db2e1de Compare November 13, 2026 00:50
@ranxianglei
ranxianglei merged commit dd0ef22 into master Nov 13, 2026
47 checks passed
@github-actions

Copy link
Copy Markdown
Contributor

📦 Merged Build Published

This merge to master was published as 0.0.1182-master.dd0ef22 (rolling master tag).

npm install -g billion-context@0.0.1182-master.dd0ef22  # this exact build
npm install -g billion-context@master  # always the newest merged build

npm view billion-context@master prints the current version at any time.

ranxianglei pushed a commit that referenced this pull request Nov 13, 2026
@github-actions github-actions Bot mentioned this pull request Nov 13, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation p3 scope/docs

Projects

None yet

Development

Successfully merging this pull request may close these issues.

All repository documentation is available in Japanese. | 仓库所有文档都提供日文版本

1 participant